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.pbio.1001337 | The Circadian Neuropeptide PDF Signals Preferentially through a Specific Adenylate Cyclase Isoform AC3 in M Pacemakers of Drosophila | The neuropeptide Pigment Dispersing Factor (PDF) is essential for normal circadian function in Drosophila. It synchronizes the phases of M pacemakers, while in E pacemakers it decelerates their cycling and supports their amplitude. The PDF receptor (PDF-R) is present in both M and subsets of E cells. Activation of PDF-R stimulates cAMP increases in vitro and in M cells in vivo. The present study asks: What is the identity of downstream signaling components that are associated with PDF receptor in specific circadian pacemaker neurons? Using live imaging of intact fly brains and transgenic RNAi, we show that adenylate cyclase AC3 underlies PDF signaling in M cells. Genetic disruptions of AC3 specifically disrupt PDF responses: they do not affect other Gs-coupled GPCR signaling in M cells, they can be rescued, and they do not represent developmental alterations. Knockdown of the Drosophila AKAP-like scaffolding protein Nervy also reduces PDF responses. Flies with AC3 alterations show behavioral syndromes consistent with known roles of M pacemakers as mediated by PDF. Surprisingly, disruption of AC3 does not alter PDF responses in E cells—the PDF-R(+) LNd. Within M pacemakers, PDF-R couples preferentially to a single AC, but PDF-R association with a different AC(s) is needed to explain PDF signaling in the E pacemakers. Thus critical pathways of circadian synchronization are mediated by highly specific second messenger components. These findings support a hypothesis that PDF signaling components within target cells are sequestered into “circadian signalosomes,” whose compositions differ between E and M pacemaker cell types.
| In the fruit fly Drosophila melanogaster, the neuropeptide Pigment Dispersing Factor (PDF) supports circadian function by synchronizing two types of pacemaker cells, M cells and E cells. The PDF receptor (PDF-R) is a G protein coupled receptor (GPCR) whose activation stimulates adenylate cyclase (AC), thereby elevating levels of the second messenger cAMP in many different pacemakers including M cells. Drosophila contains at least 12 genes that encode potential ACs. In this study, we identify the AC downstream of the PDF receptor specifically in M cells and show that PDF signals preferentially through AC3. However, other GPCRs in the very same cells do not rely on AC3. A different scaffolding protein also influences PDF responses in M cells, suggesting that signaling components are spatially grouped to allow for coupling of specific receptors with downstream components. Remarkably, in E pacemakers, AC3 disruptions have no effect. These findings suggest that distinct PDF circadian signals exist in M versus in E pacemakers, and more generally, we propose a mechanism to differentiate signaling pathways that use common components.
| The importance of biological rhythms in the anticipation and response to daily environmental changes is underscored by their conservation throughout nature. In eukaryotes, these rhythms are generated by a set of core clock genes that contrive to produce interlocked feedback loops. Both mammalian and Drosophila circadian rhythms are controlled by diverse groups of pacemaker neurons that express these core clock genes and proteins. In Drosophila these rhythms are required in ∼150 neurons, which can be subdivided into six bilateral anatomically distinct groups [1]. There appear to be two classes of pacemaker neuron in the fly brain that differ in many fundamental ways—these are termed M and E cells for historical reasons [2]–[4]. Previous work indicates that these subgroups are functionally as well as anatomically distinct and that certain cells are associated with specific components of daily locomotor behavior. Importantly, these associations are subject to specific environmental conditions, and they display considerable plasticity under different light and temperature conditions [2],[3],[5]–[8]. These pacemaker subgroups communicate to synchronize with each other to produce coherent circadian rhythms [9],[10]. Neuropeptides are critical mediators of intercellular communication between pacemaker cells in both mammals and Drosophila and a number are expressed in the Drosophila clock cell system, including the Pigment Dispersing Factor (PDF) [11]–[14].
Loss of the PDF peptide or its receptor leads to abnormalities in circadian locomotor behavior, including a reduction in morning anticipatory peak and a phase advance of the evening anticipatory peak under 12∶12 LD [14]–[17]. Under constant conditions these flies show high levels of arrhythmicity or short, weak rhythms. PDF controls the amplitude and phase of molecular rhythms of pacemaker cells [9],[18].
PDF's role in synchronization of clock cells indicates that its mechanism of action is largely within the cells of the clock network. The PDF neuropeptide is expressed by two specific pacemaker subgroups (large and small LNvs) and the PDF receptor is expressed widely, although not uniformly, throughout the circadian network in both M and E cell groups [19]. The PDF receptor signals through calcium and cAMP, although specific signaling components remain unknown [15],[17]. Signaling can be demonstrated in nearly all pacemaker cell groups in vivo [20]. Previous work indicates that M cells increase cAMP levels in response to at least two neuropeptides, PDF and DH31 [20]. The PDF and DH31 receptors belong to the same class II (secretin) G-protein coupled receptor (GPCR) family [21]. Both PDF [15],[17] and DH31 receptors [22] stimulate adenylate cyclases (AC) to produce cAMP in vitro, and in M cells in vivo [20], but the specific downstream components that differentiate the two peptide receptors remain unknown. Likewise, the basis for PDF's differential actions on the molecular oscillator in different pacemakers [9],[18] has not yet been explained.
The present study asks: What is the identity of downstream components that are associated with PDF-R signaling pathways in different circadian pacemaker neurons? Specifically, using live imaging of intact fly brains, we identify the particular adenylate cyclase (AC) isoform that is associated with PDF signaling in small LNv—commonly called M cells. Although some signaling components are common to both DH31 and PDF neuropeptide signaling, we report that DH31 signaling does not require the same AC in the small LNv cells. This suggests that PDF signals preferentially through its favored AC, while other GPCRs, in the same identified pacemaker neurons, couple to other ACs. In addition, AC3 manipulations have no effect on PDF-R expressing LNd cells, part of the E cell network. Thus in Drosophila, critical pathways of circadian synchronization are mediated by at least two, highly specific second messenger pathways.
Epac1camps is a genetically encoded cyclic nucleotide sensor that can be visualized with subcellular resolution and that responds with great sensitivity to cAMP in Drosophila neurons [20],[23]–[25]. Live brains expressing the reporter (using the gal4/UAS system) were imaged, while saline was perfused through the line and responses were measured to a bolus presentation of peptide (Figure 1A and B). M neurons were easily identifiable by their position and morphology using a Pdf-gal4 driver, and it was possible to obtain discrete readings from multiple cells within the same brain hemisphere.
In vitro assays indicate that Epac1camps has much higher sensitivity to cAMP than to other cyclic nucleotides [26]; however, it was also shown that the sensor responds to changes in cGMP levels at the Drosophila neuromuscular junction [23]. Based upon in vitro studies of PDF signaling, we hypothesized that PDF receptor activation leads to increases in cAMP, not cGMP, levels in these pacemakers [15],[17]. To test this idea, we used SNAP (S-Nitroso-N-acetyl-DL-penicillamine (1)) as a Nitric Oxide (NO) donor, which is known to stimulate cGMP production [27]. Addition of SNAP led to a measurable loss of the CFP/YFP FRET in M cells, consistent with the interpretation that the EPAC sensor detects increases in cGMP levels in addition to those of cAMP levels. SNAP responses were reduced in amplitude after pretreatment with a guanylate cyclase inhibitor 1H-[1],[2],[4]Oxadiazolo [4,3-a]quinoxalin-1-one (ODQ). Importantly, ODQ pretreatment had no effect on PDF responses. Genetic over-expression of a cAMP-specific phosphodiesterase dunce reduced the amplitude of PDF responses, but had no effect on SNAP responses in either M or E cells (Figures S1 and S2). Together, these results are consistent with the supposition that, in vivo, PDF signals through cAMP, not cGMP.
The Drosophila genome encodes at least 12 ACs, five of which are expressed broadly, or at least broadly in the central nervous system (Flybase). The remaining cyclases (ACXA-E and CG32301 and CG32305) are thought to be expressed exclusively in the male germline ([28]; Flybase). Rutabaga (Rut) is the best characterized Drosophila ACs based on a mutagenesis screen for learning and memory phenotypes [29]. Rut is expressed in the Drosophila brain and is stimulated by calcium and calmodulin [30]. However, rut mutants showed normal PDF responses in both M and E cells (unpublished data), suggesting that a different AC(s) must mediate PDF-dependent signaling. Nevertheless, there is evidence that cAMP signaling is involved in circadian physiology in Drosophila [31]. Therefore, to test the role of other AC isoforms in PDF responses in small LNv cells, we performed a transgenic RNAi screen using constructs directed against 11 of the 12 ACs. Initial controls were performed with and without UAS-dicer2, however expression of dicer2 alone showed nonspecific effects on PDF responses and therefore all experiments presented were performed without dicer expression (unpublished data). In M cells, two AC RNAi lines significantly reduced the amplitudes of PDF responses—AC3 and AC76E (Figure 1C)—although in neither case were PDF responses completely abrogated (Figure 1C, compare to second column). In agreement with the initial rut mutant results, RNAi knockdown of rut mRNA had no effect on PDF responses. These results were consistent across different GAL4 lines (Mai179-gal4 and tim(UAS)-gal4) and therefore cannot be ascribed to differences in expression pattern or strength of the specific GAL4 driver used (unpublished data).
The results using AC RNAi could potentially be explained by deleterious effects on small LNv exerted by continuous RNAi expression throughout the neurons' period of development. To evaluate this possibility, we employed a temperature-sensitive genetic system that allows for normal development, followed by conditional induction of RNAi only in the adult fly. Animals raised at a permissive temperature (18°C) had gal4 activity blocked by a temperature-sensitive gal80 transgene (tubulin-gal80ts) [32]. After normal development, the flies were then moved to a higher temperature (29°C) at which the gal80 transgene is no longer active and the gal4 transgene can drive expression of the RNAi construct, as well as the Epac1camps sensor. When tested in this manner, adult-specific knockdown of AC3, but not of AC76E, resulted in a reduction of the PDF response in adult small LNv cells (Figure 2A). This indicates that developmental effects likely cause the reduction observed in the initial RNAi screen for AC76C, while the reduction observed for the case of AC3 RNAi indicates its mediation of PDF responsiveness in adult small LNv pacemakers.
To further confirm AC3 as the candidate PDF-dependent AC and to exclude false positives (due to nonspecific RNAi knockdown), we performed further genetic tests using an independently generated AC3 RNAi line from the Harvard TRiP project (TRiP:AC3RNAi) in addition to the line used in the initial screen from the VDRC (referred to as GD:AC3RNAi) [33] that targets a non-overlapping portion of the AC3 RNA. The TRiP:AC3 RNAi line also produced a significant decrease in the amplitude of PDF responses. In addition, both the VDRC and the TRiP:AC3 RNAi lines were also tested in combination with flies that are deficient for the AC3 gene region (Df(2)LDS6), to further reduce AC3 levels. These RNAi/Df flies (hemizygous AC3 mutants) showed a marked further reduction of the response to PDF neuropeptide in small LNv cells compared to responses in either single mutant genotype: Df/+ or AC3 RNAi/+ (Figure 2B). Together these genetic experiments provide strong confirmation of our initial RNAi screening results and support the hypothesis that AC3 is the principal mediator of PDF-dependent signaling in small LNv cells. Importantly, the consequences of knocking down AC3 were highly specific to PDF: even when combined with the deficiency, AC3 RNAi had no effect on small LNv cell responses to a closely related cAMP-generating neuropeptide—DH31 (Figure 2C) [20]. This indicates that, in these same neurons, DH31-R likely signals through a different AC.
We tested the effects of UAS-rut, -ACXD, -AC76E, -AC3, and -AC78C to ask whether AC over-expression could affect PDF signaling in vivo. Novel constructs were first tested for functionality by measuring cre-LUC responses to 10 µM forskolin in hEK cells. All constructs tested showed an increased average response to forskolin compared to empty vector-transfected cells, although these did not reach significance (Figure S3). We were surprised to find that, in vivo, over-expression of AC3 completely abrogated PDF responses in M cells, while over-expression of all other constructs had no such effect (Figure 3A). This disruption was not due to developmental effects: delaying UAS-AC3 induction until the adult stage after completion of normal development (using the gal80ts system) produced the same disruption of PDF responses (Figure S4). In UAS-AC3 flies, both DH31- and dopamine-elicited cAMP increases remained intact, indicating that the cells were demonstrably healthy and could respond normally to stimulation of other Gs-coupled GPCRs (Figure 3B and C). These observations suggest that abnormally high levels of AC3 specifically disrupt the PDF signaling pathway and add further proof that AC3 is a unique component of PDF signaling in M cells.
Knocking down AC3 levels produced a diminution of PDF signaling in M cells (Figure 2): to evaluate further the specificity of this effect we wished to employ an AC3 rescue strategy. However, over-expressing the AC3 enzyme in M cells above normal levels disrupted M cell responsiveness to PDF (Figure 3A), suggesting that supra-normal levels of the AC3 enzyme can also lead to dysfunction. Therefore, we reasoned that a successful design to rescue the AC3 knockdown would require a more moderate level of AC3 over-expression. Because the gal4 system is temperature-sensitive, intermediate levels of AC3 over-expression were achieved by raising the flies at 25°C and then moving them to 18°C overnight before imaging. This temperature shift could reduce the activity of the gal4 driver, which could result in lower levels of UAS-AC3 expression. Indeed this schedule of temperature changes reduced the disruptive effect of AC3 over-expression on responses to PDF in M cells (Figure 4A, second column). We wondered whether it could also maintain effective RNAi knockdown of the endogenous gene.
We confirmed that firstly the RNAi transgene is still active under this temperature regimen (Figure 4A, third column). This UAS-AC3 RNAi line is directed against the 3′ untranslated region (UTR) of AC3 and can therefore be rescued potentially by expression of UAS-AC3, which includes only the AC3 coding region. In fact, over-expression of UAS-AC3, with a temperature shift from 25°C to 18°C at adulthood, rescued the reduction in PDF responses otherwise observed in a UAS-AC3 RNAi line (Figure 4A). Comparable over-expression of AC78C did not rescue this deficit and that result also confirms that the rescue was not due to simple dilution of the gal4 driver. Importantly, temperature down-shifted (25°C to 18°C) over-expression of AC3 alone, which should result in a small overshoot of normal enzyme levels, shows a slight reduction in PDF responses compared to control (Figure 4A). This again suggested that normal levels of receptor and enzyme are key for normal function. Together these results provide strong evidence to support the hypothesis that AC3 is a specific AC isoform in M cells whose levels are tightly controlled and that normally mediates responsiveness to PDF signaling.
We pursued the AC3 over-expression condition to further evaluate the nature of the components of the PDF receptor signalosome in M cell pacemakers. We reasoned that we could perhaps counteract an imbalance between signaling components produced by AC3 over-expression if we also over-expressed the PDF receptor. In fact, over-expressing PDF-R using a UAS-PDF-R transgene in combination with UAS-AC3 fully rescued the PDF response back to control levels (Figure 4B). The combination of AC3 over-expression with an additional copy of PDF-R (under control of its own promoter within a ∼70 kB transgene, termed PDF-R-myc; [19]) produced a partial rescue of the PDF response. The latter effect was smaller than that seen with UAS-PDF-R, presumably because the induced level of PDF-R over-expression was greater with the UAS construct. Co-misexpression of the closely related neuropeptide receptor dh31-R1 (CG17415; [22]) along with AC3 also gave a partial rescue of diminished PDF signaling due to AC3 over-expression, although these responses were still significantly lower than control and less than what we observed with co-misexpression of PDF-R and AC3. Together, these results suggest that (i) the diminution of PDF signaling that follows AC3 over-expression can be rescued by providing more PDF receptor, thus reducing the receptor/effector imbalance. It also suggests that (ii) the absolute ratio of PDF receptor to AC3 enzyme is important for normal neuropeptide signaling in M cells.
Both RNAi and over-expression screens suggested that PDF receptor associates preferentially to the AC3 adenylate cyclase in M cells, although expression profiling studies indicate that multiple AC isoforms are expressed in these identified pacemakers [34]. To determine the specificity of AC3 contributions to other peptide signaling pathways in M cells, we evaluated cAMP responses produced by other ligands for Gsα coupled GPCRs. Drosophila DH31 (Diuretic Hormone 31) is a neuropeptide closely related to mammalian Calcitonin, and the DH31 receptor (CG17415) is closely related to the Calcitonin receptor [22]. Activation of PDF receptor and DH31 receptor both lead to increases in cAMP and hence both are presumably coupled to Gsα [17],[22]; both increase cAMP in M cells in vivo [20]. RNAi knockdown of the Gsα60A subunit disrupted both signaling pathways, as expected. Interestingly, over-expression of the Drosophila G protein Gsα60A also disrupted both PDF and DH31 signaling in M cells, and responses could be restored by over-expression of the cognate receptor along with Gsα60A (Figure 5A and B). As mentioned above, neither knockdown nor over-expression of AC3 affected DH31 responses (Figure 2C). We interpret these results to suggest that both PDF and DH31 receptors are coupled to Gsα60A, but that PDF-R subsequently signals through AC3 and DH31-R through a different AC.
Scaffolding proteins play important roles in supporting assembly of specific signalosomes, which feature tight association between specific receptors and specific second messenger molecules [35]. We hypothesized that scaffolding proteins may help explain the preference of PDF-R for coupling to AC3. In Drosophila there are four known AKAP (A-kinase anchoring proteins), molecules that bind to and help co-localize many components of cAMP signaling pathways [35]. We tested the possible involvement of AKAPs as scaffolding proteins for PDF-R in M cells using gene-specific RNAi constructs. Knockdown of the AKAP nervy, but not of the other three AKAPs, reduced PDF responses to an extent similar to that produced with the AC3 RNAi (Figure 6A). As with AC3, nervy knockdown showed no effect on DH31 responses in M cells (Figure 6B). When both AC3 and nervy are knocked down together in the same M cell, PDF responses were disrupted to an even greater extent than with either RNAi alone (Figure 6C). The results from single versus double RNAi constructs were generally consistent, although the comparison between TRiP:AC3RNAi and TRiP:AC3RNAi/nervyRNAi does not reach significance (Figure 6C). This suggests that nervy also plays a role in PDF signaling in small M cells, presumably by allowing PDF signaling components to effectively localize and thus promote efficient signaling.
A number of previous studies have suggested that PDF signaling pathways differ between M and E cells. We therefore investigated PDF-R expressing LNd cell (the CRY+/PDF-R+ subset of LNd, using the Mai179-gal4 driver [18],[19],[36]) to evaluate the role of AC3 in E cell PDF signaling. We first confirmed that PDF-induced cAMP responses in these neurons are dependent upon PDF-R; flies with the strong PDF-R mutation (han5304) totally lose E cell responsiveness (as has been previously reported in M cells) [20]. As in M cells, Gsα manipulations again reduced PDF responses in E cells (Figure 7B). Previous experiments confirmed that ACs are involved in E cell PDF responses (Figure S2). The first AC we tested was rutabaga, which had proven ineffective in reducing PDF responses in M cells (Figure 1C). E cells in rut mutants produced normal PDF responses (unpublished data); responses to PDF were likewise normal following rut RNAi expression (Figure 7C). In the case of AC3, neither RNAi knockdown (combined with a AC3 Df) nor AC3 over-expression altered PDF responses in this E-type clock cell subgroup. E cell responses were robust even though these same genetic manipulations produced the most severe reductions in M cell PDF responses (compare Figure 7C with Figure 2B and Figure 3A). Notably, M cell responses were reduced even when measured in the same brains in which E cells proved responsive (unpublished data).
The foregoing data argue that AC3 mediates the cAMP generation produced by PDF in M cells. To what extent is circadian locomotor behavior affected by this disruption of this AC3 activity? Manipulations that partially reduced M cell responses to PDF (e.g., RNAi knockdown of any single AC or AKAP) did not affect locomotor rhythms (see Figure S5 and Tables 1 and 2). However, combining AC3 RNAi knockdown with a deficiency for the AC3 region produced a very strong reduction in the morning anticipation peak, as well as higher levels of arrhythmicity under constant conditions (Figure 8B and Tables 1 and 2). We observed the same effects in UAS-AC3 over-expression in PDF cells (Figure 8C and Tables 1 and 2). Over-expression of UAS-PDF-R and UAS-AC3 together slightly reduced arrythmicity in DD compared to UAS-AC3 alone (Table 2). However, the loss of morning anticipation seen in the UAS-AC3 condition is not rescued by over-expression of the PDF receptor (Figure 8D and Table 1). This suggests that, although the PDF FRET response is rescued (Figure 4B), additional (for example, temporal) aspects of PDF signaling may contribute to normal circadian behavior in LD.
Networks of pacemaker cells are synchronized by intercellular interactions [3],[9],[19]. There is strong and diverse evidence that control of cAMP levels is a critical factor underlying pacemaker rhythmicity and synchronization. Daily changes in cAMP levels in SCN neurons contribute to setting the phase, period, and amplitude of PER2 cycles and thus represent an integral component of the clock mechanism itself [37]. Furthermore, the RGS16 regulator sets the level of cAMP generation and its levels are likewise clock-controlled [38]. Regarding synchronizing agents that couple diverse pacemakers, both PDF in the fly and VIP in the mouse produce cAMP increases in response to receptor activation, and these signals ultimately have access to the pacemaker mechanism in target cells [4],[9],[10],[39]–[41]. Thus understanding the molecular components that control cAMP metabolism in pacemaker neurons, especially those downstream of receptors for the PDF and VIP modulators, are significant goals for the field.
There are at least 12 different genes encoding adenylate cyclases in the fly genome, of which the best known is Rutabaga, a calcium- and calmodulin-sensitive AC. Rut was first identified in a screen for mutations that affected learning and memory exhibited in an associative conditioning paradigm [29]. The Rut cyclase displays the properties of a coincidence detector with its activity triggered by inputs from simultaneous activation of more than one GPCR [24]. However, our studies indicate that, in M pacemakers, the PDF receptor is preferentially coupled not to Rut but to the adenylate cyclase encoded by AC3. In vitro studies suggest the AC3 cyclase may be inhibited by calcium [42]. The functional consequences of this specific signaling association, the physical basis that supports it, and the degree to which it may hold true in other PDF-responsive neurons in the Drosophila brain are important questions raised by this work.
The experiments that manipulated AC and PDF-R expression together indicate that relative levels of AC enzyme and receptor are important determinants of normal PDF cAMP responses in M pacemakers. Counterintuitively, AC3 over-expression was as effective in diminishing PDF responsiveness as was AC3 knockdown. One possible explanation is that the abnormally high levels of AC3 result in incorrect subcellular localization of signaling components, which may preclude the ability of AC3 to contribute to cAMP generation. Within M cells, only moderate expression of a UAS-AC3 transgene could restore normal PDF responses after knockdown of endogenous AC3. Likewise, over-expressing AC3 together with PDF-R could restore the balance between receptor and effector, as indicated by the return of PDF responsiveness. Although these results may not generalize to all cell types or receptor pathways, it is notable that, for this circadian signaling pathway, appropriate levels of signaling components were as important as their simple presence or absence. The reliance on proper stoichiometry between receptor and AC is further evidence to support the hypothesis that PDF-R and AC3 exhibit a specific functional association within the M class of pacemaker cells.
One possible explanation for preferential coupling of PDF-R to AC3 is simply that it is the only adenylate cyclases to be expressed in M cells. However this explanation is inconsistent with at least two notable observations—first, M cells in flies with a severe AC3 knockdown (Df2L;GD:AC3RNAi) still elevate cAMP levels normally in response to neuropeptide DH31. Second, according to recent profiling studies, multiple other ACs are normally expressed at appreciable levels in larval LNs [43] and in adult LNv [34]. Interestingly, these studies indicate that AC3 is not even the most abundant adenylate cyclase [34]. Therefore, we favor an alternative explanation—that molecular specificity dictates the composition of different receptor pathways, with PDF-R residing in privileged association with AC3 (Figure 9).
There is clear support for the concept of preferential coupling between GPCRs and specific ACs in multiple cell types, in addition to our own findings in Drosophila clock cells. Previous work in Drosophila [44] suggests that individual cyclases play specific roles in G-protein signaling associated with gustation. Furthermore, studies of the GABAergic system in the mouse pituitary indicate that Type 7 adenylate cyclase is associated with ethanol and CRF sensitivity, although mRNA for four of the nine mammalian ACs are detected by microarray in pituitary tissue [45],[46]. It has also been proposed that receptor/AC preference may depend upon environmental conditions: for example, the Type 7 preference of the CRF receptor in the mouse amygdala occurs only after phosphorylation of signaling components. Without phosphorylation, CRF receptor couples preferentially to Type 9 adenylate cyclase [47]. Thus, our results add to the body of evidence that highly specific associations between receptors and their downstream partners are key regulators of signaling.
There is clear evidence that signaling components within specific pathways do cluster, which may explain how generalized signaling molecules like cAMP and PKA are capable of targeting distinct downstream effectors. Much current work focuses on possible mechanisms for such localization, [35] and the concept of signalosomes has been proposed to describe the spatial sequestering of signaling pathway components to promote exactly this sort of specific association. Thus preferential AC3/PDF-R coupling may be achieved by localizing AC3 near to PDF receptors. Mechanisms for grouping signaling components may include their co-localization in lipid rafts; many of the components of cAMP signaling including G proteins, PDE, PKA, and cyclic nucleotide gated channels are found in lipid rafts [48] and studies in human bronchial smooth muscle cells detected three different AC isoforms, which are present in distinct membrane microdomains and which respond to different neurotransmitters and hormones [49].
In addition, it is likely that another clustering mechanism includes the formation of macromolecular structures through the use of scaffolding proteins that bind to signaling molecules, as first proposed by Stadel and Crooke [50]. Later studies showed that ACs form large complexes with β-arrestins, G proteins, and calcium channels [51]. The scaffolding protein InaD is required for normal localization of signaling components in the fly visual system including TRP and PLC [52]–[54]. Specialized signaling components such as AKAPs (A-kinase anchoring proteins) can bind to receptors as well as kinases and adenylate cyclases [35]. In Drosophila, AKAPs organize functionally discrete pools of PKA, and disruption of these signaling complexes alters normal spatio-temporal signal integration and causes a loss of anesthesia-sensitive as well as long-term olfactory memory formation in flies [55],[56]. In our study, knockdown of AKAP nervy reduced PDF responses: These results lead to a hypothesis whereby, in M pacemakers, PDF receptor preferentially couples to AC3 via a nervy-based scaffold system to produce normal circadian behavior (Figure 9). We emphasize that, while our results demonstrate a functional connection between AC3 and PDF-R, the basis for any physical connections has not yet been established.
Although our study provides an example of a specific receptor/enzyme pairing in a subset of circadian clock cells, our evidence also suggests the exact details of PDF signaling in other Drosophila pacemakers may differ. Simply put, the set of AC3 manipulations that caused a disruption of PDF responsiveness in M pacemakers had no such effect in E pacemakers. Importantly, disruption of Gsα affected both subgroups (see Figures 5A and 7B). Multiple lines of evidence have suggested that PDF signaling differs between clock cell subgroups. (i) Loss of PDF has distinct effects on PERIOD protein cycling in LNv (M cells) versus non-LNv cells (E cells). Both cell groups continued to show cycling in PER immunostaining levels and localization but, while M cells become phase-dispersed in PER cycles, E cells remain synchronized with altered phase and amplitude of PER accumulation [9],[16]. (ii) In Pdf/cry and PDF-R/cry double mutants, a subset of E cells show a severe attenuation of the PER molecular rhythm, while M cells continue to cycle normally [41],[57],[58]. Different subsets of E cells have previously been implicated in control of evening anticipation, and even when AC3 is altered in all clock cells, the evening peak retains its proper phase, again suggesting that AC3 is not a required enzyme in E type pacemaker cells (unpublished). These finding are consistent with the hypothesis that there are two functionally different PDF signaling pathways. However, although we have confirmed that adenylate cyclases are responsible for the PDF FRET responses in E cells (Figure S2), as yet we have no positive evidence regarding the contribution of any single AC in E pacemakers (unpublished data). Hence it remains to be determined how uniform the components of PDF signalosomes in the M versus E pacemaker cell types are.
How well do the observations obtained with neuronal imaging predict or correlate with circadian locomotor behavior? Manipulations of AC3 that severely disrupt PDF signaling in M cells were correlated with a loss of morning anticipation and increased arrythmicity in DD. Manipulations that only partially reduce the FRET response (for example, single AC3 or single nervy knockdown) resulted in normal circadian locomotor behavior or disruptions to some aspects but not all. The latter observations suggest that the animal is capable of compensating for reduced AC3-generated cAMP responses by M cells but not to complete loss of AC3 function (see Tables 1 and 2 and Figure S5). These data argue for a contribution to behavior by PDF signaling via AC3 in M cells and stand in contrast to a recent report by Lear et al. [10]. That group reported that PDF-R expression in E cells alone is sufficient for morning anticipation and that exclusive expression of PDF-R in M cells does not recover morning anticipation. We cannot reconcile these differences without further experimental efforts, but note that GAL80 techniques are not always sufficient to extinguish gene expression in vivo (unpublished data).
Depending on ambient conditions [12],[57], the M cells contribute to normal morning anticipatory behavior and to maintenance of rhythmicity under constant dark conditions [2],[14],[20],[58]–[60]. However, in our study M cells expressing AC RNAi remain normally responsive to at least two other neurotransmitters (DH31 and dopamine). Hence we suspect that much of the behavioral effect of knocking down AC3 in M pacemakers is mainly due to loss of PDF signaling in them despite retention of additional inputs from a PDF-independent source. Levels of PDF receptor and responsiveness to PDF are both high in small LNv cells and absent (or barely detectable) in large LNv cells [19],[20],[34]. Therefore we expect that AC3 alterations in M cells (directed by Pdf-GAL4) primarily affect PDF signaling in LNvs. In these considerations, the extent to which the AC3 behavioral phenotype is explained by PDF-R coupling to AC3 in M cells is not yet defined. AC3 appears coupled to at least one other GPCR pathway in LNvs because, in DD, AC3 knockdowns produced a more severe behavioral phenotype than did Pdf null flies (a higher percentage of arrhythmicity).
Knockdown of Gsα60A levels of the M pacemakers lengthened the period in DD, a behavioral effect opposite to those seen following loss of PDF, or M cell ablation, namely. Previous studies of Gsα60A in M cells also reported a long period phenotype [43]. Likewise selective expression in small LNv of shibiri (a dominant negative allele of the fly homolog to dynamin [61]) or of a chronically open sodium channel [60] both produce long period phenotypes [62],[63]. Although we cannot rule out a PDF-dependent role in period lengthening in our Gsα60A experiments, our imaging data suggest the lengthened period phenotype may be explained by the fact that alterations of Gsα60A impact multiple signaling pathways (see Figure 5).
Our results demonstrate in Drosophila that, in small LNv (M) circadian pacemakers, a highly specific signaling cascade is activated in response to PDF. They suggest there exists a dedicated PDF-R::AC3-dependent signaling pathway that functions to synchronize these particular clock cells. A different PDF signaling cascade is likely to operate in E pacemakers. The complete molecular details of these signaling complexes, their convergence with CRY signaling [41], and their ultimate connections to the cycling mechanism are significant issues for future studies.
Drosophila were reared on cornmeal/agar supplemented with yeast and reared at 25°C, unless otherwise indicated by experimental design. Male flies (age 2 to 5 d old) were moved to 29°C for 24–48 h before imaging to increase UAS transgene expression. For temperature shift (tubulin-gal80ts) experiments, crosses were maintained at 18°C to maintain gal80ts suppression of gal4, and males were collected and moved to 29°C for 24–48 h before imaging to allow UAS transgene expression. For temperature shift UASAC3/TRiP:AC3RNAi rescue experiments, males were reared at 25°C and moved to 18°C for 12–16 h before imaging to reduce gal4-driven expression of AC3. All gal4 lines used in this study have been described previously: Pdf(m)-gal4 [64], UAS- Epac1camps50A [20], and Mai179-gal4 [65]. The TRiP:RNAi (UAS-TRiP:AC3RNAi, UAS-TRiP:-nervyRNAi, UAS-TRiP:AKAP200RNAi), UAS Gsα60A, UAS-rutabaga, tubulin-gal80ts, and Df(2)LDS6 lines were obtained through the Bloomington Stock Center (thanks to the Harvard TRiP RNAi project) and the UAS-Gsα60ARNAi, UAS-GD:AC3RNAi, UAS-AC13ERNAi, UAS- AC78C, UAS-rutRNAi, UAS-ACXARNAi, UASACXBRNAi, UAS-ACXCRNAi, and UASACXDRNAi. UAS-yuRNAi and UAS-rugoseRNAi lines were obtained through the Vienna RNAi Stock Center.
For epifluorescent FRET imaging, living brains expressing gal4-driven uas-Epac1camps were dissected under ice-cold calcium-free fly saline (46 mM NaCl, 5 mM KCl, and 10 mM Tris (pH 7.2)). All lines tested included one copy each of gal4 (Pdf-gal4 used for small LNv cells and Mai179gal4 for PDF-R(+)LNd cells) and Epac1camps. All genotypes include one copy of each transgene unless otherwise indicated. Full genotypes are available in Table S1. For the RNAi AC screen and for pharmacological experiments, whole brains were placed at the bottom of a 35×10 mm plastic FALCON Petri dish (Becton Dickenson Labware) as in [20], incubated in HL3 saline, and substances tested by bath application. For all remaining experiments, dissected brains were placed on poly-l-lysine-coated coverslips in an imaging chamber (Warner Instruments), and HL3 was perfused over the preparation (0.5 mL/minute). Microscopy was performed through a LUMPL 60×/1.10 water objective with immersion cone and correction collar (Olympus) on a Zeiss Axioplan microscope. Excitation and emission filter wheels were driven by a Lambda 10-3 optical filter changer and shutter control system (Sutter Instrument Company) and controlled with SLIDEBOOK 4.1 software (Intelligent Imaging Innovations). Images were captured on a Hamamatsu Orca ER cooled CCD camera (Hamamatsu Photonics). Exposure times were 20 ms for YFP- FRET and 500 ms for CFP donor. Live FRET imaging was performed on individual cell bodies, YFP-FRET and CFP donor images were captured every 5 s with YFP, and CFP images were captured sequentially at each time point. Following 45 s of baseline YFP/CFP measurement the peptide was bath added/injected into the perfusion line to result in a final concentration of 10−06 M. FRET readings were then continued to result in a total imaging time course of 10 min. ODQ and dopamine were purchased from Sigma. Synthetic DH31was provided by David Schooley and PDF was produced by (Neo MPS, San Diego, CA, USA).
For all experiments reported, we collected responses from at least 10 cells that were found in at least five brains for all genotypes. A region of interest (ROI) defined each individual neuron, and for each, we recorded background-subtracted CFP and YFP intensities. The ratio of YFP/CFP emission was determined after subtracting CFP spillover into the YFP channel from the YFP intensity as in [26]. The CFP spillover (SO) into the YFP channel was measured as .397 [20]. For each time point, FRET was calculated as (YFP−(CFP * SO CFP))/CFP. To compare FRET time courses across different experiments, FRET levels were normalized to initial baseline levels and smoothed using a 7-point boxcar moving average over the 10-min imaging time course. Statistical analysis was performed at maximal deflection from the initial time point by performing ANOVA analysis followed by post hoc Tukey tests using Prism 5.0 (Graphpad Software Inc.).
Over-expression constructs were built by PCR construction from cDNA derived from adult heads (Canton S) and subcloned into P{cDNA3} and P{UAS-attb} vectors. The original AC3 clone was a kind gift from Lonny Levin (Weill Cornell Medical College).
The sequences of all primers used in this study are: AC3(BamHI) 5′: GGATCCATGGAAGCAAATTTGGAGAACGGTC; AC3(EcoRV) 3′: GATATCCTATTCTAGCAAAGACTGACATTCT; AC78C 3′: CTATAACGCATCGTTGTGGCTCTTCGATAT; AC78C nested 3′: ACTTAGACCCAGTGAGTGCGCGTACTCGG ; AC78C 5′: ATGGACGTGGAACTCGAAGAGGAGGAGGAG; AC78C nested 5′: GCATAGCAATAGACAGAATCCTCCGCCACA; AC76E 3′: CTACAATTTCCCATCGAAAGGTGTCTTTAC; AC76E nested 3′: ATCAACAGCAACTGGGTGACGATCGGTGAT; AC76E 5′: ATGGTAAATCACAATGCGGAAACTGCGAAA; AC76E nested 5′: GCCACTAGCTACACGCCACCGCTTTTCGCC; ACXD5′: ATGGACTCCTACTTCGACTCGGCC; and ACXD3′: CTAGTCTTCTTTGGTTGGCGCGGCC.
hEK-293 cells were tested using a cre-LUC reporter in response to 10 µM forskolin 24 h post-transfection with different UAS-AC constructs that had been subcloned into p{CDNA3}. All constructs were co-transfected with cre-luc and compared to empty- vector-transfected cells (0.5 µg cre-luc and 2.5 µg PDF-R and 2.5 µg AC). Four hours after forskolin addition, cells were lysed and luciferin added, followed by bioluminescence measurement using a Victor-Wallac plate reader. Measurements were performed in triplicate and normalized to vehicle-treated controls; the results represent combined activities from three independent transfections.
Male flies were loaded into Trikinetics Activity Monitors 4–6 d after eclosion. Locomotor activities were monitored for 6 d under 12∶12 light/dark and then for 9 d under constant darkness (DD) conditions. Anticipation index was calculated as in [19] as (activity for 3 h before lights-on)/(activity for 6 h before lights-on). To analyze rhythmicity under constant conditions we normalized activity from DD Days 3–9 and used X2 periodogram with a 95% confidence cutoff as well as SNR analysis [66]. Arrhythmic flies were defined by having a power value <10.
|
10.1371/journal.pgen.1005827 | Staufen1 Regulates Multiple Alternative Splicing Events either Positively or Negatively in DM1 Indicating Its Role as a Disease Modifier | Myotonic dystrophy type 1 (DM1) is a neuromuscular disorder caused by an expansion of CUG repeats in the 3' UTR of the DMPK gene. The CUG repeats form aggregates of mutant mRNA, which cause misregulation and/or sequestration of RNA-binding proteins, causing aberrant alternative splicing in cells. Previously, we showed that the multi-functional RNA-binding protein Staufen1 (Stau1) was increased in skeletal muscle of DM1 mouse models and patients. We also showed that Stau1 rescues the alternative splicing profile of pre-mRNAs, e.g. the INSR and CLC1, known to be aberrantly spliced in DM1. In order to explore further the potential of Stau1 as a therapeutic target for DM1, we first investigated the mechanism by which Stau1 regulates pre-mRNA alternative splicing. We report here that Stau1 regulates the alternative splicing of exon 11 of the human INSR via binding to Alu elements located in intron 10. Additionally, using a high-throughput RT-PCR screen, we have identified numerous Stau1-regulated alternative splicing events in both WT and DM1 myoblasts. A number of these aberrant ASEs in DM1, including INSR exon 11, are rescued by overexpression of Stau1. However, we find other ASEs in DM1 cells, where overexpression of Stau1 shifts the splicing patterns away from WT conditions. Moreover, we uncovered that Stau1-regulated ASEs harbour Alu elements in intronic regions flanking the alternative exon more than non-Stau1 targets. Taken together, these data highlight the broad impact of Stau1 as a splicing regulator and suggest that Stau1 may act as a disease modifier in DM1.
| Myotonic Dystrophy Type 1 (DM1) is an inherited disorder affecting many systems, including skeletal muscle, heart, eyes and endocrine system. DM1 is known as a ‘trinucleotide repeat disorder’ because it is caused by an abnormal expansion of a highly repeated motif within the DMPK locus. Such an expansion results in the expression of a ‘toxic RNA’, which causes misregulation of proteins involved in many essential cellular pathways. Research efforts have largely focused on misregulation of a very few splicing regulators that can be linked with many defects observed in the pathology. We have recently uncovered that the multifunctional RNA-binding protein Staufen1 is increased in DM1, and that it is capable of rescuing selected defects in DM1 cells, including alternative splicing of the INSR pre-mRNA, which is linked with insulin resistance. Given the potential impact of this novel function for Staufen1, we investigated the mechanism by which it regulates splicing, and determined that it mediates its effects through binding to conserved genomic repetitive sequences called Alu elements. We also uncovered that Staufen1 influences the splicing of numerous genes in DM1 patient cells, predictive to either improve or worsen the pathology, thus identifying Staufen1 as a novel disease modifier in DM1.
| Alternative splicing of pre-mRNAs is a phenomenon allowing multiple mRNA transcripts to be produced from a single pre-mRNA. Recent reports estimate that 95–100% of human multi-exon genes produce two or more mRNA splice variants, with a majority yielding an average of eight variants [1–5]. Generation of these variants by alternative splicing is a major mechanism responsible for the complexity of the transcriptome and proteome observed in eukaryotes [6]. Constitutive RNA splicing occurs through the recognition of the core splicing signals: the 5' splice site, branch point, polypyrimidine tract, and the 3' splice site AG by the spliceosome components. Additional cis-regulatory elements including exonic splicing silencers (ESS) and enhancers (ESE), and intronic splicing silencers (ISS) and enhancers (ISE) can influence the usage of core splicing signals. Moreover, intronic elements resembling splice sites can act as "decoy" splice sites to influence alternative splicing [7]. There is also a variety of conserved RNA secondary structures that interfere with the recognition of splicing signals and influence splice site selection [8, 9]. The binding of these regulatory elements by RNA-binding proteins can inhibit or enhance the use of core splice sites and results in alternative splicing. Regulation of alternative splicing is thus mediated through the intricate interplay between these cis-acting and trans-acting regulatory elements.
Deciphering the mechanisms that regulate alternative splicing is essential for understanding how cellular diversity and specialization are generated but, importantly, it is also critical to develop novel therapeutic approaches for a growing number of diseases caused by misregulation of pre-mRNA splicing ([10] and refs. therein). For example, in the neuromuscular disorder Myotonic Dystrophy Type 1 (DM1), an expansion of CTG repeats in the 3'UTR of the Dystrophia Myotonica Protein Kinase (DMPK) gene results in retention of CUG-containing DMPK mRNAs within specific RNA foci in the nucleus. The length of CTG repeats varies in DM1 patients and correlates with disease severity [11]. The mutant CUG-expanded mRNA causes a large misregulation of many splicing factor proteins, such as MBNL1, CUGBP1, hnRNP H, ASF/SF2 and RBFOX1 (for review see [12–14]). The misregulation of these splicing factors is reflected in the splicing defects observed in the DM1 pathology. Previously, it has been reported that 13 aberrant splicing events could be linked with the complex DM1 phenotype [15, 16], but more recent studies suggest the existence of numerous additional splicing defects in DM1 tissues [17–19]. One important example of mis-splicing in DM1 is the increase in exon 11 exclusion of the insulin receptor (INSR), which results in the overproduction of the IR-A splice variant, thereby contributing to insulin resistance in DM1 patients [20]. A variety of splicing factors acting on multiple cis-regulatory elements contribute to the splicing control of INSR and of other alternative splicing events. Indeed, a recent report suggests that CUGBP1 and MBNL1 antagonistically regulate hundreds of alternative exons and compete for binding to specific pre-mRNAs [21].
Stau1 is a highly conserved multi-functional double-stranded RNA-binding protein involved in key aspects of RNA metabolism. These include mRNA transport and localization, translation efficiency, stability, regulation of stress granule assembly, and both nuclear and unconventional cytoplasmic mRNA alternative splicing [22–32]. In mammals, Stau1 pre-mRNA is alternatively spliced to produce two major forms Stau155, Stau163 and one variant reported to not bind RNA, Stau1i [24, 33, 34]. Recently, several high-profile studies have focused on elucidating Stau1 binding sites (SBS), which are crucial for understanding Stau1’s ability to regulate mRNA metabolism [31, 35–37]. Extensive work by numerous groups has utilized various immunoprecipitation techniques to investigate SBS, which appear to be represented by a highly diverse group of RNA secondary structures. These include double-stranded RNA structures containing stems and motifs ranging in size from 5–22 base pairs (bps) to hundreds of bps long which, in turn, can contain multiple short binding sites with varying degree of perfect base pairing, displaying little to no sequence specificity [31, 35–37]. Notably, in all large-scale studies performed to date, Stau1 has been reported to bind preferentially to the primate-specific, mobile element called Alu elements. SBS, comprised from both Alu and non-Alu element containing sequences, have been found everywhere in the genome including 3'UTRs, 5'UTRs, intronic regions, coding sequences and intergenic regions [31]. This diversity of SBS location highlights the potential complexity surrounding events regulated by Stau1.
Recently, our group identified Stau1 as being significantly increased in muscle samples from adult DM1 patients and DM1 mouse models [32]. Additionally, we saw that further overexpression of Stau1 in DM1 was able to rescue key hallmarks of the pathology, such as increased export and translation of CUG-expanded mRNAs and a significant increase in INSR exon 11 inclusion [32]. Interestingly, our study revealed for the first time the ability of Stau1 to regulate alternative pre-mRNA splicing suggesting a novel role for Stau1 as a splicing regulator [32]. These data lead us to speculate that the upregulation of Stau1 represents a positive and protective adaptation in the DM1 pathology.
Here, we first set out to determine the mechanism by which Stau1 regulates pre-mRNA alternative splicing. Second, we examined the broader impact of Stau1 as a splicing regulator in the context of DM1. We report that Stau1 regulates the alternative splicing of human INSR exon 11 via binding to a region harbouring Alu elements within intron 10. Additionally, using a high-throughput RT-PCR screen, we identified numerous Stau1-regulated Alternative Splicing Events (ASEs) in both WT and DM1 myoblasts. These Stau1-induced changes in ASEs are expected to be beneficial or detrimental for the DM1 pathology. Importantly, a higher number of Stau1-regulated ASEs harbour Alu elements in intronic regions flanking the alternative exon when compared to non-Stau1 ASE targets. We thus propose that Stau1 uses Alu elements to regulate a large set of ASEs and that it acts as a disease modifier impacting on the severity of DM1.
We recently reported that Stau1 overexpression rescues specific alternative splicing defects associated with DM1, including that of exon 11 in the INSR pre-mRNA [32]. Moreover, our observation that Stau1 also promotes exon 11 inclusion in muscle cells in the absence of pathological RNA repeats suggests that Stau1 may be a bona fide splicing regulator. In order to explore this idea further, we first assessed whether Stau1 could affect INSR exon 11 alternative splicing in non-muscle cells. First, HeLa cells were transiently transfected with a Stau1-HA expression construct and the relative level of endogenous INSR exon 11 inclusion was determined using semi-quantitative RT-PCR. A high level of exon 11 inclusion was observed in these cells which agrees with previous findings [38]. Similar to our previous work in C2C12 myoblasts [32], the overexpression of Stau1-HA, as confirmed by Western blotting with HA-antibodies, resulted in a small, but reproducible ~5% increase in exon 11 inclusion (Fig 1A). To address whether Stau1 is required to maintain normal levels of INSR exon 11 inclusion, we also assessed INSR splicing event when levels of Stau1 are reduced. This would also mitigate the possibility that our observed effects on splicing were due to spurious, non-specific RNA binding of overexpressed Stau1-HA. HeLa cells were thus transiently transfected with a Stau1-targeting shRNA mix (described in materials and methods) and Western blotting was performed to assess Stau1 protein levels. This analysis demonstrated a 40% reduction of Stau1 protein level compared to CTRL (Fig 1B). This reduction of Stau1 levels caused a significant ~10% decrease in the relative inclusion of INSR exon 11 (Fig 1B). To confirm further the role of Stau1 as a splicing regulator in non-muscle cells, we extended our work to include an additional cell line, namely, HEK293Ts. In agreement with our findings in HeLa cells, the overexpression and reduction of Stau1 resulted in a significant ~5% increase and ~10% decrease, respectively, of exon 11 inclusion (S1A Fig).
The splicing of exon 11 is known to be regulated by a number of splicing factors, including, but not limited to MBNL1, CUGBP1, and hnRNP H [39–41]. These splicing proteins, similar to Stau1, are misregulated in DM1 [20, 42–44]. Thus, it was important to determine if our data are indicative of a direct effect of Stau1 on exon 11 splicing regulation or an indirect effect mediated through modified expression of other splicing regulators. In both HeLa and HEK293T cells, no significant changes were observed in the mRNA or protein levels of MBNL1, CUGBP1 and hnRNP H, upon modulation of Stau1 levels (Fig 1C and 1D and S1B Fig and S1C Fig). Finally, as we have previously confirmed, Stau1 over-expression did not differentially affect the mRNA half-lives of INSR alternatively spliced variants [32]. Thus, altogether, our results strongly suggest that Stau1 is a bona fide splicing regulator, participating in the maintenance of human INSR exon 11 alternative splicing profile.
In order to gain insights into the mechanism by which Stau1 regulates INSR exon 11 alternative splicing, we first searched for possible SBSs, which could represent cis-regulatory elements within the INSR pre-mRNA. Recently, numerous reports have emerged describing SBSs and although no single SBS has been described, one of the most highly recurring SBS reported is composed of Alu Repeat elements [31, 35–37, 45, 46]. Interestingly, an earlier report described the presence of an Alu Repeat element located in intron 10 of the INSR [47]. Closer inspection of this region via a bioinformatic analysis revealed that, in fact, there are three Alu elements located upstream of the intron 10-exon 11 boundary (Fig 2A). The fact that Alu elements are preferred SBS [45], together with the presence of Alu elements in intron 10 of the INSR, led us to propose that Stau1 may bind to these Alu elements to regulate alternative splicing of exon 11. To test this hypothesis, we selected two IR-minigene constructs [40, 41, 47]: WT and ΔAlus (in which all three Alu elements are deleted) (Fig 2A). HeLa cells were co-transfected with Stau1-HA and either the WT or the ΔAlus IR-minigene. RIP experiments were performed on cell lysates and RT-qPCR using primers specific for the intronic sequence of intron 10 was performed to identify the amount of IR-minigene pre-mRNA bound to Stau1-HA (Fig 2B). The amount of Stau1-HA bound to WT IR-minigene demonstrated a significant ~2-fold enrichment over the IgG control (Fig 2B; black bars). This degree of association was greatly reduced with the ΔAlus IR-minigene (Fig 2B; crosshatch bars), consistent with the hypothesis that Stau1 binds to the pre-mRNA of the INSR via the Alu elements located in intron 10.
Next, we investigated whether the Alu elements were necessary for Stau1-regulated splicing of exon 11 by carrying out Stau1 overexpression and knockdown experiments. First, HeLa cells were co-transfected with a Stau1-HA expression or shStau1 construct and one of the IR-minigenes; WT or ΔAlus. Overexpression of Stau1-HA induced an ~5% increase in exon 11 inclusion in the WT IR-minigene (Fig 2C), as determined by RT-PCR as above. The reduction in Stau1 led to an ~12% decrease in exon 11 inclusion in the WT IR-minigene (Fig 2D). Importantly, in the absence of the Alu elements in the IR-minigene, neither overexpression nor reduction of Stau1 resulted in a significant change in exon 11 inclusion (Fig 2C and 2D, respectively). Thus, our data demonstrate that the Alu elements located in intron 10 are essential for Stau1 splicing regulation of exon 11 of the IR-minigene.
The observation that Stau1 regulates the splicing of the INSR led us to examine whether Stau1 regulates the splicing of additional pre-mRNAs. To address this central question, we carried out a screen using a high-throughput RT-PCR screen that measured the changes in the splicing ratios of 487 selected events (as described in [18]) in MyoD-converted WT and DM1 myoblasts either overexpressing GFP or Stau1-HA. The 487 ASEs comprising the RT-PCR alternative splicing screen were chosen based on their association with the specific key terms: "muscle", "glucose metabolism", "wasting", and "ion-channel". WT (GM03377) or DM1 fibroblasts harbouring 1700 CTG (GM03132) repeats in the 3'UTR of the DMPK gene, were converted to myoblasts. Briefly, the conversion was done via two rounds of infection over 48 hours with a retrovirus engineered to express MyoD, followed by selection with Puromycin (1 μg/mL) for 5 days. Cultures were then infected with either GFP- or human Stau1-HA lentiviral particles. GFP expression was used to confirm infection efficiency 48 hours post-infection (S2A Fig). Semi-quantitative RT-PCR and Western blot analysis using MyoD specific primers and MyoD antibodies confirmed the overexpression of MyoD mRNA and protein (S2B Fig and S2C Fig). Moreover, Western blot analysis with anti-HA tag antibodies confirmed Stau1-HA overexpression (S2D Fig). Total RNA from the MyoD-converted WT and DM1 cells was then isolated and used to carry out the high-throughput RT-PCR splicing screen.
Initial heat map data from the RT-PCR screen revealed that the overexpression of Stau1 in both WT and DM1 conditions had a broad effect on the splicing profile of numerous ASEs through the observed alteration of the Percent Splicing Index (PSI) for each ASE (Fig 3A and 3B). Although several ASEs were not affected by Stau1-HA overexpression (e.g. ITGA7a; Fig 3B), a number of ASEs showed important changes in both WT and DM1 conditions (e.g. INSR; Fig 3B). To determine quantitatively whether Stau1 increase regulates the PSI of an ASE, Stau1-HA overexpression was compared to GFP CTRL from WT and DM1 cell lines, yielding a value referred to as the change in PSI (ΔPSI). A threshold of a ΔPSI ≥10% was established to denote relevant changes in splicing regulated by Stau1 (Fig 3C and 3D, S1 Table). Altogether, data from the high-throughput RT-PCR screen demonstrate that overexpression of Stau1-HA affects the splicing of 75 and 88 ASEs in WT and DM1 cell lines, respectively, with 27 ASEs common in both conditions (Fig 3D). Similar trends were seen in both WT and DM1 conditions where the majority of Stau1-regulated ASEs showed a ΔPSI between 10–30% upon Stau1-HA overexpression. Accordingly, only a few ASEs showed large (≥50%) ΔPSI when Stau1-HA was overexpressed, compared to GFP CTRL, such as INSR and NRG1 (Fig 3E and 3F). These data suggest that Stau1 does not dramatically alter the PSI of the majority of Stau1-regulated ASEs, but instead appears to fine-tune the alternative splicing of many ASEs. These results confirm our hypothesis that Stau1 is a splicing regulator and further show that Stau1 levels can alter the splicing profile of numerous pre-mRNAs both in WT and DM1 conditions.
We recently demonstrated that overexpression of Stau1 in DM1 conditions induced an increased inclusion of the INSR exon 11 [32]. This splicing modulation should be beneficial for DM1 as it reverts the INSR aberrant splicing towards WT conditions, an event that would also be predicted to reduce insulin resistance in patients. This prompted us to investigate whether the splicing events regulated by Stau1 overexpression were all beneficial for the DM1 pathology. For this analysis, only ASEs which showed a change in splicing pattern ≥10% from WT to DM1 conditions were considered. ASEs that shifted back towards WT splicing patterns when Stau1-HA was overexpressed were considered beneficial. Conversely, an ASE was considered detrimental if the overexpression of Stau1-HA in DM1 conditions exacerbated the splicing pattern observed in the pathology, i.e. opposite direction of WT. Using these criteria, 25 ASEs were classified as beneficial, whereas 8 ASEs would be potentially detrimental upon Stau1-HA overexpression in DM1 patient cells (Fig 4A and 4B). This suggests that promoting Stau1-regulated splicing in DM1 could potentially have both beneficial and detrimental effects depending on the specific alternative splicing event considered. Taken together, these results demonstrate that Stau1 is a splicing factor that regulates a broad range of splicing events and highlights the importance of Stau1 as a potential disease modifier for DM1.
We extended the results collected from our RT-PCR splicing screen by generating a comparison between ASEs altered in DM1 identified in the current study relative to those documented in a recent report by Klinck and colleagues [18]. Using the same thresholds set by Klinck et al., we identified 54 ASEs that appeared altered in the DM1 conditions that were common to both screens, including previously described events such as the INSR, ANK2 and various chloride channels (Fig 4C and S1 Table). The ASEs that were distinct between the two screens may be due to the different cell lines used in each independent study. More specifically, normal adult satellite muscle cells were used in the study by Klinck et al., and MyoD-converted myoblast cell cultures were used in our study. These differences in cell culture models may account for the variability in the DM1 associated ASEs identified. We also compared the ASEs regulated by Stau1-HA, to ASEs regulated by MBNL1 or RBFOX1 described by Klinck and colleagues. A total of 22 and 19 ASEs were identified that appeared to be co-regulated by either Stau1 and MBNL1 or RBFOX1, respectively (Fig 4D and S1 Table). A total of eight ASEs were identified as being co-regulated by all three splicing proteins, including INSR. A comparison of the direction of splicing regulation of Stau1 to either MBNL1 or RBFOX1 revealed that 51% and 63% of the ASEs co-regulated by Stau1 and MBNL1 or RBFOX1, respectively, proceeded in the same direction (Fig 4E). The fact that Stau1 regulates the same splicing events as MBNL1 and/or RBFOX1 suggests that Stau1 may act as both an agonist and antagonist to other splicing factors.
Of the ASEs affected by Stau1 over-expression in the high-throughput screen, we validated by semi-quantitative RT-PCR 13 of 19 (68%). Among the 13 ASEs that were identified as Stau1-regulated events by our screen and independently shown to be regulated by Stau1 by semi-quantitative RT-PCR validation, we found INSR, hnRNP A2B1, LRRC23, HIF1α, NRG1, FN1, ACCN3, FHL3, G6PC3, CLCN2 and CLCN6 (Fig 5A–5D and S3A–S3G Fig). Splicing analysis of four ASEs were extended to include two additional DM1 myogenic cell lines with varying numbers of CTG repeats; 500 CTGs and 50–80 CTGs (Fig 5A–5D and S2B Fig and S2D Fig). These additional DM1 cell lines were included to investigate the influence of Stau1-HA, which was overexpressed at relatively equal amounts (S2D Fig), on splicing regulation in varying degrees of the DM1 pathology. As expected, exon 11 inclusion of the INSR decreased with an increase of CTG repeats, and in all cases, Stau1-HA overexpression increased exon 11 inclusion by ≥15%, independent of the number of CTG repeats (Fig 5A). Taken together, the data suggest that Stau1 regulates the splicing of numerous ASEs even in cases of varying degrees of severity of the DM1 pathology.
Given that Alu elements in the upstream intron flanking the alternative exon 11 of the INSR is required for both binding and splicing regulation by Stau1 (Fig 2), we examined how many other Stau1-regulated ASEs contained one or more Alu elements as compared to non-Stau1 regulated ASEs. Briefly, we first mapped out the genomic regions corresponding with each primer pairs used in the RT-PCR splicing screen using UCSC Genome Browser (human genomic data v.37). Next, utilizing RepeatMasker (release v.4.0.6), the number of Alu elements that were present within each genomic region of the ASEs was determined. We then compared the number of Stau1-regulated ASEs that harboured Alu elements to non-Stau1-regulated ASE targets. In total, 80.5% of Stau1-regulated targets (ΔPSI ≥15%) contained Alu elements. In contrast, only 65.7% of non-Stau1 regulated targets contained Alu elements (Fig 6A). A similar value of 68.3% was obtained when considering the whole dataset. Focusing on the Stau1 ASEs that contained Alu elements, we again used RepeatMasker to identify the subfamilies of the Alu elements in order to see whether any particular Alu subfamilies were prevalent in Stau1 ASE targets. This analysis revealed no obvious preference toward a specific subfamily of Alu element in the introns flanking Stau1 ASEs (Fig 6B). Further analysis comparing the proportions of the major subfamilies, i.e. AluY, AluS, and AluJ, identified in our study revealed a similar distribution of Alu family proportions to those reported in primate genomes [48, 49]. It has been observed for a number of splicing factors that their recruitment either upstream or downstream of the alternatively spliced cassette-exon correlated with whether they were promoting inclusion or skipping of that exon [21, 50]. To determine if this seemed to be the case for Stau1, we examined the distribution of Alu elements relative to the alternative cassette-exon amongst top Stau1-regulated ASEs. This analysis revealed that the 35% of Alu elements were found within the upstream introns, relative to 17% in downstream introns, while the remaining 48% were found in both flanking introns (Fig 6B). However, this distribution did not seem to correlate strongly with whether Stau1 induced exon inclusion or skipping for those ASEs.
Since Stau1 has also been proposed to bind to various non-Alu RNA secondary structures [35, 37], the predicted secondary structures of the flanking introns of three non-Alu containing Stau1-regulated ASEs, were scanned for RNA secondary structures that resembled possible SBS. This additional analysis revealed the presence of secondary structures resembling known or predicted SBS within the proximity of the ASEs of hnRNP A2B1, LRRC23, and NRG1 (S4A–S4C Fig). These may serve as SBS allowing Stau1 to regulate the splicing of these pre-mRNAs when no Alu elements are present.
We report here that Stau1 regulates the alternative splicing of INSR exon 11 through its interaction with Alu elements located in intron 10. Using a high-throughput alternative splicing screen, we also demonstrate that Stau1 regulates a broad range of pre-mRNAs, many of which also harbour Alu elements within introns flanking the ASE. Importantly, although Stau1 overexpression in DM1 myoblasts did rescue splicing patterns of many pre-mRNAs towards WT, as previously observed for INSR exon 11, a number of Stau1-induced splicing changes were also found shifting away from WT patterns, and thus would be predicted to exacerbate the DM1 pathology. Taken together, these findings are consistent with the notion that Stau1 may act as a disease modifier in DM1.
In the present study, we demonstrated that modulation of Stau1 levels regulates INSR exon 11 splicing. This regulation was demonstrated in two non-muscle (HeLa and HEK293Ts) cell lines and complements our previous findings in DM1 muscle cell lines [32]. Modulating Stau1 levels did not result in major changes in the expression of other splicing factors known to regulate exon 11 inclusion, such as MBNL1, CUGBP1 or hnRNP H. Furthermore, our previous work found no protein-protein interactions between Stau1 and MBNL1 or CUGBP1 [32]. Taken together, these findings support the idea that Stau1 regulates splicing without direct protein interactions or modulation of the expression levels of these other key splicing factors. Nevertheless, Stau1 may still affect the functional activity of these splicing factors in ways that are more indirect.
Alu elements serving as cis-regulatory elements for splicing have been described. For example in the RABL5 pre-mRNA, two Alu elements that were in opposite orientation in the upstream intron were shown to affect the splicing patterns of the downstream exon 3 [51]. Another example is the regulation of the alternative splicing in the Ataxia Telangiectasia Mutated (ATM) pre-mRNA. In this case, an intronic splicing element derived from an Alu element was found to modulate the inclusion of a cryptic exon [52]. In a number of cases, binding of trans-acting factors to Alu elements is required to mediate an effect on splicing [46, 53]. For example, Zarnack and colleagues demonstrated that, in the absence of hnRNP C, thousands of Alu elements were included as exons in mRNA transcripts [53]. Splicing analysis of IR-minigenes in HeLa cells demonstrated a drastic shift in the splicing pattern to exon 11 inclusion occurred when the Alu elements in intron 10 were deleted. These findings agree with previous literature reporting similar results in HepG2 cells [47]. Thus, we propose that the presence of Alu elements serves to inhibit the inclusion of exon 11, perhaps through the recruitment of one or more Alu-element binding trans-acting factor(s), in addition to Stau1, that would then interfere with recruitment of constitutive splicing factors (see model in Fig 7). It may be informative to examine any potential interactions between Stau1 and known Alu-binding splicing proteins to further define the mechanism by which Stau1 regulates alternative splicing.
Our results show that increased Stau1 levels correlate with increased inclusion of exon 11. The specific details of what occurs once more Stau1 binds to the Alu elements remains unclear. As depicted in Fig 7, this could somehow prevent the inhibitory effect of the Alu repeats on exon 11 inclusion, either by preventing recruitment of Alu binding factors or through recruitment of a distinct factor that would instead favour inclusion. Conversely, upon Stau1 knock down, this would allow for the full inhibitory effect of the Alu elements on exon 11 inclusion (Fig 7). A number of additional factors have been reported to regulate alternative splicing of exon 11 (e.g. hnRNP H, DDX5, etc) [54] and we cannot rule out a mechanism whereby Stau1 binding to the Alu elements may modulate the action of one or more of these factors. Finally, in the context of DM1, where we know the positive contribution of MBLN1 is lost and the inhibitory effects of CUGBP1 are enhanced, the moderate increase in Stau1 levels that we have documented in patient tissues, may not be enough to completely prevent the inhibitory contribution of the Alu elements on exon 11 inclusion (Fig 7). Alternatively, Stau1’s activity might be misregulated in DM1 (e.g. through post-translational modification(s) and/or interaction with distinct factors). Further experimentation will be required to determine the precise mechanism(s) by which Stau1 contributes to INSR splicing under both normal and DM1 settings. Nevertheless, to our knowledge, this is the first example showing that the binding of Stau1 to Alu elements can regulate alternative splicing.
Strikingly, our high-throughput screen has revealed that Stau1 can regulate alternative splicing to an extent that is comparable to well-established splicing regulators like MBNL1 and RBFOX1 (see Fig 4D). An analysis of all ASEs examined in this study revealed that Stau1-regulated ASEs were more likely to harbour one or more Alu elements than those that were not Stau1-sensitive (80.5% vs. 65.7%, respectively). Amongst Stau1-regulated pre-mRNAs Alu elements were found within both introns flanking the ASE for ~50% of cases. Recent studies that investigated SBS uncovered that Stau1 preferentially binds complex, imperfectly paired duplex structures formed by the pairing of at least two Alu elements [37, 55]. Thus, it would be interesting in future work to investigate whether Alu elements found on either side of an ASE need to form duplexes in order to influence splicing decisions. Genome-wide occupancy assessment of splicing regulators, using, for example, HITS-CLIP or similar approaches, has revealed that binding of splicing regulators either upstream or downstream of an ASE is often correlated with whether it will mediate a positive or negative effect on that splicing event [50]. Using a limited dataset of ASEs most significantly affected by Stau1, we determined that when Alu elements (and thus potential SBSs) were not present in both flanking introns, they were most often found in introns upstream of the ASE (~twice as often than in downstream intron; see Fig 6). However, we found no correlation between Alu elements being present upstream or downstream of ASEs and whether the effect of Stau1 on that ASE was the induction of skipping or inclusion. We did not document within this limited dataset Alu elements positioned in very close proximity or overlapping with splice sites, but this is obviously another aspect that would need to be considered. These questions will require analysis of larger datasets in order for more conclusive patterns to emerge.
Although a majority of Stau1 ASEs harboured Alu elements in introns flanking the ASE, we observed that ~20% ASEs affected by Stau1 level modulation did not contain any Alu elements. Further analysis into the RNA secondary structure of non-Alu containing introns flanking ASEs revealed the presence of RNA duplex structures, which may represent potential SBSs. Interestingly, the three potential non-Alu SBSs in the Stau1 targets hnRNPA2B1, NRG1 and LRRC23 (S4 Fig) are all located close to the 3’SS, which could thus interfere with recognition of this site by the basic splicing machinery; a phenomenon that Stau1 could then either promote or interfere with. Several pre-mRNAs containing ASEs that were sensitive to Stau1 level modulation have been previously identified as Stau1 targets. Specifically, Stau1 binds to these targets through non-Alu element binding sites located in the coding sequence of the transcripts, including: ADPGK, AKT2, ANK1, APOBEC3C, ARF1, ARFRP1, ENSA, FN1, JAG2, NUTF2, OGDH, SHKBP1, SORBP1, TBC1D12/13, and THRA [31, 36, 37].
Over the past decades, the major emphasis in defining the molecular pathogenesis of DM1 has focused on the role of a very few specific RNA-binding proteins, such as MBNL1 and CUGBP1, in aberrant alternative splicing events in DM1. Although animal models of MBNL1 and CUGBP1 [56, 57] do reproduce many DM1 symptoms [43, 58], some are not recapitulated, suggesting that other factors, such as disease modifiers are involved. Indeed, several disease modifier proteins have recently been identified and shown to have an impact on the DM1 pathology. For example, in a Drosophila DM1 model the RNA-binding proteins TBPH (homolog of human TAR DNA-binding protein 43 or TDP-43) and BSF (Bicoid stability factor; homolog of human LRPPRC) were found be misregulated with the expression of CUG expansions resulting in altered muscle sarcomere location of these proteins [59]. Another study, done by Huin and colleagues, reported that several genetic variants of the MBNL1 gene could be associated with the severity of the disease, suggesting that these variants were acting as disease modifiers in DM1. [60]. Finally, the DEAD-box RNA helicase, DDX5/p68, found to be reduced in DM1 biopsied skeletal muscle [61], was shown to allow increased MBNL1 binding to mutant repeats which can influence splicing events misregulated in DM1 [62].
In agreement with our initial study [32], the data presented here show that overexpression of Stau1 resulted in several splicing events predicted to be beneficial for DM1, such as the rescue of the INSR exon 11. However, we also identified a number of detrimental splicing effects, which would likely exacerbate the DM1 pathology (see Fig 4A and 4B). This suggests that the upregulation of Stau1 may not represent a protective role in the DM1 pathology as previously suggested but, instead, shows Stau1 likely acts as a disease modifier for DM1 whose splicing impact can result in both beneficial and detrimental effects on the DM1 phenotype. Additionally, it is possible that Stau1 may act as a disease modifier in DM1 through non-splicing related activities. For example, we have recently shown that Stau1 negatively regulates skeletal muscle differentiation, at least in part through its regulation of c-myc translation [63]. As such, Stau1 may thus contribute to the impaired differentiation/developmental program observed in DM1 [64]. The precise overall contribution of Stau1 to the DM1 phenotype thus remains to be fully explored, but our results to date strongly indicate that Stau1 needs to be considered amongst the gene products that modulate the complex DM1 pathophysiology and its response to future therapeutic interventions.
The following cell lines were obtained from the NIGMS Human Genetic Cell Repository at the Coriell Institute for Medical Research: WT cell lines were represented by two cell lines with 0–5 CTGs repeats: GM03377 (Splicing screen) and GM01653 (Validation). DM1 cell lines used were GM03132 (1700 CTGs), GM03987 (500 CTGs) and GM03991 (50–80 CTGs). HeLa cell lines were obtained from ATCC (ATCC CCL-2) and HEK293-T cell lines were obtained from ATCC (ATCC CRL-1573).
Constructs: IR-minigenes WT and ΔAlus (aka IR-B and IR-E) were generously donated by Nicholas Webster and previously described in [47], MyoD virus (pBRIT-MyoD-(His-TEV-3FLAG)), GFP virus, hStau15-HA plasmid (described in [24]), shStau1 plasmid mix made up of two shRNAs against human Stau1 mRNA (referred to in text as shStau1) (Open Biosystems GE Dharmacon: sh06 [Clone ID:TRCN0000102306] + sh09 [Clone ID:TRCN0000102309]). Cells to be transfected were grown to ~70% confluency and transfected with 1–3 μg of DNA using Lipofectamine with plus reagent (Life Technologies:15338100), according to manufacturers protocol, for 48 hours. Virus production consisted of using plasmids previously described [32] pcDH-CMV-MCS-EF1-copGFP and pcDH-Stau1. Viral particles were produced by transient transfection of HEK-293T cells with lentiviral packaging vectors psPAX2 (Addgene:12260) and pMD2.G (Addgene:12259) using Lipofectamine 2000 reagent (Life Technologies:11668027) according to manufactures protocol. The conditioned medium containing viral particles was collected and used to transduce control and DM1 myoblasts overnight in the presence of 8ug/ml Hexadimethrine Bromide (SIGMA:H9268). Subsequent infections were performed the following day, and cells were grown for several days before analyses. Infection of cells involved growing cells until ~70% confluency, infecting first with MyoD virus, selection with Puromycin (Wisent Bio Products:400-160-EM) (1μg/mL) for 5 days, infection with either the GFP or hStau155-HA virus, confirming GFP expression after 48 and harvesting cells for RNA and protein 72 hours after initial second infection. Cells were washed with 1XPBS, scraped and lysed in 1 mL of RIPA buffer with Protease Inhibitor added prior to use. Cells with incubated in RIPA buffer for 30 min on ice and centrifuged for 15 minutes at 13,200 rpm. The supernatant was collected and stored at -20°C until use.
Following cell lysis, protein concentration was assessed using the Bio-Rad DC Protein Assay (Bio-Rad:500–0111) and protein (2–40 μg) was resolved by denaturing polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to polyvinylidene fluoride (PDVF) membranes (Immunobilon Transfer Membranes:IPVH00010). Transferred membranes were blocked with 5% milk for 30 minutes and probed with appropriate antibody in 1% milk solution for either 1 hour at room temperature or 12 hours at 4°C, with three 10 min washes with 1XPBS-0.05% Tween 20 between each antibody incubation. Antibodies included: Anti-Stau1 [1:1000] (Abcam:ab73478), Anti-GAPDH [1:10,000] (Abcam:ab8245), Anti-β-Actin [1:500] (Santa Cruz:sc-47778), Anti-CUGBP1 [1:1000] (Santa Cruz:sc-20003), Anti-hnRNP H [1:5000] (Abcam:10374), Anti-MyoD [1:300] (BD Pharmingen: 554130), Anti-MBNL1 antibody [1:300] (Abnova:H00004154), Anti-HA F7 probe [1:1000] (Santa Cruz:sc-7392). Secondary antibodies included: Mouse-anti-Rabbit HRP [1:20,000] (Jackson ImmunoResearch:211-032-171) and Goat-anti-Rabbit HRP [1:10,000] (Molecular Probes:MP 02764). Proteins on membranes were detected with Millipore-Luminata Crescendo Western HRP Substrate (WBLUR0500) and visualized on film (HyBlot CL Autoradiography Film:E3018).
RNA was isolated from whole cell lysates using Ambion TRIzol Reagent (E3018) and 2μL of collected RNA was assessed on the Take3BioPlate Reader to determine quantity (ng/μL) and quality (RNA with 260/280 ~ 2.0 was used). 500 ng of RNA was used to synthesize cDNA with random hexamers (10mM) and the Promega AMV cDNA synthesis (Promega:M5101) was carried out following manufacturer's protocol. cDNA was diluted 1:20 and 5μL (~100 ng) was used for each RT-PCR and RT-qPCR reaction. All cDNA and RNA was stored at -20°C short-term and -80°C long term.
RT-PCR was performed using Promega GoTaq DNA Polymerase (Promega:M5101) according to manufacturer's protocol. RT-PCR conditions were as follows for validation of splicing screen: 95°C for 2 min, (95°C for 30 sec, 55°C for 30 sec, 72°C for 45 sec)x32 cycles, 72°C for 10 min. Specific RT-PCR conditions used for particular primers are available upon request. Amplicons were run on a 2% agarose gel (containing 3–5 μL of EtBr 20 mg/mL) and visualized under UV light. All RT-qPCR reactions were performed using BioRad iQ SYBR Green Supermix (BioRad:170–8882) according to manufactures protocol and run with a Chromo 4 Real-Time PCR Detector. RT-qPCR conditions for all primer sets used were carried out as follows: 95°C for 2 min, (95°C for 30 sec, 60°C for 30 sec, 72°C for 45 sec)x40 cycles, 72°C for 10 min. Technical replicates of 3 were done for all RT-qPCR experiments and the average Ct values were normalized to either GAPDH or 18S (indicated in descriptions). The ΔΔCt method was used to analyze fold change of transcripts. Primers for INSR splicing analysis were previously described [32], and any of the 487 primer sets used in our RT-PCR splicing screen are available upon request. Biological replicates of ≥3 samples were done for all PCR reactions.
RNA from (WT) GM03377 and DM1-1700 CTG (GM03132) was synthesized to cDNA and subject to the screen as previously described [18]. Raw data from the screen is included in S1 Table. Analysis of data to determine top ASE included rankings of events that had the greatest PSI change (Δ) between two conditions, i.e. the ΔPSI between WT+GFP virus and WT+Stau1 virus:
ΔPSI=PSIGFP-PSIStau1-HA
(1)
Only values with a ΔPSI ≥10% between conditions were selected for additional analysis:
[(PSI(WT)GFP-PSI(DM1)GFP)(-(n)2)]*100≥10%
(2)
ASEs which contained PSI values with no data (no isoforms detected), were not included in analyses.
A minimum of three biological replicates were used to validate a change in the splicing patterns (ΔPSI) for all cell lines tested. Semi-quantitative RT-PCR was carried out with the required primers to obtain two isoforms describing an ASE for the splicing screen. Splicing patterns in each condition (WT+GFP, WT+Stau1-HA, DM1+GFP, DM1+Stau1-HA) was analyzed and classified as successfully tested and validated if a change in splicing pattern was detected in all three biological replicates and followed the same splicing pattern as predicted by our screen for all replicates (n = 3). ASEs were classified as successfully tested but not validated if the splicing pattern detected in all three biological replicates did not change between conditions or did not follow the splicing pattern predicted by the screen.
48 hours after transfection cells were treated with 1% formaldehyde to induce cross-link in vivo Stau1-HA-RNA complexes for 10 minutes at RT and reaction was quenched with 0.25 M glycine in PBS. Cells were suspended in 1 mL of RNase-free RIPA buffer and centrifuged for 15 minutes at 13, 200 rpm at 4°C. The supernatant was collected and centrifuged twice more. 40 μL of Santa Cruz Protein A/G PLUS agarose beads (Santa Cruz:sc-2003) suspended in RNase-free RIPA buffer was added to lysate was incubated with gentle rotation for 1 hour at 4°C to pre-clear. Beads were removed by centrifugation and the pre-cleared lysate was aliquoted by volume into 10% input, IgG and IP. Normal mouse IgG antibody (Santa Cruz:sc-2025) or mouse-anti HA antibody (4 μg) was added to IgG or IP, respectively, and incubated for 16 hours at 4°C with gentle rotation. 40 μL of A/G plus beads were added to IgG and IP samples and incubated for 1 hour at 4°C with gentle rotation. Stau1-RNA complexes bound to beads were pelleted at 2,500 rpm for 30 sec, supernatant removed and the pellet was resuspended in 1 mL of RNase-free RIPA buffer. These washes were repeated three times. Following final resuspension of pellet, crosslinking was reversed (1 hour at 70°C). Trizol was then directly added to the bead-RIPA solution and RNA isolation protocol was followed (as described above).
Primer pairs in the high-throughput RT-PCR splicing screen were used in a BLAST to identify the mRNA transcript ASE that Stau1 was suspected to regulate. Using CLC MainWorkbench, alignments of the DNA and mRNA transcripts revealed the exon(s) and flanking introns, which defined the ASE. Flanking introns were then analyzed with RepeatMasker (v.4.0.6) to identify the presence and subfamily of Alu element(s). Exons that made up the ASE were also analyzed for the presence of Alu elements however, none was found at that time. A total of 23 cassette exon type Alu element containing Stau1-regulated ASE targets were examined to identify the location of the Alu(s), either upstream or downstream of the ASE. These were then categorized by whether Stau1 overexpression induced exon inclusion or skipping of the cassette exon. If an intronic sequence did not contain any Alu elements, the MFE structure of the intronic sequences were manually searched for potential duplexes resembling SBS. These potential SBS were identified based on previous reports describing identified SBS, [36, 37] i.e. average size of stem length (duplex) (between 5–19 base pairs), high base pair probability and varying degrees of imperfect base pairing (preference given to longest continuous duplex structure formation). If the intronic sequence was >7,500 bps (current limit of RNAfold partition function calculations), then intron was divided into equal segments, each ≤7,500 bps, and subsequent predicted RNA secondary structure was used.
To identify statistical significance between two groups one-tailed student's t-tests were carried out on data with biological replicates n≥3. The level to determine a value as significant was set as p<0.05. Significance was denoted as follows: * p<0.05, ** p<0.01, ***p<0.001.
|
10.1371/journal.pgen.1006483 | Loss of RMI2 Increases Genome Instability and Causes a Bloom-Like Syndrome | Bloom syndrome is a recessive human genetic disorder with features of genome instability, growth deficiency and predisposition to cancer. The only known causative gene is the BLM helicase that is a member of a protein complex along with topoisomerase III alpha, RMI1 and 2, which maintains replication fork stability and dissolves double Holliday junctions to prevent genome instability. Here we report the identification of a second gene, RMI2, that is deleted in affected siblings with Bloom-like features. Cells from homozygous individuals exhibit elevated rates of sister chromatid exchange, anaphase DNA bridges and micronuclei. Similar genome and chromosome instability phenotypes are observed in independently derived RMI2 knockout cells. In both patient and knockout cell lines reduced localisation of BLM to ultra fine DNA bridges and FANCD2 at foci linking bridges are observed. Overall, loss of RMI2 produces a partially active BLM complex with mild features of Bloom syndrome.
| Cells contain specific protein complexes that are needed to correct errors during the replication and segregation of DNA. Impairment in the activity of these proteins can be detrimental to the viability of the cell and organism development. Bloom syndrome is an example of a genome instability disorder where cells cannot efficiently untangle DNA after replication. The only gene that is known to cause Bloom syndrome is the BLM helicase. In this article, we describe two affected individuals with Bloom-like features with a homozygous deletion of the RMI2 gene. The RMI2 protein has previously been shown to form a complex with BLM, topoisomerase III alpha and RMI1. Deletion of RMI2 in patient and unrelated cell lines show hyper-recombination and chromosome entanglements during cell division. Furthermore, we show that the BLM and FANCD2 proteins are diminished in the binding of DNA bridges that need to be dissolved during the late stages of cell division. Therefore, loss of RMI2 produces a milder Bloom phenotype and impairs the full activity of the BLM complex.
| Bloom syndrome (BS) is a very rare genetic disorder with features of significant growth deficiency, hypo- and hyperpigmented skin, sun-sensitive facial skin lesions, cancer predisposition in early life and male infertility [1,2]. Early cytogenetic experiments revealed clues about the underlying mechanism with patient chromosomes exhibiting hyper-recombination and genome instability [3]. The only known gene, BLM, associated with BS was identified in 1995 [4]. The gene encodes for the BLM protein that is a member of the RecQ DNA helicase family of proteins. RecQ helicases are essential for genome maintenance and are conserved across evolution.
Protein interaction studies have shown that the BLM protein is a member of a four-subunit complex that includes topoisomerase III alpha (TOP3A) [5,6] and RecQ-mediated genome instability proteins 1 [7–9] and 2 [10,11] (RMI1 & 2), collectively known as the BTR complex. The BTR promotes the dissolution of double Holliday junctions that can be formed during DNA replication into non-crossover products in a two-step process: 1) by pushing the Holliday junctions together by the helicase activity of BLM, and 2) the dissolution of hemi-catenated DNA by the cleavage and joining activities of TOP3A [12]. Crossover events between homologs in somatic cells can be detrimental to a cell’s survival as they lead to loss of heterozygosity (LOH) [13,14,15]. Notably LOH is elevated in BLM deficient cells [16]. Moreover, unresolved recombination intermediates that persist into mitosis lead to bridging and are a source of genomic instability [17]. Structure and function studies have shown that RMI1 and 2 form a heterodimer that is important in stabilising the BTR [18,19]. The BTR complex has been proposed to localise to stalled replication forks via interactions with Fanconi anaemia (FANC) subunits and Replication Protein A [20]. This super-complex is also known as BRAFT. Similar to BS, Fanconi anemia patients exhibit growth deficiencies, chromosomal breaks, heightened genomic instability and cancer predisposition [21]. Further evidence to support this connection is through structural analyses with a FANCM peptide and the RMI1-RMI2 heterodimer [22].
The FANC core complex consists of eight subunits that promote the monoubiquitination of FANCD2 and FANCI in response to sites of DNA damage where replication forks are obstructed [23,24]. FANCD2 acts at stalled replication forks to remove interstrand cross-links (ICLs) and additionally regulates homologous recombination proteins including BRCA2/FANCD1 [25–27]. BLM is known to cooperate with FANCD2 during S phase to restart stalled replication forks while also suppressing the firing of new replication origins; an activity that is independent of FANCI [28]. During mitosis, FANCD2 and FANCI subunits frequently appear at the sister chromatid anchor sites that link DAPI-negative chromatin threads also known as ultra fine bridges (UFBs) and also occasionally along the UFBs during anaphase [29,30]. FANCI/D2 sister foci in mitosis appear at chromosome arms and not centromeres and their localisation corresponds to fragile sites in the genome [29]. The foci that link UFBs during chromosome segregation imply a tethering or loading function for proteins that coat UFBs such as BLM and PICH [31,32], but this is yet to shown.
BLM, TOP3A and RMI1 are highly conserved in most eukaryotes but RMI2 is absent in some lineages including invertebrates and yeasts, suggesting that it is needed in organisms with higher genome complexity [11]. Further evidence to support RMI2’s functional role in higher eukaryotes was shown in chicken DT40 RMI2 null cells which display elevated levels of sister chromatid exchanges (SCEs). At a cellular level, whether RMI2 is required during mitosis and at an organism level, its role during development and disease predisposition are all outstanding questions. Here, we show a homozygous deletion of RMI2 in two siblings with milder clinical features of Bloom syndrome. We have additionally mutated RMI2 using CRISPR-Cas9 gene-editing to further confirm the hyper-recombination and mitotic defect phenotypes observed in the patients’ cells.
The two affected siblings are the only children of first cousin parents of Pakistani descent. Their mother had one previous miscarriage but there was no other relevant family history. The clinical features of both siblings are summarised in Table 1.
Sibling 1 (S1 Fig, Fig 1A), a male, was born by caesarean section at 36 weeks gestation. Birth weight was 2.7 kg (50th centile) but other growth parameters were not recorded. He was noted to have large numbers of café-au-lait macules in the first year of life. The café-au-lait macules were mostly <1cm in diameter, but several were >5cm (Fig 1A). There were no other features of neurofibromatosis type 1. There were also several depigmented macules. He was otherwise healthy with normal growth and development, and was an average student at school. At age six years his height was 118 cm (50th-75th centile), weight 28.8 kg (75th-90th centile) and head circumference 50.0 cm (50th centile). There was no cutaneous photosensitivity, reduction in subcutaneous fat, or feeding difficulties or recurrent infections. Neurological, cardiac, respiratory and abdominal examinations were normal and he did not have the characteristic facial appearance of Bloom syndrome. Full blood examination, electrolytes, blood glucose, liver function, and immune function were normal but alpha fetoprotein was mildly elevated.
Sibling 2 (S2), a female, was born at 37 weeks gestation following a pregnancy that was complicated by slow growth in the 3rd trimester. She was delivered by caesarean section and birth weight was 2.2 kg (<10th centile). Other growth parameters were not recorded. There were no neonatal complications. Gastro-esophageal reflux was diagnosed at age one month and was treated medically until age ten months. Multiple café-au-lait macules were noted in the first year of life with a similar pattern to her brother. Her growth was mildly delayed and she was microcephalic: at age four years her height was 95.0 cm (5th centile), weight 15.0 kg (25th centile) and head circumference 45.5 cm (1 cm below 2nd centile). S2 was otherwise healthy and hearing, vision, voice and development were normal. There was no cutaneous photosensitivity or reduction in subcutaneous fat and there was no history of recurrent infections. Neurological, cardiac, respiratory and abdominal examinations were normal and she did not have the characteristic facial appearance of Bloom syndrome. Full blood examination, electrolytes, blood glucose, liver function, and immune function were normal but alpha fetoprotein was mildly elevated.
The mild growth abnormalities of S2 and the presence of café-au-lait macules in both S1 and S2 suggested an underlying genetic defect and a chromosome microarray was requested for the affected family members. The microarray analysis in both siblings demonstrated long continuous stretches of homozygosity consistent with the parents being first cousins. A homozygous deletion was detected comprising 80 kb at chromosome band 16p13.13, resulting in the deletion of the entire RMI2 gene and the micro RNA gene, MIR548H2 (Fig 1B and S1 Fig). Both parents were heterozygous for the same deletion. Of note, BLM was not within a region of homozygosity.
To identify the exact breakpoint region, oligonucleotides were designed adjacent to the closest positive array probe at each breakpoint. Long-range PCR produced a band of approximately 6 kb for both affected children, whereas an unrelated control displayed no fragment. Sequencing of the cloned PCR product revealed a non-allelic recombination event between two Alu repeat elements, without any loss or gain of Alu sequences (S2 Fig). The deletion therefore covers a region of 84,871 bp located at chr16: 11,304,701–11,389,571 (hg38) (Fig 1C). Aside from RMI2, the deleted region contained no other coding genes.
The two Alu repeat elements share an overall sequence identity of 80% spanning 308 bp. Interestingly, a continuous stretch of 38 bp showing 100% identity between the repeats crossed the breakpoints. The deleted region does not span any copy number variable region and contains no known segmental duplication of >1000 bp as displayed on the UCSC Genome Browser.
Routine G-banding analysis on lymphocytes showed no gross chromosomal rearrangements. S1 was 46,XY in 15 metaphase cells, and S2 was 46,XX in 15 metaphase cells. Solid staining for chromosomal breaks in 100 metaphase cells revealed a higher rate in the affected siblings. 15 and 5 chromosome or chromatid breaks were identified in S1 and S2, respectively (S3 Fig). Furthermore, the presence of quadriradial chromosome formations were not observed, which are present in around 2% of Bloom syndrome cells [33]. Control lymphocytes showed no detectable chromosome breaks.
To confirm the cytological phenotype of elevated sister chromatid exchange events that is characteristic of Bloom-like syndrome, fresh peripheral blood lymphocytes were prepared for differential sister chromatid staining. Both affected siblings and two sex and age-matched controls were assayed microscopically for sister chromatid exchanges. 15 cells from each individual were examined and showed a mean of 40 and 36 chromatid crossovers for S1 and S2, respectively, compared with a mean of five crossovers for controls (Fig 2).
To examine the extent of chromosome entanglements in mitosis, fibroblast cell lines were established from the siblings and parents. These cell lines enabled a number of cytological analyses to be performed. Fibroblasts were grown on coverslips and then fixed and stained with DAPI. The presence of micronuclei are a useful biomarker for chromosomal breaks and missegregation events [34]. The number of cells containing at least one micronucleus was 4.8% and 7.4%, S1 and S2, respectively (Fig 3A). By contrast, the parents showed 1.5% and 0.89%, for P1 and P2, respectively. This equates to a 5 to 8-fold higher frequency of micronuclei in the siblings fibroblast cells.
Other features of mitotic errors were also measured. Chromatin threads or bridges connecting interphase nuclei were 0.10% and 0.28% for P1 and P2, respectively, compared with 1.7% and 1.9%, S1 and S2, respectively (Fig 3B and 3C). Larger masses of chromatin in the form of bulky DNA bridges were 0.22% and 1.5% for P1 and P2, respectively, compared with 7.5% and 9.3% for S1 and S2, respectively (Fig 4D, S6 Fig). Although both siblings share the exact same homozygous deletion spanning RMI2, overall S2 was more affected than S1 across several mitotic assays. The differences between S1 and S2 however were not statistically significant. This is consistent with her (S2) more severe clinical presentation and growth defects when compared against her brother (S1).
In order to confirm the cytological results observed in the fibroblast cells lines, we chose to create independent isogenic knockout cell lines in the near-diploid human colorectal cell line, HCT-116 using CRISPR/Cas9 gene editing. To minimise off-target mutations we adopted a double-nicking strategy [35]. Two separate guide oligonucleotide pairs were used to generate several candidate RMI2 knockout cell lines. Two knockout cell lines (1–2 and 1–3) from guide pair 1AB and one cell line (4–6) from guide pair 4AB were used in subsequent functional characterisation. Details of the mutations are provided in (S4 Fig). The three knockout cell lines were confirmed to be null for the RMI2 protein with immuno-blot (Fig 4A). These cell lines provided an independent additional line to support and expand observations from fibroblast patient cells.
The HCT-116 rate of SCEs per cell was 6.6 per cell compared with a combined average of 34 for the three KO clones (Fig 4C and S5A–S5D Fig). This equates to a 5.2-fold increase in the RMI2 knockout cell lines. Anaphase bridges showed four-fold increase in frequency when compared to wild-type cells. Whereas, lagging chromosome frequency displayed a modest increase over wild-type cells (Fig 4D and S6 Fig). Together, data from HCT-116 replicate findings from our patient fibroblast cell lines, with RMI2 null cells showing increased chromosome bridges compared to controls. In both experiments using patient fibroblast and HCT-116 cells there was no significant increase in cells showing lagging chromosomes, suggesting RMI2 and the BTR complex does not play a role in spindle-kinetochore attachment. DNA content analysis was performed on exponentially growing asynchronous cells from fibroblast and HCT-116 cell lines to determine if there was any polyploidy or aberrant cell cycling. No differences were observed between wild-type and RMI2 null cell lines (S7 Fig).
We were interested to ascertain whether RMI2 loss affected cell proliferation and UV sensitivity. A previous study in DT40 null cells had shown no effect on cell proliferation rates or sensitivity to DNA damaging chemicals [11], whereas another study using RNAi knockdown in human cells had observed a lower survival rate in cells challenged with MMS [10]. To assess whether the loss of RMI2 had an impact on cell proliferation and colony forming ability, 300 cells were plated onto dishes and grown for six days before being fixed and analysed. The RMI2 null cells showed a 2.4-fold and 8.9-fold decrease over parental wild-type cells for number of colonies and the total area that the colonies occupied per well, respectively (S5E and S5F Fig).
To test whether the RMI2 null cells were sensitive to UV light, 300 cells were plated per well and allowed to recover for one day before being exposed to UV light. The average number of colonies in the RMI2 null cells dropped to 27% of untreated cells, compared with a similar drop of 27% for untreated wild-type cells (Fig 4E). We also challenged the cells with the DNA replication inhibitor, hydroxyurea (HU) at varying doses (Fig 4F). No consistent sensitivity was observed in the knockout cell lines.
Our study analyzed bulky DNA bridges and found a significant increase due to the absence of RMI2 (Fig 4D, S6 Fig). Another class of bridge that is associated with the BTR complex activity are ultra fine bridges (UFBs) that are finer, thread-like structures not detectable using DAPI. BLM is one of several proteins that co-localise with UFBs in the later stages of mitosis with members of the BTR appearing as a streak between separating chromosomes most commonly during early anaphase [32]. UFBs occur naturally in mitosis and although the precise function of BTR in cells undergoing chromosome segregation is still to be determined, it is thought the complex aids sister chromatid decantation during anaphase [36]. It is presumed that UFBs associate with loci that contain un-replicated DNA or unresolved recombination intermediates that persist into mitosis [17]. Relevant to this study, patients with Bloom syndrome show significantly elevated levels of UFBs as a result of defective BLM [32].
To test whether RMI2 null cells also showed significant increases in UFBs, HCT-116 wild type and null RMI2 cells were stained with Plk1-interacting checkpoint helicase (PICH) protein (Fig 5), which colocalises with BLM on UFBs during anaphase [31,32]. PICH is considered a useful marker of UFBs as its localisation is independent of the BTR. The results were striking and paralleled analogous scoring in BLM disrupted cells [32]. We found that there was only a slight increase in anaphase A cells displaying PICH fibers between wild-type and RMI2 null HCT-116 clones (Fig 5A and 5C). However, by anaphase B approximately only 30% of wild type cells shows detectable PICH fibers, while the approximately 80% in null HCT-116 RMI2 null cells (Fig 5B and 5D). The data clearly indicate UFBs persist into anaphase B as a result of RMI2 disruption.
BTR complex members are a set of several proteins that co-localise with UFBs in the later stages of mitosis. We therefore analysed core components to determine if their localizations were altered in the absence of RMI2. BLM appears as a streak on UFBs between separating chromosomes, with BLM fibers most evident during early anaphase [32]. Although the precise function of BLM in cells undergoing chromosome segregation is still to be determined, it is thought the complex aids sister chromatid decantation during anaphase. Furthermore, it is presumed that UFBs associate with loci that contain un-replicated DNA or unresolved recombination intermediates that persist into mitosis [17], however the precise nature of the DNA is yet to be described. Examination of BLM localisation on anaphase fibroblast cells revealed little difference compared to controls in the prevalence of positively-staining fibers from both affected siblings and RMI2 HCT-116 null cells in anaphase B (Fig 6B and 6E). What was apparent however was the intensity of BLM (using pooled data from anaphase A and B) on the fiber was significantly weaker in RMI2 null cells in both patient and HCT-116 systems compared to controls (Fig 6A, 6C, 6D and 6F). Interestingly, in anaphase A there was small, but statistically insignificant drop in detection of BLM-positive fiber in HCT-116 RMI2 null cells relative to wild-type and an even larger decline in the analogous experiment in using patient fibroblast lines (S8 Fig).
We next examined TopoIIIα localisation onto UFBs in anaphase (Fig 7). We adopted a slightly different approach and measured the amount of anaphase B cells that showed PICH and TopoIIIα colocalisation. Our previous data (Fig 5) showed approximately 30% of anaphase cells had PICH fibers, so we asked the question how many of these PICH fibers show colocalisation with TopoIIIα. The results were very clear. For wild type HCT-116 cells 94% of anaphase B cells with PICH overlapped with TopoIIIα compared 26%, 18% and 13% for RMI2 HCT-116 null clones 1–2, 1–3 and 4–6 respectively. Together the results show RMI2 is necessary for the proper localization of the BTR complex members BLM and TopoIIIα, and provide a mechanistic link why UFBs persist during anaphase B in RMI2 null cells i.e., due to disruption of BTR subunits in anaphase.
The Fanconi anaemia (FANC) complex is needed for the repair of DNA ICLs generated during DNA replication [23]. Subunits of the FANC and BTR complexes interact together forming a super-complex known as BRAFT [22]. Furthermore, the FANCD2/FANCI subunits forms foci at regions of replication stress such as common fragile sites that anchor the BLM-staining fibers between segregating sister chromatids [17]. The FANCD2/FANCI foci on separating chromatids are visible from anaphase through to telophase [29]. We also noticed, FANCD2 can occasionally appears as a fiber across separating chromatids, reminiscent of the BLM and PICH (S9 Fig). We have examined the localisation of FANCD2 on UFBs in the family’s fibroblasts and the HCT-116 RMI2 null cells. Both cell types show a decrease in the frequency of anaphase to telophase cells containing FANCD2 foci on sister chromatids (Fig 8 and S9 Fig). Additionally, the intensity signals of the FANCD2 foci on the HCT-116 RMI2 null cells show a decrease in signal ranging from 1.9- to 2.4-fold. Taken together, these results suggest the stability of the BRAFT super-complex encompassing BLM and FANCD2 subunits is compromised through loss of RMI2.
We have identified a homozygous deletion of the RMI2 gene that results in a Bloom-like phenotype from a consanguineous kindred. The two affected siblings exhibit a variable phenotype with some overlapping features of Bloom syndrome. Sibling, S2, presented with growth deficiency and gastro-esophageal reflux, traits commonly found in Bloom syndrome children. Curiously, these indicators were absent in sibling S1 (Table 1). It is too early to tell whether homozygous deletion of RMI2 is associated with elevated risk of cancer in late childhood or adulthood. Consistent with the clinical presentation, our cell biology analyses also indicated sibling S2 was slightly more affected with mitotic assays for chromosome bridges in mitosis, bridges persisting between interphase nuclei and micronuclei all elevated compared to sibling S1. The striking feature consistent with a Bloom syndrome phenotype, is both children display café-au-lait macules (Fig 1A). These dermatological findings are often associated with childhood cancer syndromes [37]. Loss of RMI2 should therefore be added to the differential diagnosis of children presenting with multiple café-au-lait macules.
Cytogenetic investigation into genome instability showed a higher rate of SCEs and chromatid breaks (Fig 2). BLM null individuals have a 10-fold elevation in the rate of SCEs when compared to wild-type cells [3]. By contrast, we have observed a slightly lower rate at seven- to eight-fold above wild-type. This is consistent with a similar decrease when SCE rates are compared between BLM and RMI2 knockout chicken DT40 cells [11]. These data suggest that the BLM helicase displays partial activity in dissolving catenated DNA in the absence of RMI2. Indeed, in vitro addition of RMI2 to BLM-TopoIIIα-RMI1 caused a statistically significant increase in the rate and overall level of dissolution of radiolabelled double Holliday junction substrates [10]. The same study found the BLM-TopoIIIα-RMI1 complex alone still possesses significant dissolution activity; suggesting RMI2 enhances but is not essential to the enzymatic capability of the BTR complex. However, we note our in vivo analyses show removal of RMI2 has a profound affect on the stability of BLM on UFBs that likely represent a range of replication intermediates. Understanding the DNA structure of UFBs remains an important task that will provide insight into why and how UFB-associated proteins act.
The notion that BLM-TopoIIIα-RMI1 alone can still dissolve double Holliday junctions is consistent with the observation that BLM patients show a more noticeable clinical presentation compared to RMI2 affected individuals. Additionally, BLM has activities that are independent of RMI2. For instance, it is known BLM stimulates the resection activity of human exonuclease 1 [38]. It is therefore also likely that with the increase in DNA analysis capabilities and also clinical awareness that further RMI1 and also RMI2 affected individuals will be identified in the population.
The failure to dissolve catenated DNA in the affected siblings is the main trigger for downstream mitotic errors such as DNA anaphase bridges and micronuclei (Fig 3 and Fig 4). These perturbations in mitosis are thought to have an impact on the cell proliferation rate. We have investigated whether there was any link between mitotic errors and growth rates in the affected siblings but no consistent association could be found. Homozygous knockout of the RMI2 gene in HCT-116 cells showed a noticeable slowing down in cell proliferation and the ability to form colonies from single cells (Fig 4B and S5E and S5F Fig). This is in contrast to the chicken DT40 RMI2 knockout cell lines that did not display any reduction in cellular growth rate, although colony forming assays were not performed [11]. Correspondingly, one of the affected siblings showed prenatal and postnatal growth deficiency (Table 1). The variability in cell proliferation rates and impact on development is most likely to be dependent on genetic background. Furthermore, the growth deficiency phenotype observed in sibling S2 may be due to a co-existing disorder associated with the parents’ consanguinity. RMI2 mouse knockout studies will hopefully shed some light on these differences between model systems.
DNA repair disorders are commonly associated with sensitivity to DNA-damaging agents such as chemical mutagens or short wave radiation such as UV light. Bloom syndrome affected individuals are mildly sensitive to sunlight where they display sun-sensitive lesions on exposed areas such as the face [1]. However, there are conflicting reports in the literature whether BLM null cells are sensitive to UV light in vitro [4,39,40]. The affected siblings did not show any signs of sun-sensitive skin lesions on exposed areas. HCT-116 RMI2 null cells also did not show any consistent reduction in the number or size of colonies after being exposed to short-wave UV light or hydroxyurea when compared to parental wild-type cells (Fig 4). Similar results were observed in chicken DT40 RMI2 null cells when challenged with DNA-damaging chemicals such as cisplatin or methyl methanesulfonate [11]. Taken together, these data support that the BLM helicase can perform some of its DNA repair functions without the participation of RMI2.
Earlier experiments on the knockdown of RMI2 from vertebrate cells had not examined its role during chromosome segregation. Examination of RMI2 null cells in fibroblasts and HCT-116 lines has shown hallmarks of mitotic errors in the form of DNA bridges and micronuclei (Fig 3 and Fig 4). Were these chromosome entanglements due to the lack of the BTR complex localising to UFBs? We have shown both BLM (Fig 6) and TopoIIIα (Fig 7) localisation is disrupted when RMI2 is removed. Our data however show that BLM still can localise to UFBs, albeit at a significantly lower intensity. Together, this evidence supports our hypothesis that the BTR is functionally impaired during mitosis without RMI2.
Further evidence of this partial activity is illustrated with localisation experiments of FANCD2 in RMI2 null cells. Like BLM, FANCD2 sister chromatid foci are reduced in frequency and intensity, suggesting that BTR instability impacts upon important DNA repair complexes such as FANC (Fig 8). This is not without precedent as it is known that BLM co-immunoprecipitates with FANCD2 in human cells [41], and mechanistically FANCD2 and the BTR complex cooperate to restart stalled replication forks [28,42]. These studies suggest a physical and mechanistic interplay between BTR and FANCD2 in S phase under replication stress and the dependencies seemingly persist through to M phase. Curiously ours (Fig 8 and S9 Fig) and another study [29] observed FANCD2 coated UFBs. The nature of FANCD2 UFBs has not been fully explored, but it is possible they exist as backup or additional activity in resolving catenated DNA structures during anaphase.
The BLM Bloom syndrome gene was first identified over 20 years ago. Our report shows that this is the first clinical description of individuals with Bloomoid features of non-BLM subunit. Although producing a clinical presentation similar to Bloom syndrome, the hallmark features are not as severe. Independent studies using cell lines derived from homozygous affected siblings and also HCT-116 cells deleted of RMI2 both show overlapping defects with marked increase in DNA bridges during the later stages of cell division. Our data show removing RMI2 affects the stability of interacting partners in the BRAFT super complex with BLM and FANCD2 reduced on chromosomes during chromosome segregation. Significantly, those cells without RMI2 that displayed BLM fibers in anaphase showed a marked drop in signal intensity, suggesting RMI2 stabilises or activates the complex. Whether there is any residual activity of BLM and how the overall subunit composition and architecture of the BTR and BRAFT complexes is affected is not yet clear. These will be important questions for future studies. Our current study suggest both at the patient and cell biology level the effects are not as severe as lacking BLM altogether.
Family members were recruited to this study with the approval of the Hospital Research Ethics Committee at the Royal Children’s Hospital, Melbourne, Australia, ethics approval number, 28097. Written consent for the affected individuals was provided by their parents.
Genomic DNAs were isolated and purified from leukocytes using the NucleoSpin Tissue genomic DNA extraction kit (Machery-Nagel, Germany). DNA samples were processed by the Illumina Infinium method using the HumanCytoSNP—12 v2.1 (Illumina, San Diego, CA, USA) microarray platform and analysed using KaryoStudio v1.4 software (Illumina). Confirmation of the null deletions and cascade testing of the parents was performed using Affymetrix CytoScan 750K array using the manufacture’s protocols and analysed using Chromosome Analysis Suite vCytoB-N1.2.2.271 (Affymetrix, Thermo Fisher Scientific).
Fibroblast and HCT-116 cell lines were cultured in BME and RPMI, respectively. Media were supplemented with 10% FBS and penicillin/streptomycin.
Primers were designed next to the closest positive microarray probe on either side of the breakpoints. The following oligonucleotides (IDT), RM-delf (5’—CCTACTCCTCCTGCCCTTTTC—3’) and RM-delr (5’—CCTGCCTCTTTACCTGGAGTG—3’) were used in a long-range PCR amplification reaction using Phusion Hot Start II (Thermo Fisher Scientific) with the following conditions; 98°C 2 min (1 cycle), 98°C 30 sec, 61°C 30 sec, 72°C 3 min (40 cycles), 72°C 10 min (1 cycle). PCR products were A-tailed with AmpliTaq Gold DNA polymerase (Thermo Fisher Scientific) 72°C 10 min, and cloned into pGEM-T Easy (Promega) using standard methods. The plasmid insert was Sanger sequenced using primer walking at the Australian Genome Research Facility, Melbourne, Australia.
Fresh blood cells were incubated for three to four days in RPMI 1640 media/10% FBS with 20 μg/ml phytohaemagglutin. BrdU (Sigma-Aldrich) was added to a final concentration of 10 μg/ml for 30 hours followed by 0.1 mg/ml colcemid (Thermo Fischer Scientific) treatment for 45 mins before standard metaphase chromosome harvest. HCT-116 cell lines were treated for 29 hours with 10 μg/ml BrdU, followed by 0.1 mg/ml colcemid for 1.5 hours. Phosphate buffer pH 6.8 was added to cover the dried slides to a depth of 2 mm. Slides were then placed in a biosafety cabinet and were exposed to UV light at a distance of 30 cm for 45 min. The slides were briefly rinsed in dH2O and added to prewarmed 2 x SSC at 65°C for 30 min, followed by another rinse in dH2O and stained in Leishman’s stain (Sigma-Aldrich).
RMI2-null HCT-116 clones were seeded onto 6-well dishes at 300 cells in three ml of media per well in triplicate for each cell line. The next day the cells were exposed to either 2 mJ UV (254 nm) or mock treatment using a GS Gene Linker UV Chamber (Bio-Rad). Cells were then grown for six days and then rinsed in PBS, fixed in ice-cold methanol and stained in crystal violet solution. The 6-well dishes were imaged and colonies of at least 0.032 mm2 were counted using ImageJ v2.0.0.
Cell extracts preparation for immunoblotting was performed as described before [43]. In brief, cells were collected and washed once with cold PBS. The pellets were resuspended in RIPA buffer with fresh prepared EDTA-free protease inhibitor (Roche) and incubated on ice for 15 min and then sonicated. Protein concentration were determined using the Quick Start Broadford Protein Assay (Bio-Rad). 40 μg of total protein extract from each of the samples was run on 10% SDS PAGE gels (Bio-Rad). The following antibodies were used for immunoblot detection, rabbit polyclonal anti-RMI2 (1:1000) (Abcam), mouse monoclonal anti-α-tubulin antibodies (1:1000)(Sigma-Aldrich), swine anti-rabbit IgG-HRP (1:10,000)(Dako) and rabbit anti-mouse IgG-HRP (1:10,000)(Dako). ECL immuno-blotting substrate (Pierce) was used according to the manufacturer’s instructions.
Fibroblasts or HCT-116 cells were seeded onto gelatinised 22 mm x 22 mm glass coverslips in 6-well trays. After at least 24 hours, media was removed and cells were rinsed in PBS. For immunofluorescence cells were fixed with 4% paraformaldehyde for 10 minutes, permeabilised with 0.3% Triton X-100 and blocked with 3% BSA in PBS. Cells were stained with rabbit polyclonal anti-BLM (1:500)(Abcam), rabbit monoclonal anti-FANCD2 (1:500)(Abcam), mouse-monoclonal anti-PICH (Millipore, 1:200), rabbit polyclonal anti-TopoIIIα (kind gift from the Hickson laboratory, University of Copenhagen, 1:200) and mouse monoclonal anti-α-tubulin antibodies (Sigma, 1:500). Secondary antibodies were donkey anti-rabbit Alexa Fluor 488 (1:1000)(Invitrogen) and goat anti-mouse Alexa Fluor 594 (1:1000)(Invitrogen). Cells were mounted with VectaShield containing DAPI (Vector Laboratories).
For sister chromatid exchange and breakage analyses, methanol-acetic acid fixed preparations were imaged using a Zeiss Axioplan 2 microscope with a 100× objective lens. Images were analysed using AxioVision 4.7 (Zeiss). For FANCD2 images taken by DeltaVision, 36 sections (0.2 μm per section) images were taken. Images were deconvolved, and projected in 2D using SoftWoRx 4.1. Percentage of cells with symmetrical FANCD2 spots were scored and plotted. Obvious symmetrical FANCD2 spots intensity were further measured using the polygon function of SoftWoRx 4.1. For BLM fibers scoring and intensity measurements, images were captured using Zeiss Axio Imager M1 microscope and processed by AxioVision 4.7 (Zeiss). Percentage of cells with BLM fibers were scored and plotted. Line profiles across the fibers in the cells were analysed using ImageJ as described before [44].
Two independent nicking CRISPR/Cas9 guide pairs were designed using the CRISPR design tool at crispr.mit.edu. Both pairs targeted the coding sequence of exon 2. The following target sites for nicking pair #1, Guide A minus (5'—TCCCACATACTTTCATGGATGGG– 3'), Guide B plus (5'—TGGAGGTAGAAGATTTACACAGG—3') and #4 Guide A minus (5'—ATCTTCACAGCCTGCAGGCAGGG—3'), Guide B plus (5'—TCCCATCCATGAAAGTATGTGGG– 3'). Annealed oligonucleotides were cloned into the pSpCas9n(BB)-2A-GFP (PX461) vector (Addgene plasmid ID: 48140) [35]. HCT-116 cells were transfected in 6-well trays with Lipofectamine 3000 (Thermo Fisher Scientific) using the supplier's protocol. Two days after transfection, GFP-positive single cells were flow sorted into 96-well trays. Genomic DNA from clones was extracted using standard methods followed by PCR amplification screening across the CRISPR target site using the following oligonucleotides; RM-mf (5'—GATGGTGATGGGAGTGGTTC—3') RM-mr (5'–TCCTACATCCGGACTCCTTG—3'). PCR products were cloned into pGEM-T Easy (Promega) and Sanger sequenced at the Australian Genome Research Facility to confirm the presence of a knockout mutation. Three clones with knockout alleles at the DNA and protein levels were chosen for functional characterisation.
DNA content analysis was performed as previously described [43] and analysed using FACSCalibur and Cell Quest (Becton Dickinson).
Box plots were generated using beeswarm R package (https://cran.r-project.org/web/packages/beeswarm/index.html). Histograms were generated using Hmisc package (http://cran.r-project.org/web/packages/Hmisc/index.html). Statistical analyses were conducted using Student’s t test (unpaired).
|
10.1371/journal.pbio.1001775 | Faster Speciation and Reduced Extinction in the Tropics Contribute to the Mammalian Latitudinal Diversity Gradient | The increase in species richness from the poles to the tropics, referred to as the latitudinal diversity gradient, is one of the most ubiquitous biodiversity patterns in the natural world. Although understanding how rates of speciation and extinction vary with latitude is central to explaining this pattern, such analyses have been impeded by the difficulty of estimating diversification rates associated with specific geographic locations. Here, we use a powerful phylogenetic approach and a nearly complete phylogeny of mammals to estimate speciation, extinction, and dispersal rates associated with the tropical and temperate biomes. Overall, speciation rates are higher, and extinction rates lower, in the tropics than in temperate regions. The diversity of the eight most species-rich mammalian orders (covering 92% of all mammals) peaks in the tropics, except that of the Lagomorpha (hares, rabbits, and pikas) reaching a maxima in northern-temperate regions. Latitudinal patterns in diversification rates are strikingly consistent with these diversity patterns, with peaks in species richness associated with low extinction rates (Primates and Lagomorpha), high speciation rates (Diprotodontia, Artiodactyla, and Soricomorpha), or both (Chiroptera and Rodentia). Rates of range expansion were typically higher from the tropics to the temperate regions than in the other direction, supporting the “out of the tropics” hypothesis whereby species originate in the tropics and disperse into higher latitudes. Overall, these results suggest that differences in diversification rates have played a major role in shaping the modern latitudinal diversity gradient in mammals, and illustrate the usefulness of recently developed phylogenetic approaches for understanding this famous yet mysterious pattern.
| Why are there more species in the tropics? This question has fascinated ecologists and evolutionary biologists for decades, generating hundreds of hypotheses, yet basic questions remain: Are rates of speciation higher in the tropics? Are rates of extinction higher in temperate regions? Do the tropics act as a source of diversity for temperate regions? We estimated rates of speciation, extinction, and range expansion associated with mammals living in tropical and temperate regions, using an almost complete mammalian phylogeny. Contrary to what has been suggested before for this class of vertebrates, we found that diversification rates are strikingly consistent with diversity patterns, with latitudinal peaks in species richness being associated with high speciation rates, low extinction rates, or both, depending on the mammalian order (rodents, bats, primates, etc.). We also found evidence for an asymmetry in range expansion, with more expansion “out of” than “into” the tropics. Taken together, these results suggest that tropical regions are not only a reservoir of biodiversity, but also the main place where biodiversity is generated.
| The global increase of species richness toward the equator has been the subject of wonder, debates, and speculations since Darwin's times [1],[2]. Why do nearly all groups, spanning from amphibians [3], birds [4],[5], insects [6], mammals [7], and marine invertebrates [8] to micro-organisms [9], have more species in the tropics? Although more than 100 hypotheses have been proposed to explain this latitudinal diversity gradient [10],[11], the number of species in a given clade and region is ultimately explained by four major components: the time since the clade colonized the region, speciation rates, extinction rates, and dispersal events [12]. Hence, three main factors could in principle contribute to the observed high species richness in the tropics: the tropical origin of many clades, higher tropical net diversification rates (speciation minus extinction), and high dispersal rates from temperate regions to the tropics [8],[13].
Two main hypotheses related to dispersal dominate the literature. In the first, known as the “out of the tropics” hypothesis, lineages originate in the tropics, where they massively diversify, and then disperse from the tropics to the temperate regions. Under this hypothesis, dispersal is higher out of than into the tropics, thus acting “against” the latitudinal diversity gradient. In the second hypothesis, known as the “tropical niche conservatism” hypothesis, lineages originate in the tropics and have difficulties to disperse and adapt into temperate regions, thus accumulating in tropical regions [14],[15]. Under both hypotheses, the origin of diversity is tropical, such that intense dispersal from temperate to tropical regions is not considered a plausible explanation for high tropical species richness.
Dispersal effects aside, two major factors remain: time and diversification rates. The relative contribution of these two factors in explaining high tropical species richness remains highly debated [2],[16]. Some hypotheses emphasize diversification rates as the main driving force underlying the latitudinal diversity gradient: the “tropics as cradle” hypothesis emphasizes the role of high tropical speciation rates, whereas the “tropics as museum” hypothesis emphasizes the role of low tropical extinction rates [17]–[20]. Other hypotheses instead emphasize the role of time and historical contingencies [21]. Earth was mostly tropical before temperate regions started to expand ∼30–40 million years (Myr) ago, such that many groups likely have a tropical origin [16], and thus had more time to diversify in the tropics [17],[18].
Several studies, including two recent global-scale phylogenetic analyses of mammals [22] and birds [23], did not detect any correlation between latitude and diversification rates, supporting the view that the latitudinal gradient in species richness is unlinked to differences in diversification rates (e.g., [3],[12]). These findings, however, remain highly debated [24]. For example, Weir and Schluter [25] found a striking effect of latitude on speciation and extinction rates over the last ∼10 Myr in mammals and birds, with an unexpected increase in speciation rates with latitude. A latitudinal gradient in diversification rates has been suggested by several phylogenetic studies of diverse taxa [4],[6],[26],[27], as well as paleontological studies [8],[28],[29], and has given rise to many hypotheses of why speciation and extinction rates may vary with latitude [2]. It has been suggested that speciation is enhanced in the tropics by higher seasonal and longer term climatic stability [30], area effects [31], increased strength of biotic interactions [32],[33], and higher energy [34]. On the other hand, climatic variations and in particular glaciation cycles may be responsible for large-scale extinction events (and potentially enhanced speciation, [25]) in temperate regions [25],[30],[32].
Hence, the relative role of biogeographical history, speciation, and extinction in the latitudinal gradient remains unclear, representing a major scientific challenge for evolutionary biologists [2]. Here, we test the effect of latitude on speciation, extinction, and dispersal rates in the charismatic, species-rich, and globally distributed group of mammals, which displays a striking latitudinal diversity gradient ([7], Figure 1). Studies of the latitudinal gradient in mammals have mainly focused on environmental correlates of species richness [1],[31], and there is currently no consensus as to whether and how diversification rates vary with latitude in this group [2]. Weir and Schluter [25] suggested that both speciation and extinction rates increase with latitude, whereas Soria-Carrasco and Castresana [22] did not find any effect of latitude on speciation, extinction, or net diversification rates. These studies relied on sister taxa [25] or genus-level [22] analyses, thus focusing on recent diversification rates.
Here, we used recently developed biogeographic approaches (GeoSSE, [35]) that allow estimating speciation and extinction rates associated with specific biomes [36],[37]. Similar but nonbiogeographic models [38],[39] have been successfully used to detect various traits affecting diversification rates (e.g., [40]–[42]). Using these recent approaches allowed us to analyze a nearly complete phylogeny of 5,020 mammalian species covering the ∼170 Myr of their evolutionary history [43]–[45].
According to current range distribution data from the PanTHERIA database, mammalian species richness peaks near the equator (Figure 1), with 52% of all extant species living in the tropics, whereas only 25% live in temperate regions and 23% span both biomes [46]. The diversity of the eight most species-rich (>75 species) mammalian orders peaks near the equator, except that of the Lagomorpha, which is highest in Northern-temperate regions (Figure 2). We categorized each species reported in the mammalian phylogeny [43]–[45] (Materials and Methods) as living in the tropical biome, the temperate biome, or both. We analyzed the resulting worldwide data using recent biogeographic birth-death models of diversification ([35], Materials and Methods). In these models, a species present in one of the two biomes may give rise to two daughter species in this biome (rate λ), go extinct (rate μ), or disperse and expand its range in the other biome (rate d). These rates of speciation, extinction, and rate expansion may depend (or not) on the species' biome. A species occurring in both biomes (widespread species) may diversify and give rise to either one endemic plus one widespread daughter species (rate λ) or to two endemic daughter species, one in each biome (here referred to as speciation by biome divergence, rate λTempTrop). Speciation by biome divergence can occur if populations belonging to each biome experience directional selection in opposite directions leading to speciation. Widespread species may also contract their range by going extinct in one of the two biomes (rate μ).
We considered 16 alternative diversification scenarios, eight of which included speciation by biome divergence and eight of which did not (i.e., λTempTrop = 0). Within each of the set of eight scenarios, four had different rates of range expansion from the temperate regions to the tropics than the other way around (dTemp≠dTrop), and four had equal rates (dTemp = dTrop). These four scenarios consisted of (i) a scenario with equal tropical and temperate diversification rates (λTemp = λTrop and μTemp = μTrop), (ii) a scenario with speciation rates differing between biomes but equal extinction rates (λTemp≠λTrop and μTemp = μTrop), (iii) a scenario with extinction rates differing between biomes but equal speciation rates (λTemp = λTrop and μTemp≠μTrop), and (iv) a scenario with both speciation and extinction rates differing between biomes (λTemp≠λTrop and μTemp≠μTrop).
We fitted the 16 models to the phylogenetic tree of mammals, accounting for incomplete taxon sampling (Materials and Methods). The best fitting model was the model including speciation by biome divergence and with speciation, extinction, and dispersal rates differing between biomes (Table S1). Estimated speciation rates were higher—and extinction rates lower—in the tropics than in temperate regions (Figure 1, Table S1, and Figure S1), suggesting that the tropics act both as a cradle and as a museum of biodiversity. In addition, estimated rates of range expansion were higher from the tropics to temperate regions than the other way around (Figure 1 and Table S1), supporting Jablonski's “out of the tropics” hypothesis [8],[47]. Analyses on 100 trees randomly sampled from a Bayesian pseudoposterior distribution of trees [44],[45] confirmed these results (Materials and Methods). For all 100 trees, the best-fit diversification model included speciation by biome divergence and suggested the same trends in speciation, extinction, and dispersal rates, with higher speciation rates, lower extinction rates, and higher rates of range expansion in the tropics (Table S2). Estimated rates were consistent with the literature [42],[48]. These results were also robust to an alternative dating of the mammalian phylogeny obtained by incorporating dates from Meredith et al.'s study ([49], Materials and Methods, Table S3).
In our analyses of the global mammalian phylogeny, we made two major simplifying assumptions: that all lineages within a particular biome diversify at the same rate, and that diversification rates remain constant through clades' history. These two assumptions are likely violated in nature: first, diversification rates vary across lineages from a same biome for many reasons, including differences in diets [42], body size [44], or habitats [35],[50]; second, diversification rates typically vary through time [51]. To account for these two sources of rate variation, we carried a series of additional analyses at finer taxonomic resolution (i.e., on smaller phylogenies) and with more complex, time-variable models. For such analyzes, we constrained range expansion to be equally frequent from the tropics to the temperate regions than the other way around. We used this constraint to reach a reasonable trade-off between phylogeny size, model complexity, and statistical power (Materials and Methods). If dispersal rates are higher from the tropics to the temperate regions than the other way around, as estimated from the global phylogeny, constraining these rates to be equal should weaken the biome effect on diversification rather than generating a spurious effect. Trends obtained from dispersal-constrained fits to the whole phylogeny of mammals indeed tended to minimize the effect of latitude on diversification (Figure S2 and Table S4). Thus, further analyses were performed using the eight out of 16 models above with equal rates of range expansion. Comparison of uncertainties around parameter estimates for constrained versus unconstrained models on the global phylogeny suggested that constraining dispersal did not artificially reduce the uncertainty around other parameter estimates (Figures 1 and S2).
For trees corresponding to the eight richest mammalian orders, the best-fit diversification model varied across groups and trees representing these groups (Table S4). Accounting for speciation by biome divergence improved the fit of the models for the three richest groups (Rodentia, Chiroptera, and Soricomorpha) but not the others, and the estimated rates of speciation by biome divergence were in general lower than within-biome speciation rates. The estimated diversification rates for the richest orders were consistent with estimates obtained from the global phylogeny and in the literature (Figure 2, [42],[48]). There were differences across groups, yet the inferred net diversification rates were consistently higher in the tropics than in temperate regions (Figure 2, Figure S1, and Table S4). The two exceptions concerned the Lagomorpha, for which net diversification rate—following the diversity trend—was higher in temperate regions (Figure 2), and Carnivora, for which tropical and temperate net diversification rates were very similar. The inferred net diversification rate was in general positive except in the temperate regions in Chiroptera and Primates, and in the tropics in Lagomorpha.
Explanations for differences in net diversification rates between biomes differed across orders (Figure 2, Table S4, and Table S5). Higher tropical net diversification rates for Artiodactyla, Diprotodontia, and Soricomorpha were linked to higher speciation rates. In contrast, higher net diversification rates in the tropics for Primates, and in temperate regions for Lagomorpha, were linked to lower extinction rates within their corresponding biomes. Finally, higher tropical net diversification rates resulted from a combined effect of speciation and extinction in Chiroptera and Rodentia. In Carnivora, both speciation and extinction rates were higher in temperate regions, leading to a high species turnover at high latitudes. We tested the robustness of our results to potential biases in the phylogenetic tree we used. We ran our analyses on recent well-sampled phylogenies corresponding to three of the main orders (Rodentia [52], Primates [53], and Carnivora [54]) and the species-rich family Dasyuridae within Diprotodontia ([55], Materials and Methods). Even though slightly different models were selected depending on the phylogeny, trends in speciation and extinction rates were highly consistent, suggesting that these trends are strong enough to hold against phylogenetic and dating uncertainties (Table S6). The net diversification trends held in dispersal-constrained analyses as long as range expansion was constrained to not be much more frequent from temperate to tropical regions than in the other direction (Materials and Methods, Table S4). These trends also held when we completely relaxed dispersal, except in Chiroptera, Carnivora, and Lagomorpha, for which dispersal was inferred to contribute significantly to the latitudinal species richness patterns, such that unconstrained models supported different trends in diversification rates than constrained models (Figure S3).
We further refined the taxonomic scale of our analyses by considering all (seven) families with more than 100 species (Table S7). The diversification patterns for families within a given order were generally consistent with the diversification pattern corresponding to that order (Tables S4 and S5): the higher speciation and lower extinction rates in the tropics observed in Chiroptera were also found in its main family Vespertilionidae, the higher tropical speciation and extinction rates observed in Soricomorpha were found in its main family Soricidae, and the lower tropical extinction rate observed in Primates was found in its main family, Cercopithecidae. Higher tropical speciation rates were found in Bovidae, which is the main family of Artiodactyla. In Rodentia, the largest family (Muridae) had higher tropical speciation and extinction rates, consistently with the order. The two other families, however, showed divergent patterns: in Cricetidae the inferred extinction rate was higher in the tropics, and in Sciuridae the inferred speciation rate was higher in temperate regions. Still, net diversification rates were higher in the tropics than in temperate regions for all orders and families, with a single exception for Sciuridae where temperate and tropical net diversification rates were very similar.
Time variation in diversification rates can potentially bias rate estimates [51]. In particular, if speciation rates increased over time in temperate regions—for example, in response to recent glaciations cycles—this could have lead to an accumulation of splitting events towards the tips of the phylogeny. Such increase of splitting events in the recent past would resemble the “pull of the present” effect resulting from extinctions in constant rate models [56] and could thus be spuriously interpreted as a high temperate extinction rate. In order to test the robustness of our results to a potential variation of speciation rates through time, we implemented time variation in the GeoSSE biogeographic model (Materials and Methods). For the global mammalian phylogeny, a model with a linear time dependence of temperate and tropical speciation rates was indeed supported in comparison with the time-constant model (ΔAIC = 93, Table S8). The model suggested an increase in speciation rates through time in both biomes, but this increase did not affect our main findings: the inferred extinction rate remained higher in temperate regions, and speciation rates remained higher in the tropics over the majority of the history of mammals (Figure 3). Estimates obtained for speciation and extinction rates at present were very similar to estimates obtained with the time-constant model (Figure S2, Table S4, and Table S5). The time-variable model was not supported for two of the eight richest orders (Carnivora and Diprotodontia; see AICs in Table S8), suggesting that the hypothesis of time constancy may be relevant at this scale for these orders or that fitting other types of time dependencies would be required to reflect the nonlinear variation of environmental and biogeographic factors that have affected the diversification of mammals across the Cenozoic [48]. The time-variable model revealed trends in time variation (i.e., increase or decline of the speciation rate through time) that varied across orders (Table S8). Despite these variations, the estimated extinction rates remained higher in temperate regions (except in Soricomorpha and Lagomorpha, where they remained lower), and the speciation rates at present remained higher in tropical regions (except in Carnivora, where they remained lower).
The processes underlying the latitudinal diversity gradient are poorly understood. In particular, whether speciation and extinction rates also follow a latitudinal gradient is controversial [16],[22],[23],[25],[26]. The results obtained here for virtually all extant mammal species suggest that both speciation and extinction rates vary strikingly with latitude, resulting in significant differences of net diversification rates between the temperate and tropical biomes. Overall, inferred diversification patterns were consistent with the diversity gradient of each group, with higher speciation rates, lower extinction rates, or a combination of both where diversity is highest. These results were robust to potential variations of diversification rates through time and suggest that differential diversification rates may be largely responsible for today's mammalian diversity patterns.
Our results suggest that speciation rates in mammals are higher in the tropics. This was supported by analyses of the global phylogeny, and in dispersal-constrained analyses for five of the eight orders. This result was also robust to potential variations of diversification rates through time. For Primates and Lagomorpha, no latitudinal difference in speciation was detected, and for Carnivora, a higher speciation rate was found in temperate regions. These results suggest, as proposed by Soria-Carrasco and Castresana [22], that the absence of latitudinal effect on speciation they observed may arise from performing analyses at the genus level, impeding the observation of latitudinal effects that may have occurred in the early history of mammals. Using a phylogeny that covers the entire history of mammalian diversification, we found patterns in line with the hypothesis that speciation rates are higher in the tropics, potentially arising from area effects, increased specialization linked to climatic stability, niche availability, biotic interactions, and higher solar energy [2]. There is a possibility that the latitudinal diversity gradient was shaped during geological times as early as the Late Cretaceous or Paleogene [29] and that differences in diversification rates responsible for the construction of this pattern were only detectable with an approach covering the entire history of mammal diversification [22]. The relevance of this explanation for differences between our results and previous studies is, however, not entirely clear, because a lot of mammalian diversity likely arose in the last ∼20 Myr [57]. In addition, our time-dependent analyses suggest diversification differences between biomes were actually lower early in the history of mammals. Another possibility is that the phylogenetic method we used has more statistical power. In particular, it avoids averaging latitudes across species within genera, which likely weakens the association between latitude and diversification rates.
We found lower extinction rates in tropical regions for the global phylogeny and in dispersal-constrained analyses for the majority of the studied orders (Figures 1–2). A higher extinction rate in temperate regions was already reported for mammals [25] and with fossil records for other taxa [8],[28], potentially arising from higher climatic instability and glaciation cycles in temperate regions [8],[25],[30]. Our estimates of extinction rates in temperate regions were especially high for Primates, most probably due to their distinct preference for tropical forests since they originated in Asia 63–71 Myr ago [53]. Higher extinction rates in tropical regions were only found in Soricomorpha and Lagomorpha. In Lagomorpha, this pattern could come from their hypothesized temperate origin in Mongolia [58], their particularly good adaptation to grasslands (which appeared in the tropics only recently [59]), and strong negative biotic interactions in the tropics (e.g., competition and predation). Similar explanations could explain high extinction rates in Soricomorpha, with some families, such as the Talpidae, originating in temperate regions [60]. The lack of difference between temperate and tropical extinction rates in Diprotodontia may ensue from temperate marsupials having been restricted to low latitudes in the Southern Hemisphere, where they experienced few glaciations, from the extinction of tropical species following the aridification of Australia over the last ∼30 Myr [61], or alternatively from the poor statistical power linked to the paucity of temperate species in this group.
A positive correlation between speciation and extinction rates was found in Carnivora in dispersal-constrained analyses, with higher rates at high rather than low latitudes. This pattern of high turnover was proposed by Weir and Schluter [25] as a general pattern of diversification for recently diverged sister species of birds and mammals, but was observed only for Carnivora in our study. For the other groups, the difference between our results and those of Weir and Schluter [25] may come from differences in spatial scales: we analyzed the global mammalian phylogeny including species from the whole world, whereas Weir and Schluter [25] focused on the New World. In Carnivora, a temperate origin, as suggested in Felidae [62], associated with climatic oscillations and glaciations in temperate regions, may have promoted diversification, either by forcing species to move south or by restricting them to refugia separated by unsuitable habitats [63].
Our results with unconstrained dispersal mostly support Jablonski's “out of the tropics” hypothesis whereby species originate in the tropics and expand their range into temperate regions [8],[47]. This hypothesis was supported by analyses of the global phylogeny and in Rodentia, Soricomorpha, and Artiodactyla. The “out of the tropics” hypothesis has previously found paleontological [8] and phylogenetic [36] support in marine invertebrates, but has to our knowledge not been evidenced for mammals before. In Primates, Chiroptera, Carnivora, and Diprotodontia, range expansion was on the contrary estimated to be more frequent towards the tropics. Accounting for this asymmetry in range expansions did not change diversification trends in Primates and Diprotodontia (although the trend was weakened in the later). In Chiroptera and Carnivora, models with unconstrained dispersal suggested opposite diversification trends, with higher net diversification rates in temperate regions. High range expansion from temperate to tropical regions is not unrealistic. For example, the completion of the isthmus of the Panama 10 to 3 Myr ago resulted in an invasion of the tropics by temperate placental mammals which out-competed tropical marsupials [64]. However, range expansion itself tends to homogenize diversity across latitudes rather than to create a diversity gradient. Thus, high temperate net diversification rates combined with high range expansion from the temperate to the tropical regions cannot explain why there are more species in the tropics in these groups. This suggests that other factors, such as the fact that the tropics are older, are involved. This “time for speciation” hypothesis could be tested in more detail in these groups. More generally, these unconstrained results should be taken with care and tested further given the complexity of the models used relative to the size of the phylogenies [65].
Diversification estimates are only as reliable as the phylogenetic trees used to derive these estimates. Our main analyses were performed on the mammal tree of Bininda-Emonds et al. [43], improved by Fritz et al. [44], and resolved by the polytomy resolver of Kuhn et al. [45]. This tree is the most up-to-date nearly complete species-level tree for mammals, but it has weaknesses in terms of both topology and branch lengths. There is a possibility that the polytomy resolver biased the analyses; however, given that the resolutions are based on a birth-death process with equal rates across the phylogeny, our results of a strong biome effect are conservative. The robustness of the patterns across trees from the posterior distribution is also reassuring in terms of sensitivity to the random resolutions. In addition, diversification patterns in the least resolved groups, Chiroptera and Rodentia, were consistent with patterns obtained in the other groups, which are better resolved. Finally, the consistency of the diversification trends in our analyses of recent, well-sampled phylogenies gives confidence in our results. Another source of uncertainty is the dating of phylogenetic nodes. The family-level tree of Meredith et al. [49] suggests a more ancient origin of mammals and a more recent origin of extant orders (delayed by ∼11 Myr) than the tree of Bininda-Emonds et al. [38], while another family-level tree by dos Reis et al. [66] indicates similar dating results. In agreement with previous results from the literature [49], we found that diversification analyses were consistent between the two alternative dating.
Overall, we advocate that diversification rates play a crucial role in driving differences in species richness between the temperate and tropical biomes. We support the hypothesis that higher mammalian species richness in the tropics results from faster speciation and reduced extinction. This challenges previous suggestions that speciation in mammals is faster in temperate regions [25] or that the latitudinal gradient is linked to factors unrelated to diversification rates, such as the ancient origin of the tropics [2], tropical niche conservatism [16], or higher tropical carrying capacities [67]. Still, further analyses, such as historical biogeographic reconstructions [68] and diversification analyses accounting for nonlinear time dependence [48] and diversity dependence [69],[70], will be needed to fully understand the relative importance of each of these various factors. Mammals are one of the most charismatic and well-documented groups of living organisms, yet our vision of mammalian macroevolution continues to change drastically as new data are compiled and new methods are developed. Similar approaches applied to other vertebrates, insects, plants, and microorganisms will help us meet the challenge of evaluating the roles of history, speciation, and extinction in the origin of the latitudinal diversity gradient. This constitutes a necessary first step before we can fully understand the proximal ecological and evolutionary processes, correlated with latitude, which shaped current diversity patterns.
Our main analyses were performed on 100 species-level time-calibrated phylogenetic trees randomly sampled from the Bayesian pseudo-posterior distribution of trees provided by Kuhn et al. [45]. Kuhn et al. [45] used a birth–death model to resolve the polytomies in the supertree of Fritz et al. [44], which was built by completing and redating the phylogeny of Bininda-Emonds et al. [43]. This resulted in 100 phylogenies of 5,020 mammal species each. These phylogenies account for almost all mammals, as the total number of described species is 5,416 [71]. We further used this global-scale distribution of trees to obtain a distribution of trees for each of the eight most species-rich mammalian orders. We also combined these 100 trees to build a maximum clade credibility tree with the TreeAnnotator 1.7.5 (included in the BEAST package [72]). This maximum clade credibility tree, or “consensus” tree, was used for analyses on the whole phylogeny, on the orders and on the families. For convenience, we refer to all these trees as “Bininda-Emonds' trees”.
Dates from Bininda-Emonds' trees are debated [49],[66],[73]. To test the robustness of our results to these dates, we considered a tree incorporating the alternative dating proposed by Meredith et al. [49]. To redate Bininda-Emonds' consensus tree using dates from Meredith et al. [49], we used the list of nodes shared between the two studies given in table 1 from Meredith et al. [49]. Within this list, we selected the 16 deepest nodes, which correspond to ordinal and superordinal groups; these nodes were the ones that diverged the most between the two studies in terms of age estimate. We used these nodes as constraints in PATHd8 [74] to redate Bininda-Emonds' tree (Dataset S1).
Bininda-Emonds' trees offer the unique and considerable advantage to include almost all mammalian species. On the other hand, a significant number of species had no genetic data and were included in the phylogeny by a grafting and random resolution procedure. To test the robustness of our results to this procedure, we selected recent, better sampled phylogenies that did not artificially include species without genetic data. We found four publicly available phylogenies following these criteria that included species from both the tropical and the temperate regions: Rodentia (1155 spp., 50% sampled, [52]), Primates (367 spp., 98% sampled, [53]), Carnivora (286 spp., fully sampled, [54]), and Dasyuridae (Diprotodontia, 66 spp., 96% sampled, [55]).
We obtained minimum and maximum latitudinal data from the PanTHERIA database [46]. These data cover 4,668 species, including 4,536 of the species from the phylogeny. We chose −23.4° and 23.4° as the threshold latitudes defining the tropics and used the latitudinal data to characterize each species as living in the temperate biome, the tropical biome, or both (“widespread” species). We discarded all species in the phylogeny for which latitudinal data were not available, including all marine mammal species.
The GeoSSE model can account for missing species in phylogenies. Likelihoods corresponding to incomplete phylogenies are obtained by considering the probabilities that an extant species from each type (here tropical, temperate, and widespread) is sampled. In practice, these probabilities were computed as the fraction of species from each type present in the phylogeny, and directly introduced as a fixed parameter in the likelihood function. We estimated the fraction of species sampled in each biogeographic category (temperate, tropical, and widespread). We estimated the total number of species in each category by applying the fraction of temperate, tropical, and widespread species computed from the 4,668 species represented in PanTHERIA to the total number of described mammal species (5,416 species following [71]). In Bininda-Emonds' tree, for example, the estimated fraction of species represented in the phylogeny was 0.84 for temperate species, 0.83 for tropical species, and 0.85 for species spanning both biomes.
To test for an association between latitude and diversification rates, we used the Geographic State Speciation and Extinction model (GeoSSE, [35]), implemented in the diversitree R-package [75]. This birth–death model is a geographic extension of character-dependent diversification models [38],[39] including three parameters related to speciation (λTemp, λTrop, λTempTrop), two parameters related to extinction and range contraction (μTemp, μTrop), and two parameters related to range expansion (dTemp, dTrop). We used two types of analyses: analyses in which the seven parameters were allowed to vary freely, and analyses with constrained dispersal (i.e., dTemp = dTrop).
Analyses of the global phylogeny were performed on both the consensus tree and the posterior distribution of trees. We compared the 16 diversification models described in the text using the Akaike Information Criterion (AIC). We checked support for the selected model against all other models nested within it using the likelihood ratio test (p<0.05). We estimated the speciation, extinction, and range expansion rates corresponding to the best fitting model. We ran Bayesian Markov chain Monte Carlo (MCMC) analyses on the consensus tree, using exponential priors with parameters obtained from the character independent model, a 500-step burnin, and 20,000-step chain [75]. Convergence occurred within the few first steps and parameter estimates were very stable along the chain (Figure S4).
In our analyses of the eight most species-rich orders of mammals, the main families within these orders, and the four more recent phylogenies described above, we reduced the number of parameters in our model by constraining dispersal (i.e., fixing dTemp = dTrop). It is now well recognized that phylogenies have limited statistical power, particularly those of small size [76],[77]. Fitting complex models to such phylogenies is not recommended. In particular, a study specifically designed to test the robustness of character-dependent models—such as the GeoSSE model used here—recommended to simplify models by reducing their number of parameters when phylogenies are small [65]. Given our primary interest to analyze the effect of biomes on speciation and extinction rates, we constrained dispersal. We performed the same maximum likelihood and MCMC analyses as on the global phylogeny, using the eight models with constrained dispersal.
We tested the robustness of our results to the hypothesis that range expansion is symmetric (dTemp/dTrop = 1). First, we ran the best-fitting models while constraining the ratio dTemp/dTrop to other values, using the consensus tree used for MCMC analyses. We considered 150 values ranging from 0 to +∞, thus encompassing scenarios in which range expansion from the tropics to temperate regions is more frequent (dTemp/dTrop<1) and scenarios in which range expansion from temperate regions to the tropics is more frequent (dTemp/dTrop>1). For each value of dTemp/dTrop, we assessed whether the trend in net diversification rate was conserved. For example, if we found a higher net diversification rate in the tropics under the initial hypothesis of symmetrical range expansion (dTemp/dTrop = 1), we assessed if it remained higher in the tropics with the new ratio. This yielded a lowest and highest ratio dTemp/dTrop such that trends were conserved. Second, we ran unconstrained models, such as the ones fitted to the global mammalian phylogeny (described above).
To test the robustness of our results to time variation in speciation rates, we relaxed the hypothesis of rate constancy in the GeoSSE model. We modified the make.geosse function from the diversitree package [70], which computes likelihood functions associated with the different biogeographic models, by integrating the implementation of time dependency available in the Binary State Speciation and Extinction (BiSSE) model (codes are available in diversitree, make.geosse.t function). Speciation rates were assumed to vary linearly through time, such that λ(t) = λ0+rt, where λ0 is the speciation rate at present and r controls the rate of change in speciation rate through time, and t measures time from the present to the past. Extinction rates and dispersal are assumed constant through time μ(t) = μ, d(t) = d. We included time variation in speciation rates in the best-fit time-constant model (i.e., the model reported in Tables S4 and S5), with dTemp = dTrop. These analyses were performed on the consensus “Bininda-Emonds” tree for the global phylogeny and each order.
|
10.1371/journal.pntd.0004386 | Concerted Efforts to Control or Eliminate Neglected Tropical Diseases: How Much Health Will Be Gained? | The London Declaration (2012) was formulated to support and focus the control and elimination of ten neglected tropical diseases (NTDs), with targets for 2020 as formulated by the WHO Roadmap. Five NTDs (lymphatic filariasis, onchocerciasis, schistosomiasis, soil-transmitted helminths and trachoma) are to be controlled by preventive chemotherapy (PCT), and four (Chagas’ disease, human African trypanosomiasis, leprosy and visceral leishmaniasis) by innovative and intensified disease management (IDM). Guinea worm, virtually eradicated, is not considered here. We aim to estimate the global health impact of meeting these targets in terms of averted morbidity, mortality, and disability adjusted life years (DALYs).
The Global Burden of Disease (GBD) 2010 study provides prevalence and burden estimates for all nine NTDs in 1990 and 2010, by country, age and sex, which were taken as the basis for our calculations. Estimates for other years were obtained by interpolating between 1990 (or the start-year of large-scale control efforts) and 2010, and further extrapolating until 2030, such that the 2020 targets were met. The NTD disease manifestations considered in the GBD study were analyzed as either reversible or irreversible. Health impacts were assessed by comparing the results of achieving the targets with the counterfactual, construed as the health burden had the 1990 (or 2010 if higher) situation continued unabated.
Our calculations show that meeting the targets will lead to about 600 million averted DALYs in the period 2011–2030, nearly equally distributed between PCT and IDM-NTDs, with the health gain amongst PCT-NTDs mostly (96%) due to averted disability and amongst IDM-NTDs largely (95%) from averted mortality. These health gains include about 150 million averted irreversible disease manifestations (e.g. blindness) and 5 million averted deaths. Control of soil-transmitted helminths accounts for one third of all averted DALYs. We conclude that the projected health impact of the London Declaration justifies the required efforts.
| Neglected tropical diseases (NTDs) are a group of infectious diseases that occur mostly in poor, warm countries. NTDs are caused by various bacteria and parasites, such as worms. They can either be cured or prevented through drugs and other interventions, such as control of insects that spread the infection. The London Declaration is a statement by various organizations, including the World Health Organization (WHO) and pharmaceutical companies that donate the necessary drugs. The declaration endorses targets for disease reductions by 2020, as recently formulated in the WHO Roadmap, to be achieved by rigorous application of available interventions. We explore how much health can be gained if these targets are indeed achieved. We estimate that in such case 5 million deaths can be averted before 2030 and also that huge reductions in ill-health and disability can be realized. Over the period 2011–2030, a total health gain would be accomplished of about 600 million disability adjusted life years (DALYs) averted. DALYs are a measure of disease burden, consisting of life years lost and years lived with disability. This enormous health gain seems to justify similar investments as for e.g. HIV or malaria control.
| Neglected tropical diseases (NTDs) are considered a special category of infectious diseases, distinct from the major killers HIV, tuberculosis and malaria, which have been the main focus of attention and funding for developing countries over the past decades. NTDs are largely confined to (sub)tropical resource-constrained regions, where they cause substantial morbidity, disability and even mortality, as documented by the recent Global Burden of Disease (GBD) estimates [1–4], and consequently have high socioeconomic impact [5,6]. Most NTDs are either curable or preventable, but in practice there exist major barriers to the effective implementation of control. Fortunately, international commitment to NTD control has rapidly increased in recent years. In 2012, the World Health Organization (WHO) formulated a ‘Roadmap’ towards ambitious control and elimination targets [7]. By endorsing the London Declaration on NTDs, several private and public sector organizations committed to meet those targets [8]. For five NTDs—lymphatic filariasis (LF), onchocerciasis, schistosomiasis, soil-transmitted helminths (STH) and trachoma—the primary control strategy is preventive chemotherapy (PCT). For four other NTDs—Chagas’ disease, human African trypanosomiasis (HAT), leprosy and visceral leishmaniasis (VL)–control programs rely on case detection with innovative and intensified disease management (IDM), sometimes in combination with other measures such as vector control. Guinea worm (dracunculiasis) is confined to just a few residual foci in Africa and close to being eradicated. For LF, trachoma, HAT and leprosy the target is elimination by 2020, and for the others it is currently control [7,9].
The London Declaration was formulated to accelerate progress towards the WHO Roadmap targets by sustaining or expanding existing drug donation initiatives; providing funding to support NTD programs, strengthen drug distribution, and research and development; and enhancing collaboration and coordination on NTDs at (inter)national levels [8]. To further motivate and justify these efforts it is important to know their expected health gains. We therefore aim to estimate the global health impact of meeting the WHO Roadmap targets in terms of averted morbidity and mortality, expressed in years lived with disability (YLD), years of life lost (YLL), and disability adjusted life years (DALYs). YLD reflects the number of prevalent cases of each considered disease manifestation multiplied by a disease-specific disability weight between 0 (perfect health) and 1 (equivalent to death), whereas YLL reflects the number of deaths times a standard life expectancy at the age of death in years. The number of DALYs is the sum of both measures (DALYs = YLD + YLL).
Two datasets were used in our calculations. First, the GBD-2010 estimates regarding NTDs were made available to us by the Institute for Health Metrics and Evaluation (IHME), Seattle, USA [3,10]. Second, UNPOP demographic data and projections were obtained from the website of United Nations Department of Economic and Social affairs [11]. The GBD-2010 data consist of three burden estimates: prevalent cases, years lived with disability (YLD) and years of life lost (YLL). These estimates were available for 1990 and 2010, per country, age group and sex. Prevalent cases were provided per disease manifestation (sequela), whereas YLD and YLL were only provided as totals per NTD. Table 1 gives an overview of all 31 sequelae considered in the GBD calculations for the London Declaration NTDs. Guinea worm was not included in the GBD study and is therefore not considered here. For STH, burden estimates were available for ascariasis, hookworm disease and trichuriasis separately. Background documents justifying and describing the underlying assumptions of the GBD estimates, including disability weights, were also kindly made available to us. GBD estimates were structured according to the following age groups: 0–6 days, 7–27 days, 28–364 days, 1–4 years, 5–9 years, …, 75–79 years, and 80+ years. We combined the four youngest age groups into a 0–4 years group. For irreversible sequelae (see below), the number of prevalent cases was redistributed into 1-year age groups, using a smoothing method that minimizes the squared differences between successive years, under the constraint that 5-years totals equal the available data. The demographic data were already available in 1-year age groups.
The GBD estimates of the number of prevalent cases for all 31 sequelae and 5 causes of death (HAT, VL, STH-ascariasis, Chagas’ disease and schistosomiasis) in 1990 and 2010 were taken as the basis for our calculations. Estimates for other years were obtained by interpolating between 1990 and 2010, and further extrapolated until 2030, under the assumption that the 2020 WHO Roadmap targets were met and sustained beyond 2020. Health impacts were assessed by comparing the results of achieving the targets with the counterfactual, construed as the health burden had the 1990 situation continued unabated. Prevalent cases (both remaining and counterfactual) were translated to YLD and YLL, and summed to arrive at DALYs. The health impact of reaching the targets was expressed as DALYs averted over the decades 2011–2020 and 2021–2030.
All calculations were carried out in duplicate in Microsoft Excel, and verified using R. All results (totals and country-specific values), underlying calculations and assumptions are available as an open-access web-based dissemination tool (https://erasmusmcmgz.shinyapps.io/dissemination/). A detailed step-wise explanation of our methodology is given below.
Sequelae were first categorized as either reversible or irreversible (Table 1), depending on whether treatment of the underlying infection would remove the sequelae in a relatively short time, say, within a couple of years at most. For all reversible sequelae, interventions were considered to affect their prevalence, while for irreversible sequelae this was their incidence. Linear interpolation (at the log-scale for irreversible sequelae) was carried out between 1990 (or the start-year of large-scale control efforts) and 2010 for prevalence rates (i.e. the number of prevalent cases divided by population size) per sequela, country, age group and sex. Absolute numbers were then calculated from these interpolated prevalence rates, using the demographic UNPOP data. For 2020 (and beyond), WHO Roadmap targets were interpreted in terms of prevalence (for reversible sequelae) or incidence (for irreversible sequelae) levels, based on discussions with—mostly WHO—disease experts (Table 2). Trends in incidence and prevalence during the intervening years (usually 2010–2020) were obtained through linear interpolation between the 2010 levels (GBD data) and the interpreted targets. We then translated the calculated trends into absolute numbers of remaining cases using UNPOP projections for the period 2011–2030, and compared this with the counterfactual situation of no additional control efforts, to assess the impact of meeting the targets. The counterfactual was construed as the health burden that would have been expected had the 1990 epidemiological situation (i.e. disease incidence or prevalence) continued unabated. Whenever the 2010 prevalence of a sequela exceeded that of 1990, we took 2010 as the counterfactual.
Interpolation for irreversible sequelae, such as blindness as a result of onchocerciasis, was carried out at the level of incidence, because even after elimination of infection these sequelae will persist until the death of the last patient. The annual incidence density λ(a,t) at age a and calendar time t of an irreversible condition is given by the following equation
∂s(a,t)∂a+∂s(a,t)∂t=−λ(a,t)⋅s(a,t)+[1−s(a,t)]⋅μ(a,t)⋅s(a,t)
where s(a,t) denotes the susceptible fraction (i.e. 1 –prevalence) of the population and μ(a,t) the excess mortality rate among those affected. In a stable endemic situation (i.e. without cohort effect, thus ∂s(a,t)/∂t = 0) and without excess mortality (i.e. μ(a,t) = 0), λ(a,t) can be obtained from a single cross-sectional survey by taking the differences in the logarithmic age profile of the fraction susceptible. However, because the cross-sectional age profiles of GBD 1990 and 2010 for each sequela differed, we annualized the differences (on a logarithmic scale) in these profiles to obtain an estimate for ∂s(a,t)/∂t. We further assumed the excess mortality rate to be independent of age and calendar time, and have a pre-set value μ* = 0.0, 0.05 or 0.10, dependent on the severity of the sequela (Table 1). The value of μ* was chosen after consultation of the disease experts and crudely reflected the mortality rates as used in the GBD calculations. The resulting incidences were calculated back to prevalences (of remaining cases) by ‘exposing’ cohorts to the derived age and time-specific incidence densities and excess mortality rates.
Predicted prevalent cases for each sequela were then translated to YLD, using two matrices of multiplication factors (one for the year 1990 and one for 2010) that we had derived from the GBD data as follows. Whenever an NTD had one sequela (e.g. trachoma), the GBD YLDs in 1990 and 2010 were divided by the number of prevalent cases in the same year to arrive at country, age and sex-specific multiplication factors that capture disability weights, the underlying case-mix (e.g. severe vs. mild disability, where applicable), and correction of burden estimates for co-morbidity, as used in the GBD 2010 study [2]. For NTDs with multiple sequelae (e.g. onchocerciasis) we followed the same procedure, but using a weight for each sequela based on an estimate of the average disability weight using GBD documentation (Table 1), because the YLD data provided by the GBD study did not separate the contributions of different sequelae. We treated all multiplication factors as constants. Remaining cases after 2010 were multiplied by the factors in the 2010-matrix, and for 1990–2010 an interpolation of the multiplication factors in both matrices was used. For counterfactual cases we used the multiplication factors in the 1990-matrix, or both matrices when 2010 was used as counterfactual (i.e. similar to the approach for remaining cases).
Regarding our mortality calculations, we first translated GBD YLLs in 1990 and 2010 to actual country, age, and sex cause-specific mortality rates, using the age and sex-specific residual life expectancies as applied in the GBD study [1]. For HAT, VL and ascariasis, where mortality is closely linked (in time) to infection prevalence, these rates were treated as prevalent cases (of reversible sequelae) as described above and back-calculated to YLLs. For Chagas’ disease and schistosomiasis, where mortality is closely linked to late sequelae, we followed a different procedure. Similar to the calculation of YLDs for NTDs with multiple sequelae, we related YLLs in 1990 and 2010 to prevalent cases of selected sequelae, using a weight representing their lethality. For schistosomiasis, mortality was related to hematemesis (weight = 50), ascites (1.0) and schistosomiasis infestation (0.01). For Chagas’ disease, these were heart failure (10) and chronic heart disease (1.0).
Using the above method, some irreversible sequelae—in particular for Chagas’ disease and LF—showed for some countries values of λ(a,t) < 0, due to unrealistic fast declines in the GBD prevalence estimates between 1990 and 2010. Here, we chose alternative prevalences, but still within the confidence limit (Cl) provided by the GBD study, as follows. We reduced the GBD 1990 ‘Mean’ prevalence to 0.25 ‘Mean’ + 0.75 ‘Lower Cl’, and we increased the GBD 2010 ‘Mean’ prevalence to 0.25 ‘Mean’ + 0.75 ‘Upper Cl’.
The GBD 2010 estimates for leprosy appeared to be mistakenly based on overall leprosy new case detection (incident cases) instead of prevalence of (irreversible) cases with leprosy grade 2 disability, on which the disability weights are based. We therefore performed a recalculation to arrive at grade 2 disability prevalences as follows. First, we took from the WHO-published global leprosy data for 2010 the proportion of newly detected cases with grade 2 disability, which was 6% [12]. Secondly, prevalence of leprosy cases with grade 2 disability in virtual birth cohorts was accrued at a rate determined by this incidence density, while assuming a steady-state until 1990 and a linear decreasing incidence to 2010. We further assumed that excess mortality due to leprosy is negligible (μ* = 0.0). These prevalence values constituted the ‘GBD data’ on which our calculations were based.
Fig 1 shows the global trends in remaining and averted DALYs, distinguished into YLD of reversible and irreversible sequelae and YLL. According to the original GBD 2010 data (dark-colored bars), the health burden of onchocerciasis, STH, Chagas’ disease, HAT and VL has clearly decreased over the period 1990 to 2010. For LF, schistosomiasis and leprosy, the absolute burden has increased, but not as fast as would be expected from the counterfactual. Thus, for these NTDs, the relative burden has decreased, when correcting for population growth. Only for trachoma (and in some countries for schistosomiasis), the GBD-estimated burden has increased faster than would be expected from the demographic trends over the period 1990–2010.
Meeting the 2020 targets will lead to a substantial health-impact for all NTDs (Fig 1). It is clearly visible that reversible sequelae (green) are disappearing faster than irreversible sequelae (brown). This makes the health impact of reaching the targets for LF, trachoma and leprosy over the first two decades somewhat less spectacular compared to that for the other NTDs, of which the burden is mainly caused by reversible sequelae or death. Another important factor determining the overall health impact is population growth and other demographic developments, as expressed by the counterfactual. NTDs that are prevalent in Asia (LF, STH, leprosy and VL) show a slower rise of the counterfactual compared to the NTDs mainly confined to Africa (onchocerciasis, trachoma and HAT) or South America (Chagas’ disease).
Overall, meeting the targets of London Declaration NTDs will avert about 600 million DALYs in the two decades after 2010, nearly equally distributed between PCT and IDM-NTDs, with the former mostly (96%) attributable to averted disability, whereas the latter largely (95%) results from averted premature death (Fig 2). These health gains include about 150 million averted irreversible disease manifestations, in particular chronic heart disease due to Chagas’ disease, bladder pathology due to schistosomiasis, and hydrocele and lymphedema due to LF (Table 3). In addition, approximately 5 million deaths are averted, mainly from VL and HAT, and to a lesser extent Chagas’ disease (Table 4).
Of the 600 million DALYs overall averted in the period 2011–2030, in the ideal situation of meeting the WHO Roadmap targets of London Declaration NTDs, about 30 million will be realized in the year 2020, increasing to 40 million in the year 2030. This is of the same order of magnitude as the current annual health burden of any of the ‘big three’ infectious diseases, HIV/AIDS, TB and malaria, which accounted for about 80, 50 and 80 million DALYs, respectively, in 2010 [3]. Clearly, for these three infections elimination is a more remote perspective than for the nine NTDs targeted by the London Declaration. Thus, the ongoing efforts to control the big three seem to justify similar investments in NTD control. In addition, it can be expected that for several of these NTDs control efforts will lead to a cessation of transmission over vast regions, after which further control can be discontinued and investments wound down adding to the value of this investment for future generations.
STH accounts for one-third (34%) of the averted DALYs, almost entirely due to avoided disability. This perhaps surprising finding can be easily explained by the wide-spread distribution of STH [13]. Importantly, approximately half (46%) of the averted STH-burden would be realized in China. This brings to the fore the sensitivity of our results to the choice of counterfactual. That is, our assumption that the situation of 1990 would continue unabated may be questioned for several countries, including China, which have experienced unprecedented economic and social development over the past decades [14]. For example, the health impact for STH would be about halved if the situation in 2010 were used as the counterfactual, as can roughly be concluded from Fig 1, but such a drastic correction would certainly not be reasonable for many endemic countries in Africa and Southern Asia. On the other hand, socioeconomic development may also have facilitated the spread of NTDs, in particular schistosomiasis, of which large outbreaks followed the construction of dams and irrigation schemes [15]. HAT perhaps follows more erratic patterns, reflecting e.g. civil unrest, war and also ecological circumstances [16], so that the year 1990 may not be representative of the actual counterfactual over 1990–2020. Trachoma and schistosomiasis showed large increases in GBD prevalence from 1990 to 2010, which may well reflect an underestimation of the 1990 burden. Consequently, this may have led to underestimating both the counterfactual and the health impact. Another potential source of underestimation of the health impact for some NTDs may be that the largest gains are achieved in the initial years of programs, followed by a slow down towards the target year, as it becomes harder to reach the more marginalized populations. Furthermore, by using a fixed excess mortality rate μ* for irreversible sequelae (where applicable) we may have somewhat overestimated health impacts for these sequelae as treatment is likely to improve over time. However, since the remaining cases get older at the same time, possibly experiencing a higher mortality, we may have introduced some underestimation as well. Clearly, by using a uniform methodology we have introduced (perhaps occasionally substantial) under or overestimation of NTD and country-specific results, but we are confident that the overall bias in our estimated health impact of reaching the targets will be small.
Almost half (44%) of the overall health impact is attributable to averted deaths, in particular from visceral leishmaniasis and HAT, and to a lesser extent Chagas’ disease, followed by schistosomiasis and STH (ascariasis). In our calculations, we followed the GBD accounting philosophy which assigns all DALYs (i.e. residual life expectancy at the age of death) resulting from a death to the year in which it occurred [1], whereas DALYs attributable to morbidity are accrued during the years that individuals suffer [2]. Moreover, remaining life expectancies were based on the demography of Japan, according to the fundamental concept that all people are entitled to the best life expectancy in the world, irrespective of e.g. country of residence and socioeconomic status. Clearly, other methodologies might have distributed health gains differently over time.
Our calculations depend strongly on the estimates made in the GBD study [1–3]. These estimates are notably uncertain for NTDs, given the paucity of data on their geographic spread and control. Most GBD 1990 and 2010 estimates for NTDs show very wide confidence intervals, often ± 50% the mean, but sometimes with an upper confidence limit up to 5 times the mean. As a consequence, our predictions (all based on GBD point estimates) are subject to at least a similar degree of uncertainty. Also, the GBD disability weights used are still under heavy debate, such as the relatively low value for blindness as compared to itching [17]. Furthermore, our calculations are confined to the 31 sequelae considered in the GBD study, and discussions continue about whether additional sequelae need to be considered. In particular, the choice not to include so-called subtle morbidities, such as impaired cognitive development due to STH and schistosomiasis, or poor mental health from stigma and discrimination due to the disfigurements caused by LF and leprosy, is considered an important omission by many [13,18–21]. Our results also depend upon the interpretation and formulation of the WHO Roadmap targets [7,9], which occasionally are ambiguous. Consulting disease experts at WHO has resulted in agreement about interpretations for most NTDs, even though sometimes the targets were considered too general or utopic.
In addition to the intrinsic value of averting human suffering and death, this health impact of reaching the targets will also give rise to major economic and societal improvements, such as increased productivity and avoided (often catastrophic) out of pocket payments for treatment and care, which can be assigned monetary values. In particular, the currently ignored subtle morbidities are likely responsible for major societal impacts.
We realize that the targets are ambitious, and may for instance be jeopardized by challenges in drug distribution, disease surveillance and health care access. Also, systematic non-compliance in mass-drug administration, population groups currently not eligible for treatment, and development of drug or insecticide resistance could be serious threats, as demonstrated in a recent collection of studies by the NTD Modelling Consortium focusing on the question whether we are on track to reaching the goals [22]. Furthermore, even if the targets are reached by 2020 it is essential that control and surveillance are continued to avoid rebounding effects, certainly for those NTDs where elimination of transmission cannot be expected.
In conclusion, NTDs together constitute a major health burden, comparable to any of the three major infectious diseases HIV/AIDS, TB, and malaria. Achieving internationally agreed targets of NTD control and elimination will bring about major gains in health and reductions in human suffering. Much of this will be achieved by avoiding morbidity rather than mortality as many of the parasites involved, such as soil transmitted helminths, rarely kill their hosts. This also implies that our impact assessment depends on the valuation of health states as used by GBD, a valuation that inevitably is somewhat subjective and open to debate. We did not consider the costs involved in reaching these targets, but a recent assessment demonstrated that these are relatively modest [23], indicating that the cost-effectiveness of interventions to control NTDs will likely be high. One thing is certain however: as NTDs are disorders that disproportionately affect the poor, their control will considerably improve global equity.
|
10.1371/journal.pntd.0002834 | Prevalence and Serological Diagnosis of Relapse in Paracoccidioidomycosis Patients | A review of 400 clinical records of paracoccidioidomycosis (PCM) patients, 93 with the acute/subacute (AF) and 307 with the chronic form (CF), attended from 1977 to 2011, selected as to the schedule of release for study by the Office of Medical Records at the University Hospital of the Faculdade de Medicina de Botucatu – São Paulo State University – UNESP, was performed to detect cases in relapse. The control of cure was performed by clinical and serological evaluation using the double agar gel immunodiffusion test (DID). In the diagnosis of relapse, DID, enzyme-linked immunosorbent assay (ELISA) and immunoblotting assay (IBgp70 and IBgp43) were evaluated. Out of 400 patients, 21 (5.2%) went through relapse, 18 of them were male and 3 were female, 6∶1 male/female ratio. Out of the 21 patients in relapse, 15 (4.8%) showed the CF, and 6 (6.4%) the AF (p>0.05). The sensitivity of DID and ELISA before treatment was the same (76.1%). DID presented higher sensitivity in pre-treatment (80%) than at relapse (45%; p = 0.017), while ELISA showed the same sensitivity (80% vs 65%; p = 0.125). The serological methods for identifying PCM patients in relapse showed low rates of sensitivity, from 12.5% in IBgp70 to 65.0% in IBgp43 identification and 68.8% in ELISA. The sensitivity of ELISA in diagnosing PCM relapse showed a strong tendency to be higher than DID (p = 0.06) and is equal to IBgp43 (p = 0.11). In sum, prevalence of relapse was not high in PCM patients whose treatment duration was based on immunological parameters. However, the used methods for serological diagnosis present low sensitivity. While more accurate serological methods are not available, we pay special attention to the mycological and histopathological diagnosis of PCM relapse. Hence, direct mycological, cytopathological, and histopathological examinations and isolation in culture for P. brasiliensis must be appropriately and routinely performed when the hypothesis of relapse is considered.
| Paracoccidioidomycosis (PCM) is a systemic mycosis caused by fungi of the Paracoccidioides brasiliensis and Paracoccidioides lutzii complexes, which live in the soil and affect mainly rural workers in the most productive period of their lives. PCM can relapse after effective treatment because quiescent fungi can reactivate. However, the physician has to differentiate a relapse from another disease. In the diagnosis of relapse, the classical double agar gel immunodiffusion test (DID) and the modern enzyme-linked immunosorbent assay (ELISA) and immunoblotting (IB) were evaluated. The frequency of PCM relapse and its diagnosis were the objective of this study. The prevalence of relapse was low and the new serological tests presented a little higher sensitivity than DID. Since the serological tests presented only a moderate sensitivity, direct mycological, cytopathological, and histopathological examinations and isolation in culture for the fungi must be appropriately and routinely performed when the hypothesis of relapse is considered.
| Paracoccidioidomycosis (PCM) is a systemic mycosis caused by thermo-dimorphic fungi from the Paracoccidioides brasiliensis complex and the Paracoccidioides lutzii complex [1]. Confined to Latin America, PCM is endemic to the area extending from Mexico to Argentina [2]. Although incomplete, the available data indicate a higher incidence of such mycosis in Brazil, where they are frequently diagnosed in the State of São Paulo [3].
PCM is known to be able to reactivate despite effective treatment because of quiescent fungi remain and disease relapse is possible. However, a few studies have investigated the relapse of paracoccidioidomycosis. A study that was conducted with 58 patients who were infected with paracoccidioidomycosis and treated with itraconazole indicated that there was a relapse in 8 (13.8%) of the cases, where 50.0% of them occurred after 36 months of discontinued treatment [4].
The gold standard for diagnosing PCM is either the direct visualization of characteristic multiple-budding cells in biological fluids and tissue sections, or fungus isolation from clinical specimens [5]. Serological tests are useful for diagnosis, severity assessment and follow-up, especially the double agar gel immunodiffusion test (DID) [6].
The ELISA (Enzyme-Linked Immunosorbent Assay) test has been the subject of a number of publications in regard to the detection of circulating antibodies PCM patients [7], although it has not been included in most clinical laboratories.
This paper is aimed at evaluating the ELISA test and its ability to detect antibodies against Paracoccidioides brasiliensis in PCM patients in relapse.
A prospective study was conducted with patients who were receiving medical attention at the Paracoccidioidomycosis Outpatient Service in the Infectious Diseases Center of the University Hospital in Botucatu Medical School – UNESP. The need to adhere to the treatment given, as well as, when indicated, the suppression of alcohol intake and smoking, was reiterated in all of the outpatient visits.
This study was approved by the Research Ethics Committee of FMB-UNESP. Written informed consent was obtained from all participants. In this study IRB was signed by all the adult patients and by one of the parents of the children. We had no IRB signed by the closest relative or the legal representative.
The criteria for the inclusion of the patients in the study were as follows: (A) Upon diagnosis: (a) PCM confirmed cases, characterized by the presence of a suggestive clinical condition and the identification of typical forms of P. brasiliensis yeast phase in one or more clinical materials; (b) PCM probable cases, characterized by the presence of suggestive clinical conditions and specific serum antibodies detected by the results of the DID test. (B) After treatment: patients showing PCM relapse, characterized by the recurrence of signs and symptoms indicative of PCM, with or without the identification of the typical forms of P. brasiliensis yeast phase in any clinical specimen, and/or the serology reaction to the DID test. Those in this group have also been given the appropriate therapy and have shown clinical cure, experienced a normalization of their erythrocyte sedimentation rate (ESR) and a serology regression to negative values, and have continued antifungal treatment for at least one year after serological cure.
The exclusion criteria were as follows: presence of other systemic diseases of infectious, inflammatory or neoplastic source, such as co-morbidity; pregnancy; lactation; history of hypersensitivity or severe side effects in response to azole or cotrimoxazole; concomitant use of medicines that can interact with such antifungals or change their serum levels or concomitant use of other antifungals.
Eight patients with mycologically confirmed and PCM who did not relapse constituted the control group; 122 serum samples from these patients were serologically evaluated by the double agar gel immunodiffusion test (DID) and the enzyme-linked immunosorbent assay (ELISA).
The categorization of patients and the assessment of disease severity in each patient were carried out according to Mendes [8] and the Paracoccidioidomycosis Surveillance and Control Guideline [6] by the infectious disease MD responsible for attending the patients.
A review of 400 clinical records of PCM patients, 93 of whom with the acute/subacute (AF) and 307 with the chronic form (CF), attended from 1977 to 2011 selected as to the order of release for study by the Office of Medical Records in the University Hospital of the Faculdade de Medicina de Botucatu – São Paulo State University – UNESP, was performed to detected cases patients in relapse. A standard form was filled in per patient containing their name, sex, age, enrollment code at the Hospital, date of admission to the service, presence of previous treatment, clinical form, treatment start date, and results of diagnostic and serological tests.
Likewise, a review of the serological records of PCM patients was conducted by the Tropical Diseases & Mycology Research Laboratory – Department of Infectious and Parasitic Diseases, in Botucatu Medical School – UNESP.
Treatment was deemed appropriate if the signs and symptoms of the disease were gone, there was a normalization of the erythrocyte sedimentation rate (ESR) and a negative reaction to DID could be observed during one year of antifungal medicine administration and for at least one year without such therapy.
The categorization of cases of relapse was characterized by the reappearance of signs and symptoms that were compatible with PCM, with or without the identification of the typical P. brasiliensis yeast forms in any clinical specimen, or the reaction to DID after the appropriate treatment in patients with confirmed or probable PCM. Five types of relapse have been taken into account: (a) clinical, serological and mycological, characterized by the reappearance of signs and symptoms compatible with PCM, with or without identification of the typical P. brasiliensis yeast forms in any clinical specimen, and a positive DID serology test; (b) clinical and serological, characterized by the reappearance of signs and symptoms compatible with PCM and a positive DID serology test; (c) clinical and mycological, characterized by the reappearance of signs and symptoms compatible with PCM, with the identification of the typical P. brasiliensis yeast s in their common forms in any clinical specimen; (d) clinical, characterized by the reappearance of the clinical signs of PCM, responsiveness to sulfamethoxazole-trimethoprim combination (TMP-SMX); and (e) serological, characterized only by a positive DID response without clinical signs.
The P. brasiliensis 113 (Pb-113) yeast phase culture filtrate, prepared at the Clinical Mycology Laboratory of Araraquara Pharmaceutical Sciences College – UNESP, was used to run DID, ELISA and immunoblotting (IB) tests.
The serum levels of anti-Pb antibodies were determined through a DID test which was run according to the specifications of Restrepo [9] at the Tropical Diseases Experimental Laboratory in Botucatu Medical School – UNESP.
Non-diluted sera were tested first and then diluted by ½, with serial dilutions (×2).
For each test, positive and negative control sera were included.
The serum levels of anti-Pb antibodies were determined through an ELISA test [10]–[11], which was run at the Tropical Diseases Experimental Laboratory in Botucatu Medical School – UNESP, according to the standard protocol set forth by the Mycosis Immunodiagnosis Laboratory of Adolfo Lutz Institute (São Paulo, Brazil).
The cutoff point was determined by drawing the ROC (receiver operator characteristic) curve for 200 PCM patients and 100 healthy subjects (control subjects) from Botucatu Blood Bank. The final end point was statistically defined according to the specifications made by Frey et al. [12] for a 95% confidence interval, and was equal to 0.710 optical density.
Such serological techniques were conducted at the Mycosis Immunodiagnosis Laboratory of Adolfo Lutz Institute, in São Paulo, Brazil.
Electrophoresis that involved the transfer of proteins contained in polyacrylamide gels to nitrocellulose membranes was carried out according to a methodology described by Towbin et al. [13], as amended by Passos [14].
A 2×2 comparison of the sensitivity of the serological methods that were used was carried out through a binomial test, according to the specifications made by Siegel [15].
A comparison of the curves representing the time to achieve serological response in different groups of patients and serological techniques was conducted by using the ranks for the two methods, and regression was calculated for each of them, which were then compared to both regressions, through F-test [16].
A kappa coefficient was ascertained by means of SAS – statistical analysis system, Version 6.12, SAS Institute Inc., USA. The kappa coefficient was interpreted as follows: (a) mild, when lower than 0.4; (b) moderate, from 0.4 to 0.79; (c) strong, from 0.8 to 0.99; and (d) perfect, when equal to 1.0.
For each such statistic test, differences were set up at p≤0.05.
Out of 400 patients, 21 (5.2%) went through relapse, 3 of which (14.3%) were female and 18 (85.7%) were male, resulting in a male/female ratio of 6∶1. The prevalence of male patients was 85.7% among the patients in relapse, 88.1% among the patients who did not relapse and 88.0% among the 400 studied patients, with no difference between patients in relapse or not (p = 0.72).
Out of the 21 patients who experienced a relapse, 15 (71.4%) showed a chronic form and 6 (28.6%) showed an acute/subacute form. The incidence of relapse as to clinical form was the same: 15/307 (4.9%) in the CF and 6/93 (6.5%) in the AF (p = 0.60).
Out of the 15 patients in relapse with chronic PCM, 5 of them showed the severe chronic form (SCF), 9 of them showed the moderate chronic form (MCF), and 1 of them showed the mild chronic form (MiCF), while the 6 patients with acute/sub-acute PCM showed the severe acute/subacute form (SAF).
Fourteen out of 21 patients in relapse were treated with TMP-SMX and 7 with ITC. The duration of the treatment with TMP-SMX lasted 33 months, ranging from 28 to 91, and 48 (21–135) with ITC.
These 21 patients showed five diagnosis patterns of relapse: a) positive clinical, serological and mycological – 3; b) positive clinical and mycological – 7; c) positive clinical and serological – 4; d) only positive serological – 2; e) only positive clinical – 5. Regarding as to four of these patterns, in addition to clinical signs, a positive mycological and/or serological test was noted. The five patients who presented only clinical signs were diagnosed as relapse patients based on the exclusion of the other possible etiologies and progress to cure after resumption the treatment with TMP-SMX. In contrast, 2 patients only had a serological relapse, which progressed to a negative response after resuming of the antifungal treatment. The case concerning 9 (47.4%) patients with clinical and/or mycological relapse should be highlighted because the DID tested negative.
Relapses occurred from 46 to 296 months (96 on the average) after treatment introduction and from 4 to 267 months (60 on the average) after treatment discontinuation. The time of relapse did not show any differences compared to the clinical forms: (a) due to treatment introduction: AF = 77.5 months (38–296); CF = 112.0 (46–234); p>0.05; (b) due to treatment discontinuation: AF = 38.5 months (4–267); CF = 75.0 (14–168); p>0.05.
Thus, relapse is defined in this paper as the reappearance of symptomatology suggesting PCM (the etiology of which has been confirmed by mycological and/or serological tests by DID), specific antibodies serum level regression to negative (non-reactive), and persistence of such conditions for at least one year after antifungal therapy discontinuation in patients with prior etiological diagnoses of and successful treatment for PCM, which included clinical cure and the normalization of erythrocyte sedimentation rate. This approach is common in clinical practice, and this paper is aimed at addressing the problem using other methods of serological diagnosis, rather than trying to identifying the source of new infectious processes; otherwise, the research design would have to have been completely different.
The prevalence of relapse out of the 400 patients was equal to 5.2%, i.e., lower than the 13.8% that was observed by Marques [4], who followed up with 58 patients. The prevalence of relapse did not vary according to clinical form, a finding that confirms those of Marques [4]. The diagnosis of relapse was late, i.e., within an average of 60 months after antifungal therapy discontinuation. Marques [4] findings revealed that 50.0% of the reactivations could be observed 36 months after antifungal therapy discontinuation, while our findings presented that 66.6% of the reactivations could be observed after the same period.
Few authors address the subject of PCM relapse, an event that is not rare which can bring serious repercussions to patients, since preserved organs may be impaired and the injured can worsen. In addition, a low prevalence of serological relapse can lead to complications in future diagnoses because it can suggest another manifesting disease, which may lead the MD to opt for unjustified treatments, thus delaying the diagnosis and treatment of the PCM relapse.
This study appraised different serological methods for PCM relapse diagnoses, from the DID reaction, which has been the method of choice due to its specificity, positive predictive value, repeatability and simplicity [6], [17], to immunoblotting with gp43 identification, which to the best of our knowledge and belief, is routinely available only at the Adolfo Lutz Institute (São Paulo). A comparison of both moments, the initial pretreatment and relapse, disclosed that the ELISA test had a higher sensitivity than the DID reaction, which provided several patients with a diagnosis that was not discovered by the gel precipitation test. These findings evince that the ELISA test should be the method of choice for diagnosing PCM relapse. The sensitivity of EIA is still only 65% so a negative result does not rule out relapse infection.
Several mechanisms might explain negative serological results in some relapse patient. One is that the fungus causing the relapse may not be P. brasiliensis, a second species could induce the production of DID-undetectable antibodies when using the same antigen as that in the initial reaction. Another explanation would be the modification of the antigenic composition of quiescent fungi, which would initiate the production of modified antibodies that are undetectable by methods designed on the basis of the typical antigens of P. brasiliensis.
Two other findings suggest a third hypothesis to explain the absence of antibody detection patients in relapse. The first is the polymorphism of the gene encoding the P. brasiliensis gp43 immunodominant antigen [18]. The second is the finding of two different isolates of P. brasiliensis in different organs of the same armadillo [19], a finding also observed in a patient with two genetically distinct isolates of P. brasiliensis. Clinical isolates were obtained from injuries on different anatomic sites and characterized by means of the RAPD technique. The genotypic evaluation showed more than 28% variability between these fungal isolates, therefore suggesting that different genotypes of P. brasiliensis may infect the same patient and induce active disease [20]. Thus, it is possible that some patients have been infected by more than one of P. brasiliensis isolate, with a positive serology to one of them; the other isolate, responsible for the relapse, could not be serologically recognized [21].
Other studies confirm these hypotheses. The cryptic species of Paracoccidioides brasiliensis, S1, PS2, PS3 and Paracoccidioides lutzii, recently identified [22]–[24], have implications on PCM immunodiagnosis [25]–[26], and presented a regional distribution. Such results suggest that there are differences in the fungus antigenic composition of P. brasiliensis. Arantes et al. [27] collected aerosol samples for the environmental detection of Paracoccidioides ssp., by placing a cyclonic air sampler at the entrance of the armadillo burrow in Botucatu (SP). Most ITS sequences showed a high similarity with homologous sequences of P. lutzii in the GenBank database, suggesting that this species – Paracoccidioides lutzii – may not be exclusive in Midwest Brazil.
Studies involving other microorganisms also confirm such hypotheses. Ghannoum et al. [28] analyzed clinical isolates of Cryptococcus neoformans obtained from five patients with recurrent cryptococcal meningitis and demonstrated the different composition of sterols between relapse and pre-treatment isolates, which indicates that the sterols had been modified by therapy or that patients were infected with new isolates with different sterol compositions.
Soll et al. [29] monitored Candida albicans strains isolated from different body sites of a single patient in three events of recurrent vulvovaginal candidiasis. The strains were evaluated by using Southern blot hybridization. They observed three different strains of C. albicans colonizing five sites, at the time of the first infection. Although the same strain of C. albicans had been responsible for the three vaginal infections, different colony phenotypes manifested with every new infection.
Such findings suggest that the action of antifungal and/or immune response of patients could lead to changes in the antigenic composition of fungi that are responsible for relapse, which would induce the production of antibodies that do not recognize the antigens used in routine serological tests. Hence, diagnostic methods based on proteomics could contribute to the solution of this problem. Evaluations in that direction are already in the course of pilot study in our Service. Another possible method of diagnosis of PCM relapse would be the analysis of circulating [30] and/or urinary [31] antigens, the methods of which have been evaluated successfully in the initial diagnosis of this mycosis and in follow-up with these patients. Nevertheless, such methods are not yet routinely available in clinical laboratories, especially those that are public.
Polymerase Chain Reaction (PCR) with primers specific for P. brasiliensis detection in clinical materials and other molecular method could also be excellent options for diagnosing PCM relapse, however, these techniques have not been incorporated yet into the routine of clinical laboratories [32]. In addition, PCR sensitivity is low in serum samples [33], the aim of our study, although higher in other clinical materials [32]–[35].
As a final point, while more sensitive and specific serological methods are assayed, we should continue to give special attention to the mycological diagnosis of PCM relapse. Such methods include the gold standard of diagnosing PCM. Hence, direct mycological [36]–[37], cytopathological [36]–[38], histopathological [36] and cultivation tests for P. brasiliensis [36]–[39] must be appropriately and routinely conducted when following up with such patients.
|
10.1371/journal.pgen.1003686 | Introns Regulate Gene Expression in Cryptococcus neoformans in a Pab2p Dependent Pathway | Most Cryptococccus neoformans genes are interrupted by introns, and alternative splicing occurs very often. In this study, we examined the influence of introns on C. neoformans gene expression. For most tested genes, elimination of introns greatly reduces mRNA accumulation. Strikingly, the number and the position of introns modulate the gene expression level in a cumulative manner. A screen for mutant strains able to express functionally an intronless allele revealed that the nuclear poly(A) binding protein Pab2 modulates intron-dependent regulation of gene expression in C. neoformans. PAB2 deletion partially restored accumulation of intronless mRNA. In addition, our results demonstrated that the essential nucleases Rrp44p and Xrn2p are implicated in the degradation of mRNA transcribed from an intronless allele in C. neoformans. Double mutant constructions and over-expression experiments suggested that Pab2p and Xrn2p could act in the same pathway whereas Rrp44p appears to act independently. Finally, deletion of the RRP6 or the CID14 gene, encoding the nuclear exosome nuclease and the TRAMP complex associated poly(A) polymerase, respectively, has no effect on intronless allele expression.
| Cryptococcus neoformans is a major human pathogen responsible for deadly infection in immunocompromised patients. The analysis of its genome previously revealed that most of its genes are interrupted by introns. Here, we demonstrate that introns modulate gene expression in a cumulative manner. We also demonstrate that introns can play a positive or a negative role in this process. We identify a nuclear poly(A) binding protein (Pab2p) as implicated in the intron-dependent control of gene expression in C. neoformans. We also demonstrate that the essential nucleases Rrp44p and Xrn2p are implicated in two independent pathways controlling the intron-dependent regulation of gene expression in C. neoformans. Xrn2p regulation seems to depend on Pab2p whereas Rrp44p acts independently. In contrast, the other exosome nuclease Rrp6p and the TRAMP associated poly(A) polymerase Cid14p do not appear to be implicated in this regulation. Our results provide new insights into the regulation of gene expression in eukaryotes and more specifically into the biology and virulence of C. neoformans.
| Introns, discovered in 1977, are genomic sequences that are removed from the corresponding RNA transcripts of genes [1]. First considered just as elements to be removed for correct gene expression, it has since become obvious that they participate in many aspects of gene regulation. Actually, the presence of introns and their splicing by the RNA-protein complex named spliceosome [2] affect gene expression by different means [3] including transcription, polyadenylation, mRNA export, mRNA localisation, translation efficiency and the rate of mRNA decay (see [4] for review). Most eukaryotic genes contain introns although the proportion of genes containing introns is highly variable between organisms. For example, whereas 92% and 78% of the genes in human and plant genomes contain introns, respectively, [5], [6] introns are found in only 5% of the genes in the yeast Saccharomyces cerevisiae [7]. Furthermore, the influence of introns on gene expression differs from one organism to another and from one gene to another.
In mammals, the expression of most of the genes is reduced in the absence of splicing but the effect of introns on gene expression is generally modest [8]. In contrast, the expression of some genes like the β-globin gene or the purine nucleoside phosphorylase gene has been shown to be highly intron-dependent [9], [10]. Introns act mainly at a post-transcriptional level and their absence reduces nuclear and cytoplasmic mRNA accumulation, alters efficient mRNA 3′end formation and consequently reduces nuclear mRNA export [8], [11], [12]. Introns seem also to regulate mRNA translation efficiency [8], [11], [12]. Similarly in plants most mutations can be complemented by cDNA sequences suggesting that most genes do not require introns for expression. For a few genes however, IME (intron-mediated enhancement) of gene expression has been demonstrated [13]. IME has been shown to act at a post transcriptional level and to be, at least for some genes, independent of splicing per se [14], [15]. More recently, IME has been shown to regulate 3′UTR formation and, to a lesser extent, translation [15]. However, in both plants and mammals, the pathway by which mRNAs transcribed from intronless alleles are degraded has not been described [16].
In fungi, the information is even more sparse, with most data coming from studies on S. cerevisiae, in which introns are rare and generally not necessary for gene expression [17], [18]. In a few examples however, introns have been shown to be necessary for gene expression, controlling the export of mRNA from the nucleus [19], [20]. More recently, introns have been shown to be key modulators of ribosomal protein gene expression in the baker's yeast [21]. In the other hemiascomycete yeasts in which the percentage of intron containing genes goes from 2.4% in Candida glabrata to 14.5% in Yarrowia lipolytica [22], introns do not seem to be necessary for gene expression although no specific studies have been reported. Similarly, in Schizosaccharomyces pombe in which 47% of the genes contain introns [23], these are generally not necessary for gene expression [24]. In filamentous fungi like Aspergillus nidulans or Neurospora crassa, cDNA sequences have been widely used for the production of heterologous or homologous proteins [25]–[27] suggesting only a moderate influence of introns on gene expression. In two cases however, one in Podospora anserina and one in Trichoderma viride, introns were reported to be necessary for gene expression but the mechanisms by which this regulation occurs have not been studied [28], [29]. Similarly in basidiomycetes, although intron density is generally higher than in ascomycetes [30], only a few cases of alteration of gene expression by the elimination or the addition of introns have been described [31]–[35]. In Schizophyllum commune, the addition of one intron in a GFP reporter gene has been shown to increase gene expression by altering mRNA accumulation rather than the level of transcription although no further description of the mechanisms by which this regulation occurs has been reported [31].
Cryptococcus neoformans is a capsular basidiomycete yeast mainly studied because it is responsible for opportunistic infections in patients presenting a cellular immune deficiency (mainly AIDS patients) that are fatal if left untreated [36]. The presence of an antiphagocytic polysaccharide capsule and the production of the antioxidant melanin are its two major virulence factors [37], [38]. The genome (20 Mb) sequences of five strains, two of serotype D, one of serotype A, and two of serotype B are now complete [39], [40]. The sequences of the 14 chromosomes of the serotype D strains were annotated using 21000 cDNA sequences isolated from a normalized library. Of the 6574 predicted genes, 80% had confirmed transcripts associated with them. Interestingly, C. neoformans genes are intron-rich and more than 98% of them have been reported to contain introns. Thus, C. neoformans has probably the intron-richest annotated genome described to date. These introns (5 on average per gene) are very small in size (67 bp) whereas exons have a size (250 bp) close to the human ones [40], [41].). Alternative splicing has been reported to be very common in C. neoformans and intron retention represents its most common manifestation [40], [42]. Finally, the fact that the proteome of C. neoformans contains numbers of proteins sharing sequence similarities with known metazoan SR proteins ([43]; Janbon unpublished data) as well as the identification of a DEAD-box helicase as a central regulator of multiple virulence factors [44] suggest that intron-dependent regulation of gene expression might play a major role in C. neoformans biology and virulence.
In this article, we have addressed the importance of introns for gene expression in C. neoformans. We have shown that introns are necessary for mRNA accumulation for some genes but not for others. We also demonstrated that introns can play a positive or a negative role in this process. Finally, we showed that the nuclear poly(A) binding protein Pab2 and the exosome nuclease Rrp44p are implicated in this intron-dependent regulation of gene expression in C. neoformans. Our results also suggested that Xrn2p might act in the same pathway as Pab2p.
We previously reported that the CAS3 gene contains 12 introns, all of them but the last one (intron 12) being located within the CDS [45]. We performed RACE experiments and noticed that among the five 5′end cDNAs sequenced, two were copies of RNA molecules not spliced in the intron 1 whereas the intron 2 was spliced. In order to identify the different types of CAS3 mRNA molecules present in the cell we sequenced a large number of full length cDNAs. Poly(A) RNA molecules were purified from C. neoformans var. neoformans cells growing in YPD and used for RT-PCR experiments. After separation by gel electrophoresis and purification, 3 pools of 15 full length cDNA molecules were cloned and sequenced (see Material and Methods). As presented in Figure 1A, a large diversity of CAS3 RNA molecules was identified, ranging from completely spliced molecules to completely unspliced ones. Although these experiments were not quantitative, the pattern of splicing observed revealed that some introns were more rarely spliced than others. Introns 1 and 12 were spliced in only 51% and 18% of the 45 sequenced molecules, respectively. RNA-Seq data alignment pattern analysis confirmed that all the introns from this gene can display a certain level of intron retention (Janbon, unpublished data). With the obvious exception of intron 12 which lies in the 3′UTR, all introns of this gene contain at least one in frame stop codon suggesting that none of these intron-containing mRNA molecules could encode a protein.
Evidence for intron retention at the CAS3 locus suggested that at least part of the regulation of this transcript was dependent on introns. So as to analyse the influence of intronic sequences on CAS3 expression we replaced the CAS3 wild type allele by a version without introns (cas3Δi). The co-transformation procedure used here allowed a complete allele replacement at its original locus without any further modification of the local genomic landscape (see Material and Methods). The main phenotype associated with the deletion of CAS3 is a modification of the capsule structure that can be revealed using anti-capsule monoclonal antibodies [46]. As shown in Figure 1B, the capsule structure of the strain bearing cas3Δi was similar to the one in which the gene had been deleted (cas3Δ) suggesting that the intronless allele was not functional. Moreover, Northern blot experiments showed that very little CAS3 specific RNA was present in the cas3Δi strain (Figure 1B). Introns are thus necessary for CAS3 expression.
To verify whether the importance of introns on gene expression was a general feature in C. neoformans, we cloned cDNAs from the genes UXS1, CAP10, UGE1 and CAS4 under the control of their own promoter. These constructs were used to transform the corresponding deletion mutant strains. We then compared mRNA levels of intronless and wild type alleles by Northern blot analysis. As presented in Figure 1C, the influence of introns on mRNA level was gene-dependent. Some genes, like CAP10, UGE1 and to a lesser extend CAS4 were highly intron-dependent whereas others like UXS1 did not depend on the presence of such sequences to be expressed.
The intron-dependent regulation of mRNA accumulation could act at different levels. Thus, the absence of RNA in the cas3Δi strains could be due to the absence or a very low level of transcription or/and to a decrease of the stability of the corresponding RNA leading to a complete or nearly complete degradation of it. To answer this question, we performed nuclear run-on experiments, thus measuring the frequency of transcription initiation of the different alleles largely independently of the effects of RNA stability [47]. The ratio of the CAS3 specific signal versus the ACT1 specific was not altered by the absence of introns (1.26±0.25) when compared to the wild-type allele (1.29±0.47) (Figure S1). These results suggested that the absence of introns does not alter the transcriptional activity of the gene but rather greatly alters the stability of RNA molecules transcribed from the cas3Δi allele.
So as to better analyse the influence of the introns on gene expression, we constructed a series of alleles bearing different numbers of introns at different positions. These alleles were integrated at the wild-type locus following the same procedure used previously. The level of mRNA was then measured by Northern analysis and confirmed by RT-qPCR (Figure 2). As previously shown, in the absence of introns, very few transcripts can be detected (less than 1% of the wild-type). Surprisingly, the presence of one intron was not enough to restore any expression of the gene as demonstrated by the analysis done with the strains NE293, NE295 and NE300 where the introns 12, 1, or 11 were present, respectively. Even with 2 introns, the expression of the gene remained barely detectable (see strains NE294, NE457 and NE449). The intron 12 and to a lesser extent the intron 1 appeared to play a negative regulatory role in CAS3 expression. Indeed, in strains NE298 and NE299 in which the CAS3 allele lacks the intron 1 and 12, respectively, the expression of this gene went up 2–3 fold. The negative role of the intron 12 in CAS3 gene expression was confirmed by comparing the expression of the CAS3 alleles from the strains NE454 (without intron 12) and NE453 (with intron 12). By comparison the absence of intron 2 influenced poorly the level of expression of the gene (strain NE456). The presence of the other introns regulated CAS3 expression in a positive way. In fact, excepting the regulation by introns 1 and 12, the more introns were present in CAS3 the better the gene was expressed. The identity of the introns did not seem to be important. Indeed, deletion of introns 2 to 6 (strain NE296) altered CAS3 mRNA level to the same degree as the deletion of introns 7 to 11 (NE453). Accordingly, the negative effect of the intron 12 appeared to be more dependent on its position than on its sequence as the re-positioning of the intron 2 at the intron 12 position in the cas3Δi12 allele results in a wild type expression of this intron swapped allele (127% of mRNA accumulation as compared to the wild type) (Figure S2).
Finally, dot blot assays using an anti-capsule antibody were performed to see whether CAS3 mRNA levels correlated with the phenotype of the corresponding strains. Results shown in Figure S3 demonstrated that a level of expression of CAS3 of at least 37% (Figure 2) of the wild-type mRNA level is associated with a wild-type capsule phenotype.
We aimed to identify elements involved in the degradation of the RNA molecule transcribed from the intronless allele cas3Δi. We constructed and screened an insertional library of C. neoformans mutants using a dot blot assay and the anti-capsule monoclonal antibody CRND-8. More than 5000 mutant strains were tested and fourteen strains were identified as having a low but detectable reactivity with this antibody (data not shown). Northern blot analysis confirmed that all of them expressed the cas3Δi allele at a low level (data not shown). Analysis of the position of the insertion site in the first mutant strain studied, revealed that it was within the gene CNB04570 coding for a protein of 210 amino acids sharing 49% and 32% of amino acid sequence identity with the nuclear poly (A) binding protein of S. pombe [48] and the human one, respectively [49]. Like its human and fission yeast counterparts, the C. neoformans Pab2p sequence presented a single RNA binding domain and an arginine-rich C-terminal domain. However, the poly-alanine domain present in the human protein in which mutations associated with the genetic disease named Oculopharyngeal muscular dystrophy (OPMD) have been identified [49], was absent in both fungal proteins.
We deleted PAB2 using a nourseothricin marker (see Material and Methods). As previously reported in S. pombe, the pab2Δ mutants grow less well at 15°C as compared to the wild-type strains. In C. neoformans, we also found that this mutation results in an alteration of the growth rate at 30°C and 37°C (Figure 3A) and an increased sensitivity to SDS 0.01% as compared to the wild-type. We also studied the classically associated virulence phenotypes and found no evidence of modification of the capsule size or structure and no alteration of the urease production (data not shown). In contrast, we observed a small but reproducible reduction in melanin production (Figure 3A).
A pab2Δ cas3Δi strain was constructed by selecting adapted progenies after crossing the single mutant strains. Analysis of the expression of this intronless allele in a pab2Δ genetic background confirmed that this protein regulates cas3Δi expression. Indeed, as shown in Figure 4A, whereas PAB2 deletion did not increase the expression of the CAS3 wild type allele, it restored the expression of the intronless allele cas3Δi up to 12% of the wild type (Figure 4A, central panel, grey bars). We also confirmed by ELISA using another anti-capsule monoclonal antibody (Mab 302) that the level of mRNA accumulation in these strains correlated with the phenotype of the corresponding strains (Figure 4A, right panel).
We noticed the presence of two additional bands present in Northern blots for the pab2Δ cas3Δi mutant strain (Figure 4A, left panel, lane 4). These bands are probably products of partial degradation of the transcript or the result of partial transcription of the gene. Indeed, hybridizing with oriented RNA probes or with probes specific for the 5′ or 3′ends of the gene demonstrated that these additional bands correspond to the 5′end of the sense transcripts (Figure S4).
We also purified RNA from nuclei isolated using a similar protocol as the one used for the run-on experiments (see Material and Methods) and compared the accumulation of RNA obtained with this nuclei enriched fraction with the ones obtained with RNA extracted from intact cells. As shown in Figure 4A (middle panel), we observed a more pronounced accumulation of the cas3Δi mRNA in the nuclei enriched fraction (white bars), this more pronounced accumulation being exacerbated by a pab2Δ mutation. In good agreement with the localisation experiment data (see below), these results also suggested a nuclear role for Pab2p in the control of intronless allele expression.
Finally, we checked that Pab2p could also modulate the expression of intronless alleles of other genes by constructing a pab2Δ cap10Δi double mutant strain. As shown in Figure S5, the PAB2 deletion also restored the cap10Δi mRNA level close to the wild type level confirming the role of Pab2p in the control of intronless allele expression.
To functionally characterise the biophysical properties of Pab2p, we expressed an N-terminal 6XHis-tagged version in E. coli and purified the recombinant protein (see Material and Methods). We tested the affinity of this recombinant Pab2p towards a 30-mer poly(A) RNA coated in a 96-well plate well (see Material and Methods). We found that Pab2p recognized poly(A) oligonucleotides in a dose dependent way (data not shown) and that this recognition was specific to RNA as the same protein presented very little affinity to a 30-mer poly(A) DNA (Figure 4B). Similarly to what has been observed in S. pombe [48], competition assays suggested also that the binding was specific to poly(A) sequences as a poly(C) RNA sequence was not able to compete the binding of Pab2p to poly(A) (Figure 4B).
Next we constructed a GFP::PAB2 allele to localize Pab2p within the cell. We transformed a pab2Δ cas3Δi strain and checked that the transformant selected grew as well as the wild-type strain at all temperatures tested. The functionality of the GFP::Pab2p fusion was confirmed by Northern blot which demonstrated that the fusion protein decreased expression of the allele cas3Δi (data not shown). Examination, by fluorescence microscopy showed a pattern of Pab2p of fluorescence consistent with nuclear localisation (Figure 4C). These results suggested strongly that Pab2p is a nuclear poly(A) binding protein.
Pab2p has been recently shown to interact with the two nucleases of the exosome (i.e. Rrp44p and Rrp6p) to control the synthesis of snoRNAs and the expression of meiotic genes [50]–[52]. It was thus very tempting to hypothesize that this multi-protein complex could regulate the expression of cas3Δi by degrading the RNA transcribed from this allele.
We identified the RRP6 (gene CNC03940) and the RRP44 (gene CND00800) homologues in the genome of C. neoformans and constructed corresponding mutant strains. As in the model yeasts S. cerevisiae and S. pombe, RRP6 was not essential and we were able to delete this gene in C. neoformans. The phenotypes associated with the RRP6 deletion were compared with the ones associated with the PAB2 deletion. As presented in Figures 3A and 3B, pab2Δ and rrp6Δ strains presented a similar growth defect at 30°C. However, in contrast to what we observed with the pab2Δ mutant strains, the rrp6Δ mutants growth defect is not exacerbated when the cells are incubated at 15°C or 37°C and no hyper-sensitivity to SDS 0.01% was observed. The size and structure of the capsule and the urease production were not affected by the deletion of the RRP6 gene (not shown). Interestingly, we observed the same slight defect in melanin production in pab2Δ and rrp6Δ mutant strains when the cells were grown on Niger medium (Figure 3B). Successive unsuccessful attempts to delete RRP44 suggested that this gene is essential in C. neoformans as it is in S. cerevisiae and S. pombe [53], [54]. We thus expressed this gene under the control of the GAL7 promoter which has been shown to be strictly regulated by the presence of galactose in the medium and can be used as a regulatable promoter in promoter swap experiments [55]. On galactose, these cells displayed no specific phenotype although RRP44 was clearly over-expressed (Figure 5B and 5C) whereas on glucose, the PGAL7::RRP44 strains failed to grow, confirming that this gene is essential in C. neoformans (Figure 3C).
To identify the nuclease that regulates expression of intronless CAS3, the cas3Δi allele was next introduced into the rrp6Δ and rrp44 mutant strains. Moreover, we constructed all possible double mutant strains (rrp6Δ pab2Δ, rrp6Δ PGAL7::RRP44 and pab2Δ PGAL7::RRP44) and we introduced the cas3Δi allele in all of them. Deletion of RRP6 did not restore even partially the expression of cas3Δi (Figure 5A), suggesting that Rrp6p is not the nuclease degrading the RNA transcribed from the cas3Δi allele. Surprisingly, RRP6 deletion in a pab2Δ background resulted in reversion to a complete absence of expression of the cas3Δi allele.
We next tested the influence of Rrp44p on the control of cas3Δi expression. To do so, we grew the different strains under the non-restrictive condition (galactose) and then transferred them to the restrictive condition (glucose). Preliminary experiments had shown that as early as 2 hours after the transfer of the cells to glucose medium no RRP44 specific mRNA could be detected by Northern blot analysis when this gene was expressed under the control of the GAL7 promoter (not shown). We compared mRNA levels of CAS3 and cas3Δi under non-restrictive conditions and after 10 hours under restrictive growth conditions in the different genetic backgrounds. Whereas CAS3 mRNA levels were similar in all mutant strains tested, incubation of PGAL7::RRP44 cells under restrictive conditions (glucose) restored the expression of the intronless allele cas3Δi up to 9% of the wild type (Figures 5B and 5D). Similar results were obtained after shorter (6 h) incubation times. This mRNA level was not increased in the rrp6Δ PGAL7::RRP44 double mutant confirming that Rrp6p is not implicated in this regulation. Interestingly, RRP44 appeared to be up-regulated in the absence of RRP6 suggesting a potential explanation for the absence of cas3Δi mRNA in the rrp6Δ pab2Δ double mutant (Figure 5A). The analysis of the double mutant pab2Δ PGAL7::RRP44 revealed a synergic effect of these mutations. As shown in Figures 5C and 5D, the double mutant strains expressed the intronless allele up to 34% of the wild type. Accordingly, the level of mRNA correlated with the phenotypes of the corresponding strains (Figure 5D). These results strongly suggested that Rrp44p and thus the exosome participates in the degradation of mRNA transcribed from the intronless allele cas3Δi. They also demonstrated that the exosome is acting mainly independently of Pab2p suggesting the existence of a least two pathways regulating intronless expression in C. neoformans.
Several RNA species including snRNA, snoRNA, tRNA and rRNA are targeted to degradation by the exosome following polyadenylation by the TRAMP complex [56]. We thus addressed whether this nuclear complex has a role in the regulation of cas3Δi expression. Cid14p has been shown to represent the catalytic subunit responsible for the TRAMP complex poly(A) polymerase activity in S. pombe [57]. We deleted the single homologous gene (CNK02250) in the C. neoformans genome. Neither the cid14Δ strains nor the cid14Δ pab2Δ double mutant strains had any growth phenotype at any temperature tested (30°C, 37°C, 15°C) (not shown). Moreover, no alteration of CAS3 or cas3Δi mRNA levels could be observed (Figure S6). Thus, Cid14p and the TRAMP complex do not seem to be implicated in the regulation of the expression of intronless alleles in C. neoformans.
Finally, as Pab2p has been previously implicated in poly(A) tail length control [58], we performed poly(A)-tests to examine the length of the poly(A) tail in the wild type or in the absence of Pab2p or Cid14p (see Material and Methods). The consequences of these gene deletions on the poly(A) tail length of 10 C. neoformans genes including CAS3 were tested. However, we observed no reproducible modification of length of the poly(A) tails in any of the mutants tested (Figure S7, data not shown).
In keeping with a primarily nuclear degradation of cas3Δi (as supported by the nuclear localisation of Pab2p and an enrichment in nuclear fractions) and given the results obtained for the two exosomal nucleases (Rrp6, Rrp44p), the major nuclear 5′→3′ exonuclease Xrn2p/Rat1p appeared to be the most promising candidate for Pab2p-assisted degradation of intronless mRNA. This idea was further encouraged by the finding that deletion of PAB2 not only stabilises full-length cas3Δi but also truncated fragments corresponding to the 5′ end of the sense transcript thus suggesting the involvement of a 5′→3′ exonuclease. In addition, Xrn2p and Rat1p have been shown to be involved in the degradation of unspliced transcripts in human and yeast [59], [60].
Given that XRN2 is likely essential in C. neoformans, we placed the gene (CNF01810) under the control of the GAL7 promoter, similar to the strategy applied for RRP44. As expected, cells failed to grow under restrictive conditions (glucose) showing that XRN2 is indeed essential for C. neoformans viability. However, depletion of XRN2 did not lead to any stabilisation of the cas3Δi transcript neither in a wildtype nor in a pab2Δ context (Figure 6A). Instead a slight decrease of cas3Δi mRNA was observed upon XRN2 depletion in pab2Δ strains (Figure 6A). Accordingly, when we compared the levels of RRP44 expression in wildtype and xrn2 mutant cells, we found an increase in RRP44 expression upon XRN2 depletion (Figure 6B). Likewise expression of XRN2 is elevated in cells depleted for RRP44 (Figure 6B), which might partially explain the rather minor stabilisation of cas3Δi transcripts in rrp44 cells.
To circumvent this compensatory effect and thus be able to evaluate the role of Xrn2p, we compared cas3Δi mRNA levels when overexpressing either RRP44 or XRN2 in wildtype and pab2Δ backgrounds. To this end, wildtype, pab2Δ, rrp44, xrn2, pab2Δ rrp44 and pab2Δ xrn2 strains were grown in inducing conditions (galactose) and the levels of cas3Δi expression were measured by Northern analysis. Note that upregulation of the intronless allele was reproducibly observed when the strains were grown in galactose (see Figure 5C). Overexpression of RRP44 or XRN2 in a PAB2 wildtype context led to the nearly complete degradation of cas3Δi mRNA preserved by growth in galactose suggesting that both nucleases are implicated in the degradation of these mRNA molecules (Figure 6C). On the other hand, overexpression of RRP6 did not lead to any destabilisation of the cas3Δi transcript thus confirming that Rrp6p has no central role in this regulation (data not shown).
In the absence of Pab2p, RRP44 overexpression led to a strong decrease of cas3Δi mRNA accumulation and to nearly complete elimination of the two additional bands observed by Northern blot in a pab2Δ single mutant context (Figure 6C). In contrast, XRN2 overexpression in a pab2Δ strain led to a very moderate or no decrease of cas3Δi mRNA accumulation. These results were confirmed by RT-qPCR (Figure 6D) although the effect of Xrn2p is probably overestimated in this assay due to the choice of primers specific for the transcript's 5′ end. In conclusion, these overexpression experiments confirm on the one hand that the action of Rrp44p is mainly independent of Pab2p and on the other hand suggest a rather Pab2p-dependent role for Xrn2p in the degradation of the intronless mRNA.
The role of introns on gene expression has been the focus of a large number of studies during the last decades [4]. Most of these studies are coming either from mammals or plants or have been performed using the intron-poor micro-organism S. cerevisiae as a model. In these organisms, the replacement of a wild-type gene by an intronless allele generally has a modest effect on gene expression suggesting that introns are more a source of protein diversity or/and regulation of gene expression than a sine qua non condition for a gene to be expressed [8], [18], [21], [61]. In contrast, expression of some genes like the human β-globin or the plant ERECTA genes is highly dependent on the presence of introns [10], [15]. Most of the intron-dependent regulation occurs at a post-transcriptional level although the different steps necessary for the production of a mature mRNA, including transcription and splicing are mutually dependant. [62]. In these cases the presence of introns in a transcript can affect 3′end formation, mRNA export from the nucleus and mRNA stability [8], [63]. However, the pathway(s) by which mRNA molecules transcribed from intronless alleles are recognized and degraded remain unknown [64].
The intron density of the pathogenic yeast C. neoformans is probably the highest yet known for an organism having a completely annotated genome. In fact, a recent re-annotation of the C. neoformans var. grubii genome based on RNA-Seq data showed that 99% of the expressed genes have at least one intron (Janbon, unpublished data). Moreover, 11.5% and 4.1% of genes in C. neoformans have been shown to have 5′ and 3′ UTR introns, respectively [65]. It has also been previously published that alternative splicing is very common in C. neoformans [40], [42]. More recently, a link between transposon pre-mRNA splicing and RNAi dependent degradation has been demonstrated in C. neoformans [66]. Altogether, these data suggested a central role for intron metabolism in the biology and the virulence of C. neoformans. In this study, using the gene CAS3 as a model transcript, we showed that alternative splicing can affect all introns from a single gene although their spliceability appeared to be intron-dependent. We also demonstrated that introns are necessary for the CAS3 gene expression in C. neoformans. Three other tested genes have the same intron-dependence of gene expression whereas another one (UXS1) can be expressed without introns. This insensitivity to the lack of introns does not appear to depend on the number of introns. Indeed CAP10 has only 3 introns, UGE1 only 4 whereas CAS4 and UXS1 have 9 and 7 introns, respectively. Moreover, it probably does not depend on the presence of an intron in the UTR as CAS3 is the only one of the presently studied genes to possess such an intron. It has to be noted that the fact that intronless bacterial antibiotic resistance genes are commonly used for mutant construction in C. neoformans does not contradict this observation. Indeed, all these genes are expressed under the control of the ACT1 promoter, in which an intron is present [67]–[69].
Our results demonstrated that most introns play a positive role on mRNA accumulation and that the absence of introns does not alter the level of transcription as measured by run-on transcription assay. These results are similar to what has been observed in mammals, in the fungus S. commune or for IME in plants in which the regulation of gene expression by introns acts mainly at a post-transcriptional level [8], [31], [61]. In contrast to what has been observed in most cases however, one intron is not enough to restore gene expression. Even with two introns the mRNA level remained below 3% of the wild-type. Most introns played a positive role on gene expression and their action seemed to be more cumulative than specific as previously reported for the ERECTA gene in A. thaliana [15]. The two most external introns (1 and 12) played a negative role on CAS3 mRNA accumulation. Run-on experiments suggested no transcription rate alteration associated with the deletion of either one of these introns (data not shown) suggesting also a post-transcriptional regulatory mechanism. More investigations are obviously needed to understand the role of these introns on mRNA accumulation.
The absence of introns results in an important reduction of mRNA accumulation. Deletion of the PAB2 gene partially stabilized mRNA transcribed from the intronless allele. In contrast, the analysis of the cid14Δ strains suggests no apparent role for the TRAMP complex in this regulation [52]. Pab2p has been shown to interact physically with the two nucleases (Rrp6p and Rrp44p/Dis3p) from the exosome in S. pombe [50]. In C. neoformans, although the deletion of RRP6 encoding the nuclear exosome nuclease has no effect on the accumulation of mRNA transcribed from cas3Δi, the analysis of the level of cas3Δi mRNA in a RRP44 conditional mutant under a restrictive condition strongly implicates this multiprotein complex in this regulation. The analysis of the double mutant strain pab2Δ rrp44 demonstrated a synergic effect of the two mutations suggesting that these two proteins could act in two independent pathways. [70], [71]. As suggested in other Pab2p dependent pathways described to date, Pab2p could be a facilitator for the degradation of cas3Δi mRNA through recruitment of another nuclease. Our double mutant strains analysis and overexpression experiments suggest that the nuclear 5′→3′ exonuclease Xrn2p might represent a good candidate. This model could explain the synergic effect of the pab2Δ rrp44 double mutation. Thus, in the single rrp44 mutant strain, XRN2 is over expressed and can partially compensate the effect of Rrp44p depletion whereas in the absence of Pab2p, XRN2 over-expression would have much less effect on the degradation of cas3Δi transcripts. The fact that Xrn2p/Rat1p has been previously shown to be involved in the degradation of unspliced mRNA in human and yeast [59], [60] sustains this model although no genetic interaction between Xrn2p and Pab2p has been reported to date.
Very recently, Pab2p has been shown to be involved in three different RNA processing and degradation pathways in S. pombe. Thus, together with the exosomal nucleases Rrp6p and independently of the TRAMP complex it controls polyadenylation and synthesis of snoRNAs [50], meiotic gene expression in the Mmi1-dependent pathway [51], [52], [72] and targets ribosomal pre-mRNA RPL30-2 [73]. Similarly, in the meiotic gene expression and pre-mRNA RPL30-2 regulation, a synergic effect was observed when RRP44 and PAB2 were mutated suggesting here also that the effect of Rrp44p could be mainly independent of Pab2p although the other elements involved in this Rrp44p-dependent pathway remain to be identified.
The role of Pab2p in the intronless gene expression regulation remains mysterious. In S. pombe, Pab2p is recruited to the nascent mRNA before 3′ end formation and polyadenylation and controls the length of poly(A) of only a subset of RNAs [48], [50], [71]. It also physically interacts both with the exosome nucleases and the poly(A) polymerase Pla1p [52], [71]. In the absence of Pab2p, some cas3Δi mRNA molecules are exported from the nucleus and translated although most of them are still degraded. The subcellular localisation of Pab2p together with the analysis of the accumulation of the mRNA transcribed from the intronless allele in the nucleus, suggested strongly a nuclear role for this protein although Pab2p has been shown to be able to shuttle to the cytoplasm in S. pombe and in Drosophila [70], [71]. The kinetic of degradation of intronless mRNA in C. neoformans might be the result of a disequilibrium between mRNA export from the nucleus and degradation. Thus, when not enough introns are present the altered dosage of mRNA binding proteins would result in an extended retention time of mRNA in the nucleus giving time to the nucleases to degrade them. The absence of Pab2p would slow down this degradation giving time to some mRNA molecules to be exported in a “take the money and run” strategy [74] (Figure 7).
In terms of evolution, the comparison of S. cerevisiae and C. neoformans provides a fascinating example of opposite evolutionary choices. Whereas S. cerevisiae has lost almost all its introns and has largely simplified its RNA metabolism (i.e. loss of RNAi pathway, only one SR protein, absence of EJC-like complex…), C. neoformans has conserved and maybe increased its intron number and appears to have a very complex RNA metabolism. The selective pressure that has maintained introns in one organism and has eliminated them in another one is unknown. It has to be noted that C. neoformans is not a unique example among basidiomycete fungi. Thus, genes from Coprinus cinereus and Phanerochaete chrysosporium have an intron density close to that of C. neoformans [30]. In two other pathogens, Ustilago maydis for the plants and Malassezia sp. for human for example, the number of genes with introns is small [75], [76]. This specificity might be related to the fact that C. neoformans is an opportunistic human pathogen living in the environment. As such the diversity of signals to which it can be exposed in the human body or in soil for example is huge. Indeed, this organism has to cope with a large number of different stresses and probably needs a very flexible metabolism. It is tempting to hypothesize that its complex RNA metabolism provides a mechanism to achieve such flexibility.
C. neoformans strains used in this study all originated from the serotype D strain JEC21 [77] and are listed in Table S1. The strains were routinely cultured on YPD medium at 30°C [78]. Synthetic dextrose (SD) was prepared as described [78]. The capsule sizes were estimated after 24 h of growth in capsule-inducing medium at 30°C as previously described [79]. Melanin and urease production were assessed after spotting 105 cells of each strain on Niger or Christensen agar medium, respectively [69], [80]; the plates were read after 48 h of incubation at 30°C. The bacterial strain Escherichia coli XL1-blue (Stratagene) was used for the propagation of all plasmids.
Cells were routinely harvested after being grown up to 5·107 cells/mL in YPD. RNA was extracted with TRIZOL Reagent (Invitrogen) following the manufacturer's instructions. Total RNA (5 µg) was separated by denaturing agarose gel electrophoresis and transferred onto Hybond-N+ membrane (Amersham) and probed with [32P]dCTP-radiolabelled DNA fragments. The banding pattern was quantified with a Typhoon 9200 imager and Image Quantifier 5.2 software (Molecular dynamics).
Total RNA was extracted from JEC21 cells growing on YPD. mRNA was purified using Oligotex Direct mRNA Mini Kit (Qiagen) following the manufacturer's instructions. SMARTer RACE cDNA Amplification Kit (Clontech) was used to synthesize the cDNA. CAS3 cDNA was PCR amplified using the primers CAS3a and CAS3AR (see Table S2) and was analysed by agarose gel electrophoresis. The presence of a smeary pattern on the gel suggested the presence of different types of cDNA molecules. The amplified fragments were gel purified in three different pools of sizes and 15 cDNA molecules from each pool were cloned in a pGEMT plasmid and sequenced.
An insertional mutant library was constructed in a NE292 (MATa cas3Δi ura5) background using the Agrobacterium tumefaciens strain EHA105 transformed with the plasmid pPZP-NEO1 as previously described [81]. A total of 5796 colonies were transferred from the transformation plates to 96-well plate wells containing 75 µL of capsule inducing medium [79] supplemented with adenine (20 mg/L) and uracil (5 mg/L). The mutants were then tested with the anti-capsule monoclonal antibody CRND-8 [82] as previously described [83]. Positive mutant strains were isolated and tested a second time using the same strategy. Total RNA was extracted and the presence of CAS3 mRNA was analysed by Northern blot.
PAB2 cDNAs were amplified by PCR and inserted into the pQ-30 E. coli expression vector (Qiagen). The E. coli BL21 transformant strains were grown in 50 mL of YT containing ampicillin (50 µg/ml) and kanamycin (30 µg/ml) to an OD600 of 0.5; gene expression was induced by adding 1 mM of IPTG and incubation for 4 hours at 37°C. The cells were then disrupted by sonication and centrifuged at 3000·g. The supernatant was recovered and the recombinant proteins were purified by affinity chromatography on a Ni-NTA column (Qiagen) following the manufacturer's procedures. The protein solution was adjusted to 20% (w/v) glycerol (final concentration 140 µg/mL) and stored in aliquots at −80°C.
These experiments were conducted in 96-well Streptavidin coated plates (Nunc). For each sample and concentration to be tested one well was washed three times with 200 µl of washing buffer (Tris 25 mM, Nacl 150 mM, pH 7.2, BSA 1% wt/vol, Tween 20 0.05% wt/vol). Each well was then incubated for 2 h at room temperature and under agitation (700 rpm) with 100 µl of washing buffer containing 0.1 µM of poly(A) 30-mers oligonucleotide 5′ biothinylated. Unbound oligonucleotides were then eliminated through three washes with 200 µL of washing buffer. Each well was then incubated with 100 µl of recombinant Pab2p solution (0.70 mg/mL) for 30 min at room temperature under agitation (700 rpm). After three washes with 200 µL of washing buffer, the quantity of poly(A)-bound protein was estimated using an anti-His peroxidase linked monoclonal antibody (Qiagen) and OPD (O-phenylenediamine dihydrochloride) (Sigma) following the manufacturer's procedures. After 10 min of incubation at room temperature, the colorimetric reaction was stopped by the addition of H2SO4 4% (v/v) and the optic density was measured at 492 nm. For the competition assays, 10 µM of unlabeled poly(A) or poly(C) (Sigma) was added to the protein solution.
To localize the Pab2 protein, the PAB2 gene under the control of its own promoter was joined in-frame to a sequence encoding the GFP protein at its N-terminal end. Primers used for amplification are listed in Table S2. A pab2Δ strain was transformed with a plasmid containing the URA5 marker and the Pab2-fluorescent protein fusion by biolistic delivery [84]. Transformants were grown on minimum medium and analyzed for fluorescence.
The pBluescript (Stratagene) based plasmid pNE247 contained a 4067 bp DNA fragment containing the complete C. neoformans CAS3 gene PCR amplified using the primers CAS3F and CAS3R (see Table S2) and cloned at the NotI site. This plasmid was used to construct all the CAS3 alleles presented in this study. For the intronless allele, the cDNA from CAS3 was amplified, cloned and sequenced (see above). A completely spliced molecule was digested with SphI and PstI and cloned at the SphI-PstI site of pNE247, thus replacing the wild type gene by an intronless version under the control of its own promoter. 5 µg of the resulting plasmid pNE254 were NotI digested and mixed with 1 µg of the URA5 containing plasmid pNE10 [79] digested with NotI. This DNA solution was used to transform the strain NE128 (MATa cas3Δ:ADE2 ura5) [46] by biolistic transformation. The transformants were selected on a minimum medium containing adenine at 20 mg/L. After three days at 30°C, the transformation plates were transferred to room temperature and one week after transformation some colonies developed a pink phenotype suggesting that the ADE2 gene had been lost and thus that the cas3Δ::ADE2 allele has been replaced by the cas3Δi allele. The pink colonies were then cultured in liquid YPD so as to loose the unstable pNE10 plasmid [79]. Ura− strains were selected on FOA. The absence of the cas3Δ::ADE2 allele and the correct integration of the intronless allele were confirmed by PCR. The absence of additional integrations in the genome was confirmed by Southern blot. Two independent mutant strains were selected and stored at −80°C. Similar strategies were used to construct the other alleles and the other mutant strains.
For nuclei purification, 500 mL of culture (OD600 = 3) were harvested by centrifugation. Spheroplasts were prepared as previously described [85] and re-suspended in 3 mL of lysing buffer (Pipes 10 mM pH 6.9, sucrose 0.5 M, CaCl2 5 mM, MgSO4 5 mM, DTT 1 mM) containing a complete set of antiproteases (Roche). The spheroplasts were then mechanically disrupted and the intracellular organelles were separated from cellular debris and unbroken cells by centrifugation. Nuclei were purified by differential ultracentifugation (1 hour, 161 000·g, 4°C) through a separation buffer (Pipes 10 mM, sucrose 2.1 M CaCl2 5 mM, MgSO4 5 mM, DTT 1 mM) containing a complete set of antiproteases (Roche). Nuclei were then washed twice with conservation buffer (TrisHCl 50 mM pH 8.3, glycerol 40%, MgCl2 5 mM, EDTA 0.1 mM pH 8), re-suspended in 500 µl of conservation buffer and stored in aliquots at −80°C. For each run on experiment 100 µL of nuclei suspension was used following a protocol previously described [47]. The radioactive transcripts produced were used to hybridize a serial dilution of DNA spotted on a nylon membrane. The plasmid pNE428 containing the 1558 bp fully spliced JEC21 CAS3 cDNA amplified with the primers CAS3a and CAS3AR and cloned in the pGEMT plasmid (Clontech) was used as CAS3 specific DNA. The plasmid pNE435 containing the DNA 519 bp DNA fragment PCR from JEC21 genomic DNA using the primers ACT1F and ACT1R cloned in pGEMT was used as ACT1 specific DNA. The plasmid pGEMT alone was used as negative control. The intensity of the signal was quantified with a Typhoon 9200 imager and Image Quantifier 5.2 software (Molecular dynamics). Each experiment was repeated twice using two independent nuclei preparations. The same protocol of nuclei preparation was used to isolate the nuclear RNA fraction. Electrophoretic analysis of these RNA samples showed a clear decrease in the rRNA proportion confirming the quality of our preparation (data not shown).
The genes described in this report have been deleted by biolistic transforming a serotype D strain using a disruption cassette constructed by overlapping PCR as previously described [45]. The primer sequences used are given in Table S2. The transformants were then screened for homologous integration as previously described [46]. The plasmid, pNAT used to amplify the NAT selective marker was kindly provided by Dr Jennifer Lodge (Saint Louis University School of Medicine). The plasmid pPZP-NEO1 used to amplify the NEO selective marker was kindly provided by Dr Joe Heitman (Duke University). Multiple mutant strains were obtained through crosses of single mutant strains on V8 medium as previously described [45]. Progenies were selected on minimum medium to which different amino acids were added. Their genotypes were determined by PCR. The mating types of the strains were determined by testing them on V8 medium in the presence of tester strains of known mating type.
A four way overlap PCR gene deletion was used to generate the promoter-specific exchange cassettes of RRP44 and XRN2, which included a nourseothricin and a neomycin cassette, respectively. The primers used in these experiments are listed in Table S2. The GAL7 promoter was used as the inducible promoter [55]. The 694 bp upstream the RRP44 ATG and the 601 bp upstream the XRN2 ATG were replaced by the 556 bp present upstream of the GAL7 gene. Transformants were screened for homologous integration as previously described [86].
The ePAT and TVN-PAT reactions were performed using 1 µg of input RNA as previously described [87]. The sequences of the primers used for PCR amplification are listed in Table S2. The cDNA was column-purified using NucleoSpin Gel and PCR Clean-up columns (MACHEREY-NAGEL). Specific PCR products were analysed by 2% high resolution agarose gel (Ultra pure 1000; Life Technologies) pre-stained with sybr safe (Life Technologies) and imaged against a 100 bp ladder (New England Biolabs) using an LAS 3000 imager and multigauge software (Fujifilm).
Total RNA was subjected to an initial DNaseI (Roche) treatment to eliminate contaminating genomic DNA. 1 µg of the DNaseI treated RNA was then reverse transcribed using the kit QuantiTect Reverse Transcription (Qiagen) following the manufacturer's instructions.
Quantitative PCR assays were performed according to Bio-Rad manufacturer's instructions using 96-well optical plates (Thermo Scientific) and an iCycler iQ (170–8740, Biorad).
Each run was assayed in triplicate in a total volume of 25 µL containing the 5 µL cDNA template at an appropriate dilution, 1× Absolute qPCR SYBR Green Fluorescein (Thermo Scientific) and 320 nM of each primer. The primers used are listed in Table S2. PCR conditions were: 95°C/15 min for one cycle, 95°C/30 s for 40 cycles. Amplification of one single specific target DNA was checked with a melting curve analysis (+0.5°C ramping for 10 s from 55°C to 95°C). The Ct values obtained in triplicate were averaged and normalised to that of the housekeeping gene ACT1 using standard curves. To verify the absence of genomic DNA contamination, negative controls in which reverse transcriptase was omitted were used. Three independent biological replicates were performed.
|
10.1371/journal.pcbi.1007259 | Pulse transit time estimation of aortic pulse wave velocity and blood pressure using machine learning and simulated training data | Recent developments in cardiovascular modelling allow us to simulate blood flow in an entire human body. Such model can also be used to create databases of virtual subjects, with sizes limited only by computational resources. In this work, we study if it is possible to estimate cardiovascular health indices using machine learning approaches. In particular, we carry out theoretical assessment of estimating aortic pulse wave velocity, diastolic and systolic blood pressure and stroke volume using pulse transit/arrival timings derived from photopletyshmography signals. For predictions, we train Gaussian process regression using a database of virtual subjects generated with a cardiovascular simulator. Simulated results provides theoretical assessment of accuracy for predictions of the health indices. For instance, aortic pulse wave velocity can be estimated with a high accuracy (r > 0.9) when photopletyshmography is measured from left carotid artery using a combination of foot-to-foot pulse transmit time and peak location derived for the predictions. Similar accuracy can be reached for diastolic blood pressure, but predictions of systolic blood pressure are less accurate (r > 0.75) and the stroke volume predictions are mostly contributed by heart rate.
| Recently there has been a strong trend for self-monitoring of your cardiovascular health and new wearable sport trackers and mobile applications are coming to the market everyday. However, such solutions are mostly taking advantage of heart rate measurement. Other health indices such as blood pressure and pulse wave velocity reflecting to the condition of cardiovascular system would also be of great interest, but such solutions for continuous monitoring are barely existing or are at least unreliable. In this paper, we use computational modelling to assess theoretical capabilities of such measurements. We concentrate on predicting health indices using on pulse transmit time type of measurements. Such measurements could be carried out, for example, with photopletyshmography sensor or an optical sensor already found from several wearable sport trackers. We use cardiovascular modelling to create a database of “virtual subjects”, which is applied with machine learning to construct predictors for health indices. Our findings suggest that aortic pulse wave velocity and diastolic blood pressured could be predicted with a high accuracy, but predictions of systolic blood pressure are less accurate.
| This paper considers continuous monitoring of cardiac health using computational modelling. Stiffening of the arterial wall, such as aorta, causes reduction in the pulsatile properties in the vascular tree, accelerates the vascular premature ageing and predisposes to the dysfunction of the heart, brain and other organs [1, 2]. Aortic stiffness can be measured by using invasive methods or medical imaging such as ultrasound [3] and MRI [2]. Another indicator reflecting the cardiac performance is stroke volume (SV), which is typically measured using Doppler ultrasound [4]. However, these imaging modalities typically require special expertise and are only carried out clinically. On the other hand, aortic stiffness is associated with the unfavourable changes in the diastolic and systolic blood pressures (DBP/SBP), which can have several negative consequences in cardiac function and structure [1]. Ambulatory home measurements of DBP and SBP use the techniques based on inflated cuffs, but continuous recording is still cumbersome. It would be helpful to find unobtrusive methods for the long-term monitoring of these cardiac indices during the daily activities and sleep.
Arterial stiffness is often assessed by measuring pulse wave velocity (PWV), which is increased in stiffer arteries. The PWV can be estimated by measuring arrivals of pulse waves at two arterial sites:
PWV = distancebetweenthesites traveltimebetweenthesites .
The travel time is commonly referred as pulse transit time (PTT). Arrival of the pulse wave to distal arterial sites can be easily measured by using a photoplethysmogram (PPG), which is an optical non-invasive sensor that can be placed, for example, in a wearable device [5]. On the other hand, in order to predict aortic stiffness reliably, the first arterial site should be located at the beginning of aorta (for measurement of aortic valve opening). However, a measurement of valve opening can require a device such as phonocardiograph, ultrasound or MRI.
To overcome this difficulty, PTT is often approximated using pulse arrival time (PAT) which uses the R- wave of electrocardiogram (ECG) as a reference timing [6]. However, there exists controversy in the clinical accuracy of using PAT in the predictions due to variations in pre-ejection period (PEP) from the R-wave to aortic valve opening [7, 8]. An alternative approach is to approximate the reference with a measurement from another distal site near aorta. For example, the gold standard for aortic PWV measurement is to measure differences of pulse arrivals to carotid and femoral arteries.
The estimation of blood pressure from arrival of pulse waves has also been largely studied; see e.g. [6, 9, 10]. Although promising results have been reported, clinical use of these techniques is still limited. Haemodynamic alterations can have significant effects on the accuracy [11].
A common problem with the clinical use of the above methodologies is that the development and validation of the methods typically require a large set of measurements from real human subjects with sufficient variety. Such data collection can be a very difficult and expensive task.
A preliminary assessment of the methods without extensive data collection can be carried out using simulators. For example, Willemet et al [12, 13] proposed approach to use cardiovascular simulator for generation of a database of “virtual subjects” with sizes limited only by computational resources. In their study, the databases were generated using one-dimensional (1D) model of wave propagation in a artery network comprising of largest human arteries [14]. Such 1D models provide computationally efficient way to simulate blood circulation and are also used in several other applications [15]. There are also studies validating 1D simulations against real measurement [16–18]. The virtual database approach was used to assess accuracy of pulse wave velocity measurements for estimation of aortic stiffness [12] and the accuracy of pulse wave analysis algorithms [13].
The aim of our study is to assess theoretical limitations for the prediction of aortic pulse wave velocity (aPWV), blood pressures (DBP/SBP) and SV from PTT/PAT measurements. We apply a similar virtual database approach to find correlations between these cardiac indices and PTT/PAT timings measured from different locations. In particular, we train Gaussian process regressor to predict the cardiac indices using different combinations of PTT and PAT measurements. The regressor model is trained using a large set of virtual subjects generated using 1D cardiovascular simulator, and the results are validated using another set of virtual subjects. The result of study can give preliminary implications for the accuracy of such predictions in rather ideal circumstances.
Our study is based on the 1D haemodynamic model of entire adult circulations introduced by Mynard and Smolich [19]. It includes heart functions and all larger arteries and veins for both systemic and pulmonary circulation. As heart is included to the model, it can also simulate variations in PEP that are essential in the comparison of PTT and PAT timings.
This paper is organized as follows. Cardiovascular model, generation of virtual subjects and prediction methods are described in Methods and Models section. Results section contains numerical experiments. We will finish with Discussion.
In this section, we begin with a short summary of cardiovascular model and present its numerical discretization.We will also describe the generation of the database of virtual subjects and the computation of Gaussian process predictions.
The blood circulation model is based on the 1D haemodynamic model described in [19], which basically extends commonly used 1D wave dynamics model (see e.g. [14]) with heart functions and realistic arteria and venous networks including pulmonary and coronary circulations. The components of the model are shortly summarized below, see [19] for more details.
The database is created by running the cardiovascular model repeatedly. The model parameters are varied to reflect variations between individual (virtual) subjects.
In [12, 13], the seven parameters were varied: elastic artery PWV, muscular artery PWV, the diameter of elastic arteries, the diameter of muscular arteries, heart rate (HR), SV and peripheral vascular resistance. In their study, the parameters were varied by specifying a few possible values for each parameter and the cardiovascular model was run for all of the resulting 7776 combinations. However, in our study, the cardiovascular model has significantly more model parameters (e.g. parameters related to heart model and valves, vascular beds, …). Such systematic variation of all essential parameters would lead to excessively large number of combinations.
In this study, we choose “sampling” approach in which the model parameters are varied randomly. Our aim is to choose random variations that would represent healthy subject and, where applicable, the range of the parameters is of similar range as in [12]. Some choices can be rather subjective due to the limited amount of (probabilistic) information from related physiological quantities. Our goal is to choose variations to be wide enough so that “real world” can be considered as a subset of the population covered by the variations. However, if more sufficient information about parameters becomes available, it should be rather straightforward to carry out the analysis with the adjusted distributions.
In the following, the superscript (s) refers to a virtual subject for which the parameters are specified. The overbar notation (e.g. L ¯) refers to the values used in [19] (the baseline). Unless otherwise mentioned, the variations are chosen to be normally distributed. Furthermore, the statements such as 10% relative variation should be understood in terms of standard deviations instead explicit ranges of the parameter. We use slightly unconventional notation N ( μ , X % ) to denote the Gaussian distribution with mean μ and the standard deviation σ = X/100μ (i.e. X% variation relative to the mean/baseline). The uniform distribution is denoted as U ( a , b ).
We apply Gaussian process regression for the computation of predictors. GPs are widely used, for example, in machine learning, hydrogeology and analysis of computer experiments (e.g. see [25–27]). GPs also provide flexible predictors that can handle non-linear relationship between input data and the response variable as well as uncertainties in the data. However, we note that any other class of regressions capable of nonlinear relationships can also be used for the analysis. For example, similar results can be achieved with multivariate adaptive regression splines [28].
A GP is a stochastic process f (z) (z ∈ R d) such that f (z1), …, f (zn) is a multivariate Gaussian random variable for all combinations of z1, …, zn. It can be described by the specifying mean function μ (z) = E(f (z)) and the covariance function k (z, z′) = cov(f (z), f (z′)). For more details, see e.g. [25].
Consider a case in which the inputs z are a vector of PTT or PATs and possibly HR and y is the response variable (aPWV, DBP, SBP or SV). We model the response variables as
y ( z ) = h ( z ) T β + f ( z ) + ϵ , (34)
where h (z) is a vector of (deterministic) basis functions, β is a vector of basis function coefficients, f (z) is a GP with zero mean and covariance function k (z, z′), and ϵ is an Gaussian white noise. The first term represents mean behavior of the GP model. The GP term models non-linear relationship between input data and the response variable as well as correlated uncertainties in the data.
Training data comprises of input-output pairs {(zi, yi); i = 1, …, N }. We assume that yi’s are output of the above model i.e. yi = y (z1). Furthermore, let Z ′ = ( z 1 ′ , … , z p ′ ) be inputs for which we want to calculate predictions. Then Y = (y1, …, yN) and Y ′ = ( y ( z 1 ′ ) , … , y ( z p ′ ) ) are both Gaussian and the conditional distribution of Y′ given Y is (see e.g. [25], Appendix A.2]),
p ( Y ′ | Y ) = N ( μ Y ′ + Σ Y ′ Y Σ Y - 1 ( Y - μ Y ) , Σ Y ′ + Σ Y ′ Y Σ Y - 1 Σ Y Y ′ ) (35)
where μY and ΣY denotes the mean and covariance of Y and ΣYY′ is the cross-covariance of Y and Y′. The means and covariances can be calculated by pluggin in the model (34), which gives
μ Y ′ | Y = h ( Z ′ ) T β + k ( Z ′ , Z ) ( k ( Z , Z ) + σ ϵ 2 I ) − 1 ( Y − h ( Z ) T β ) (36) Σ Y ′ | Y = k ( Z ′ , Z ′ ) − k ( Z ′ , Z ) ( k ( Z , Z ) + σ ϵ 2 I ) − 1 k ( Z , Z ′ ) (37)
where h (Z′) and k (Z′, Z) are shorthand notations for the vector and matrix with the components h ( z i ′ ) and k ( z i ′ , z j ), respectively. The above conditional mean gives us an prediction of Y′ with a confidence estimate given by the conditional covariance.
In this study, the covariance function are chosen to be Matern kernel function with ν = 3/2 with a separate length scales for each input parameter. This kernel function can be written as
k(z,z′)=σ2(1+3r)exp(−3r),r=(∑md(zi−zj)2ℓm2)1/2 (38)
where σ2 is the variance and ℓm are the length scales for each input. We note that the choice of the kernel function does not have a large effect to the results as our sample size is large. For example, our experiments show that use of the squared exponential covariance function gives very similar results with differences of the same scale as the prediction uncertainty.
The predictors are computed using fitrgp function in MATLAB Machine Learning Toolbox which provides numerically efficient implementation for the GP regression. The basis functions h(z) are chosen to be linear. The fitrgp function also estimates hyperparameters θ(β,σϵ2,σ2,ℓ1,…,ℓd) by minimizing the negative loglikelihood,
L ( θ ) = - log p ( y | Z , θ ) = 1 2 y T Σ θ - 1 y + 1 2 log det Σ θ + n 2 log 2 π (39)
where Σ θ = k ( Z , Z ; θ ) + σ ϵ 2 I. The optimization is carried out using a subset of observations to avoid high computational load. The parameters of fitrgp related to this hyperparameter optimization are chosen to be the default values.
In this section, we apply GP regression to predict aPWV, DBP, SBP and SV using combinations of different type of PTT/PAT timings and HR as input. We train a GP predictor separately for each considered combination as described above. For validation, we apply the trained predictor to the test set and calculate Pearson correlation between the predictions and ground truth values. Tables A-H in S1 Appendix also report 95% confidence intervals (CI) for the Pearson correlations (BCa bootstrapping intervals [29]). Each table also highlights selected predictions with largest Pearson correlations. However, we note that the order of Pearson correlations should be considered as indicate rather than a definite order of performance due to the uncertainty especially when differences are small.
Fig 6 shows predictions of aPWV for a selected set of combinations when the measurement location is LCA. Table A in S1 Appendix summarizes the results for the complete set of combinations.
The results show that using PTTff or PTTD as a single input gives moderate accuracy and predictions using either HR, PTTp, or DAT are insufficient. Performance can be improved by combining multiple different timings. For example, the accuracy is significantly improved if both PTTff and PTTp are used for predictions (r = 0.90). Furthermore, including also DAT provides the accuracy of r = 0.94, and adding other timings does not significantly improve accuracy any further.
Measurements from RCA provide less accurate predictions (Table B in S1 Appendix): for example, the combination of PTTff, PTTp, PTTD and DAT provides one of highest accuracies for RCA (r = 0.79), but is still only moderate. Such results can be expected as pulse waves travel shorter distance in aorta and also travel through brachiocephalic artery (see Fig 1) inducing additional variations to the (average) wave speeds.
Performance of wrist measurements (LRad / RRad) are even worse (see Table C in S1 Appendix for LRad; results for RRad are similar). For example, the highest accuracy (r = 0.73) can be achieved with the combination of PTTff, PTTp, PTTD and DAT. This is also expected as relative large part of the arterial tree to these measurement locations are comprised of brachial and radialis arteries with their own variations to PWV. On the other hand, measurements from lower limb could provide better performance: for right femoral artery, we can achieve r = 0.75 using PTTff and r = 0.84 using PTTff, PTTp, PTTD and DAT (Table D in S1 Appendix). In this case, pulse travels though the whole aorta to reach these measurement locations.
As mentioned above, in practice, the R-peak location in ECG signal is often used as a surrogate to aortic valve opening due to simpler measurement. However, using PATs gives only mediocre accuracy compared to PTT due to the physiological variations in PEP [7, 8]. Our finding are similar, see for example, Fig A and Table E in S1 Appendix for LCA. The highest accuracy is r = 0.79 (e.g. PATff, PATp, PATD and HR) which is significantly worse compared to using PTTs.
Another approach to avoid measurement of aortic valve opening is to consider differences of pulse arrival times to two distal locations. Such setup also allows us to avoid the influence of PEP variations. Results for measurement between LCA and Fem can be seen in Fig B and Table F in S1 Appendix: difference of PTTff gives r = 0.76 which is slightly better than using normal PTTff measurement from Fem, but not as good as normal PTTff measurement from LCA. The highest accuracy (r = 0.87) can be obtained, for example, with PTTff, PTTp, PTTD and HR. The predictions of PWV that use the difference between LCA and RCA or the difference between LRad and RRad are less accurate (r ≈ 0.75 − 0.78 at best); see Tables G and H in S1 Appendix.
Figs 7 and 8 show predictions for DBP and SBP for selected PTT time combinations when measurements are taken from LCA; see also Table A in S1 Appendix for all combinations. For DBP, predictions using PTTff as a single input achieves very low accuracy (r = 0.33). Significantly more accurate predictions can be achieved using HR (r = 0.85) or DAT (r = 0.86). For SBP, the performance of PTT based predictions is better but still quite low (r = 0.58 for PTTff and r = 0.60 for PTTp). Predictions can be improved by adding additional input timings. For DBP, combining PTTff with HR or DAT gives r = 0.92 and the highest accuracy r = 0.94 is obtained with PTTff, PTTp, PTTD and DAT. Additional input timings also improves performance of SDB predictions: PTTff and HR/DAT results in r = 0.735 and the highest accuracy is r = 0.75 (PTTff, PTTp, PTTD and DAT). Findings the other measurements locations are similar; see Tables B, C and D in S1 Appendix.
We also consider predictions from pulse arrival times (i.e. using R-peak as a reference timing). Compared to PTT times, the results are of mixed accuracy; see Table E for PAT measurements from LCA. For DBP, using PATff as single input yields insufficient predictions (r = 0.19), but PATp gives moderate accuracy (r = 0.67). Combinations of different PAT timings can even achieve higher accuracy than using PTTs: for example, r = 0.95 with PATff and DAT and r = 0.96 for PATff, PATp, PATD and DAT. For SBP, PATff provides slightly better accuracy compared to PTTff(r = 0.62), but otherwise results are similar.
As with aPWV, we consider differences of pulse transit/arrival times measured with two sensor. Measuring between LCA and Fem gives very similar performance to PTT measurements from LCA (Table F in S1 Appendix). However, other considered setups provide less accurate results: see Table G in S1 Appendix for differences between LCA and RCA measurements and Table H for differences between measurements from radialis arteries.
Results show that HR has largest contribution to the predictions of SV, meanwhile performance with pulse transit or arrival timings (without HR information) can only provide moderate accuracy at best. For example, Fig G and Table A in S1 Appendix show the predictions using measurements from LCA. Predictions with HR as a single input reaches r = 0.81, but predictions using PTTff or PTTD are insufficient estimates (r < 0.25) and predictions with PTTp are of moderate accuracy (r = 0.60). SV can be predicted with good accuracy with DAT, but this is due to the strong correlation between HR and DAT as mentioned above. Furthermore, significant improvements will not be achieved by combining several inputs. For example, highest accuracy is r = 0.83 which can be obtained, for example, with PTTff, PTTD and HR. Results are similar for all other measurement setups; see Tables B-H in S1 Appendix.
This paper assessed theoretical limitations for the prediction of aortic pulse wave velocity (aPWV), DBP/SBP and SV from pulse transit and arrival time measurements. We applied a virtual database approach proposed by Willemet et al [12, 13] in which a cardiovascular simulator is used to generate a database of virtual subjects. In this work, we applied one-dimensional haemodynamic model by Mynard and Smolich [19] to construct a simulator for entire adult circulation. This simulator was used to generate a large database of synthetic blood circulations with varied physiological model parameters. The generated database was then used as training data for Gaussian process regressors. Finally, these trained regressors were applied to another synthetic database (test set) to assess capability of regressors to predict aPWV, SDB, DBP and SV using different combinations pulse transit/arrival time and HR measurements.
The results indicate that aPWV and DBP can be estimated from PPG signal with a high accuracy (Pearson correlation r > 0.9 between true and predicted values for measurement from left carotid artery) when, in addition to foot-to-foot PTT time, information about the peak and dicrotic notch location is also given as input to the predictor. The predictions of SDB were less accurate (r = 0.75 at best). For SV, accurate predictions were mostly based on heart rate, with only a very minor improvement in accuracy when PTT timings were also included as inputs.
As this was entirely in silico study, it is not guaranteed that the result can be applicable to the real world as is. However, the aim of the study was to give preliminary results about correlations between the cardiac indices and PTT/PAT timings and the applicability of such predictions. The hope is that the results could to be extended to real clinical applications in future research.
The limitations to be addressed in future are the following. First, the cardiovascular model has its limitations. Although previous studies have shown that similar cardiovascular models can be used to simulate human physiology relatively well [16–18], not all physiological phenomena are fully covered in the Mynard’s model. One example of such phenomenon is respiration. The effect of respiration can be important as the breathing and cardiac cycles are in a close interaction. Several physiological factors, such as the changes in the intrathoracic pressure and the variation in the interbeat intervals modulate the cardiac mechanics and blood outflow from the heart. Even the timing of the shorter cardiac cycles coupled with the longer respiratory cycles has effects on the central circulation. When we considering a healthy heart, the effects of respiration can perhaps be managed by interpreting different virtual subjects to represent inspiratory and expiratory phases of the breathing. Other phenomena that are not covered by the model are, for example, gravity and baroreceptors. Furthermore, lumped parameter models that are used for heart and vascular beds were relatively simple approximations. However, new analytical methods allow us to bridge the models and human bodily functions [30].
The chosen baselines and variations of the model parameters were chosen to represent healthy subject. The choices, however, can be subjective due to the limited amount of (probabilistic) information. Our attempt were to produce variations such that the virtual population covered by the chosen parameter variations includes real physiological variations. We, however, emphasize that the presented approach is not limited to the chosen parameters variations and it can be adjusted if more precise information becomes available.
Due to the limited phenomena covered by the model, the results may not be reliable when considering subjects with medical conditions. For example, the simplified heart model and variations of related model parameter may not present subjects with heart diseases.
In this study, we only considered pulse transit and arrival type of time information as the input to the predictor. Predictions could potentially be improved with other kinds of additional information. For example, aortic PWV predictions could be improved by using information about the distances between aorta and/or measurement points. Information about arterial path lengths could have been easily used in our simulation analysis, but in practice such information would require clinical measurements such as MRI [21, 22]. On the other hand, the arterial path length are often estimated using the body lengths or measuring distances of certain points in the body [21, 22]. Such information was not used in this simulation study as precise statistical knowledge of connection between such body measurement and arterial length was not available. Instead, Gaussian process regressors implicitly marginalize predictions over different arterial lengths that are present in the virtual database.
Ultimately it would be beneficial to develop approaches that do not need reference measurement (aortic valve opening/R-peak). For example, Choudhury et al [31] presented a machine learning algorithm which uses raise times and pulse widths derived from PPG signal to predict DBP and SBP. Furthermore, deep learning approaches could perhaps be used to infer optimal information from PPG waveform. These are subject of our future research.
|
10.1371/journal.pbio.1001142 | A Blind Circadian Clock in Cavefish Reveals that Opsins Mediate Peripheral Clock Photoreception | The circadian clock is synchronized with the day-night cycle primarily by light. Fish represent fascinating models for deciphering the light input pathway to the vertebrate clock since fish cell clocks are regulated by direct light exposure. Here we have performed a comparative, functional analysis of the circadian clock involving the zebrafish that is normally exposed to the day-night cycle and a cavefish species that has evolved in perpetual darkness. Our results reveal that the cavefish retains a food-entrainable clock that oscillates with an infradian period. Importantly, however, this clock is not regulated by light. This comparative study pinpoints the two extra-retinal photoreceptors Melanopsin (Opn4m2) and TMT-opsin as essential upstream elements of the peripheral clock light input pathway.
| The circadian clock is a physiological timing mechanism that allows organisms to anticipate and adapt to the day-night cycle. Since it ticks with a period that is not precisely 24 h, it is vital that it is reset on a daily basis by signals such as light to ensure that it remains synchronized with the day-night cycle. The molecular mechanisms whereby light regulates the clock remain incompletely understood. Here we have studied a cavefish that has evolved for millions of years in the perpetual darkness of subterranean caves in Somalia. Like many other cave animals, these fish display striking adaptations to their extreme environment, including complete eye degeneration. We show that despite evolving in a constant environment, this blind cavefish still retains a circadian clock. However, this clock ticks with an extremely long period (nearly 47 h), and importantly it does not respond to light. We reveal that eye loss does not account for this “blind” clock. Specifically, mutations of two widely expressed non-visual opsin photoreceptors (Melanopsin and TMT opsin) are responsible for the blind clock phenotype in the cavefish. Our work illustrates the great utility of cavefish for studying the evolution and regulation of the circadian clock.
| The circadian clock is a highly conserved, physiological timing mechanism that allows organisms to anticipate and adapt to daily environmental changes and it is synchronized primarily by light. In mammals, intrinsically photosensitive retinal ganglion cells serve as the principal circadian photoreceptors [1]. In non-mammalian vertebrates, photoreceptors located outside of the retina (in the pineal complex and in the deep brain) have also been implicated in the regulation of the circadian timing system [2]. At the core of the vertebrate circadian clock is a transcription translation feedback loop mechanism composed of activator and repressor clock proteins [3]. Light-induced expression of certain clock genes represents a key step in the relay of lighting information to the core clock machinery [4],[5].
The zebrafish (Danio rerio) represents a fascinating model to study the mechanisms whereby light regulates the clock. The “peripheral” clocks in most zebrafish tissues and even cell lines are entrained by direct exposure to light [6]. However, fundamental questions concerning the identity of the widely expressed photoreceptor molecules and how they signal to peripheral clocks remain unanswered. To date, a set of widely expressed opsins, one cryptochrome homolog, and flavin-containing oxidases have all been implicated as candidate peripheral photoreceptors [7]–[9].
In certain extreme environments such as caves some fish species have remained completely isolated from the day-night cycle for millions of years [10]. They show convergent evolution, sharing a range of striking physical “troglomorphic” properties including notably degeneration of the eyes during early development. However, many aspects of cavefish biology still remain incompletely understood. Does evolution in constant darkness lead to loss of other aspects of photoreceptor function including regulation of peripheral circadian clocks by light? Furthermore, do these remarkable animals even retain normal circadian clocks?
In this report we explore the circadian clock and its regulation by light in Phreatichthys andruzzii, a Somalian cavefish that shows an extreme troglomorphic phenotype. We compare the circadian clock mechanism of this cavefish with that of the zebrafish with the goal of identifying key components of the light input pathway. We reveal that P. andruzzii possesses a clock that is entrained by periodic food availability, displays a long infradian period, and lacks temperature compensation. However, importantly, this cavefish clock is no longer entrainable by light. Strikingly, in the cavefish we encounter mutations in the candidate non-visual photoreceptors Melanopsin (Opn4m2) and TMT (teleost multiple tissue)-opsin and we provide direct evidence for a light-sensing function of these non-visual opsins in the regulation of vertebrate peripheral clocks.
We chose to study a species of cavefish expressing an extreme “troglomorphic” phenotype, P. andruzzii (Figure 1B). This Somalian cavefish evolved from surface dwelling ancestors, isolated in a totally dark environment beneath the desert at a nearly constant temperature for 1.4–2.6 million years [10], approximately 1 million years longer than the well-studied cavefish Astyanax mexicanus [11]. P. andruzzii shows total eye degeneration, no scales, and complete depigmentation [12].
As a first step we wished to explore the circadian clock and its regulation by light in this cavefish species. We initially measured locomotor activity in cavefish exposed to a 12∶12 light-dark (LD) cycle (Figures 1D,F and S1B). We documented striking arrhythmic locomotor activity that contrasts with the clear rhythmic, diurnal pattern observed for zebrafish under the same conditions (Figures 1A,C,E and S1A). In order to explore at the molecular level this lack of behavioural rhythmicity in cavefish, we next subcloned a set of P. andruzzii clock gene homologs with the aim of examining their expression pattern under LD cycles (Table S1). These genes were selected since their zebrafish counterparts are either clock- or light-regulated. Sequence analysis confirmed a close similarity between zebrafish and P. andruzzii genes (Table S2 and Figure S2), consistent with both species belonging to the Cyprinidae family. We measured the expression of a subset of clock-regulated (Clk1a, Clk2, Per1, Per1b) and light-regulated genes (Per2, Cry1a, and Cry5) [4],[5],[8],[13],[14] in vivo in adult tissues and in whole larvae from both species (Figure 1G–J and Figure S3). In the zebrafish, in agreement with previous reports, exposure to a LD cycle results in robust rhythmic expression of these genes (Figure 1G,I and Figure S3A,B,E) [6],[14]. Remarkably, arrhythmic gene expression was encountered in cavefish tissues and larvae (Figure 1H,J and Figure S3C,D,F,G), even during the first day of life when eye rudiments still persist [12]. Importantly, these clock gene expression patterns are consistent with the behavioural activity profiles observed under LD cycles for both species. Circadian clocks are encountered in most vertebrate cell types and even in cell cultures in vitro. Furthermore, the persistence of circadian clocks in vitro has enabled many more detailed mechanistic studies. Therefore, we established cell lines from adult cavefish caudal fins (CF cells) to test whether P. andruzzii cells in vitro also lack rhythmic gene expression under LD conditions. Again we observed arrhythmic clock gene expression in P. andruzzii cells (Figure 2C,D) that contrasts with the robust rhythms documented in zebrafish cells (Figure 2A,B) [6],[15],[16]. Thus, we failed to detect rhythmic clock gene expression in vivo or in vitro in P. andruzzii under LD cycles. So our data reveal that either P. andruzzii lacks the circadian clock itself or it has a clock lacking a functional light input pathway.
To distinguish between these two hypotheses, we first assessed whether the cavefish circadian clock could be entrained by an alternative environmental time signal (zeitgeber), periodic food availability. Adult cavefish and zebrafish were fed once at the same time each day for one month under constant dark conditions, and during this period, locomotor activity was measured (Figure 3). For both species we observed food anticipatory activity (FAA, [17]), a characteristic increase in locomotor activity encountered a few hours prior to mealtime (Figure 3A,B,D,E) and a strong entrainment of rhythmic locomotor activity (Figure 3C,F). FAA is indicative of regulation by a food entrainable oscillator (FEO, [18]); thus we tested circadian clock gene expression in various tissues of both species during the final day of food entrainment and then two days of fasting under constant conditions (Figure 4A). Consistent with previous results [19], rhythmic clock gene expression (Clk1a and Per1b) was observed in zebrafish brain, heart, fin, and liver (Figures 4B,D,F and S4A), with the only exception of Clk1a in the zebrafish brain (Figure 4B, red trace). However, in all cavefish tissues including the brain, robust circadian rhythms of both Clk1a and Per1 expression were observed (Figures 4C,E,G and S4B). In both species, differences in the phase and amplitude of rhythmic expression for each gene were observed between different tissues. Importantly, these results point to P. andruzzii having a functional clock that is entrainable by feeding but not by LD cycles. This contrasts with the situation in zebrafish where both light- and food-entrainable oscillators are present.
We also tested the effect of alternative zeitgebers on clock gene expression in cavefish CF cells. Transient treatments with glucocorticoids are widely used to induce rhythmic gene expression in cultured cells [20]. CF cells transfected with a clock-regulated zebrafish reporter construct (zfPer1b-Luc) [13] were treated transiently with 100 nM dexamethasone, an agonist of the glucocorticoid receptor (Figure S5; Figure 5B, green trace) [20]. Surprisingly, this induced a bioluminescence rhythm in cavefish cells that persisted for almost three cycles with an extremely long period (τ = 43 h at 25°C). This contrasts with the circadian bioluminescence rhythm observed upon dexamethasone treatment of zebrafish cells (τ = 24.2 h at 25°C) (Figure 5A, green trace). These results reveal the existence of an abnormal circadian clock in P. andruzzii that displays an infradian period under constant conditions. We next wished to determine whether other circadian clock properties are abnormal in the cavefish. One highly conserved feature of the circadian clock is that its period remains relatively constant over a physiological range of temperatures (so-called temperature compensation) [21]. Thus, we measured the period of rhythms induced by dexamethasone pulses in CF and zebrafish cells held at two additional constant temperatures 22°C and 29°C (Figure 5). With an increase in temperature the period length of the cavefish clock decreased significantly (τ = 47 (22°C), 43 (25°C), and 38 h (29°C)) revealing reduced temperature compensation with Q10≈1.35, while in zebrafish cells, as expected, a relatively constant period was observed (τ = 23.6 (22°C), 24.2 (25°C), and 24.6 (29°C)) h, respectively (Q10≈1 [22]). Thus, together our results indicate that P. andruzzii possesses a circadian timing system with an aberrant core clock mechanism.
Could the discrepancy between the striking infradian period of the P. andruzzii clock (τ>30 h) and the period of the LD cycle (T = 24 h) have been the origin of the observed arrhythmicity under LD cycles? In such a scenario, the cavefish clock might still be entrainable by light. In zebrafish cell lines, exposure to brief light pulses results in advances or delays in pre-existing clock gene expression rhythms depending on the precise time when the cells are illuminated [13]. Thus, we wished to test whether light is able to regulate the cavefish clock in a similar manner. We synchronized rhythmic clock gene expression in CF cells by dexamethasone treatment and then exposed the cells to a 15 min light pulse at five different time-points distributed throughout the 43 h cycle (Figure S6) [13],[23]. None of the light pulses changed either the phase or levels of rhythmic clock gene expression. Hence, we conclude that P. andruzzii indeed possesses a truly blind clock. Thus, this cavefish represents a powerful complementary model for exploring the function of the light input pathway in vertebrates.
Our systematic cloning and sequencing strategy failed to detect any mutations significantly affecting clock gene coding sequences (unpublished data). Light-induced transcription of clock genes represents a key step in photic entrainment of the zebrafish clock [4],[5]. Therefore, we speculated that mutations in promoter sequences of light-inducible clock genes could account for the cavefish blind clock phenotype. We have previously defined an essential light responsive module (LRM) containing E- and D-box enhancer elements in the zebrafish per2 promoter [5]. In the LRM promoter region from the cavefish per2 gene, we identified several base substitutions compared with the zebrafish sequence, including one in the E-box element (Figure 6A). We wished to test whether the cavefish LRM was still able to direct light-induced gene expression. With this aim we transfected zebrafish cells with a cavefish per2 promoter driving luciferase expression (cfPer2-Luc) (Figure 6B, green trace). These cells exhibited robust light-driven reporter expression comparable to that of the zebrafish per2 promoter (zfPer2-Luc) (Figure 6B, blue trace). Conversely, zebrafish zfPer2-Luc transfected into cavefish cells failed to show light-induced expression (Figure 6C, green trace). Together, these results indicate that mutations disrupting the cavefish light input pathway should lie upstream of directly light-regulated clock gene promoters.
Mutations affecting peripheral photoreceptors could also account for the blind cavefish clock. Although the identity of the teleost peripheral photoreceptor remains unclear, candidates include the opsins Melanopsin (Opn4m2) and TMT-opsin that are widely expressed in most tissues [9],[24]. Melanopsin was originally isolated from the photosensitive melanophores of Xenopus [25]. Subsequently, orthologs of Melanopsin were isolated from other non-mammalian vertebrates including zebrafish [26]. TMT-opsin was originally identified by virtue of its opsin sequence homology, and to date has only been isolated from teleosts [9]. We chose to clone and characterize these two opsins in the cavefish. We documented mRNA expression of both Melanopsin and TMT-opsin in various cavefish tissues including the CF cell line (Figure S7), consistent with previous results that revealed widespread expression for the zebrafish homologs [9],[24]. Furthermore, sequence alignment with other teleost homologs revealed strong conservation (unpublished data). Interestingly, however, premature stop-codons were encountered in the coding sequences of both TMT-opsin and Melanopsin (at the C-terminus of the 5th transmembrane domain and the N-terminus of the 6th transmembrane domain, respectively) (Figure 7A). These C-terminal truncations were the result of frame shift mutations in the two coding sequences (an insertion of one T in TMT-opsin at position +654 and deletion of one G in Melanopsin at position +842 relative to the ATG initiation codons of the zebrafish homolog sequences). Both opsin mutations would be predicted to eliminate binding of the essential chromophore, retinaldehyde, that normally occurs at the 7th transmembrane domain. Therefore, we decided to test whether ectopic expression of the zebrafish homologs of these two opsins would rescue light-inducibility of a zfPer2-Luc reporter in cavefish cells. Simply supplementing the culture medium with retinaldehyde failed to induce rhythmic expression of zfPer2-Luc (Figure 7B). Strikingly, upon cotransfection with single opsin expression vectors, zfPer2-Luc was robustly induced during the light phase and subsequently decreased during the dark phase (Figure 7C, blue trace and 7D, red trace). In contrast, expression in cavefish cells of zebrafish Melanopsin and TMT-opsin carrying mutations introducing premature stop codons equivalent to the two cavefish opsins (zfOpn4m2K286X and zfTMTY224X) failed to rescue light inducible zfper2-Luc expression (Figure 7C and D, grey traces). These data are consistent with the predicted loss of the chromophore binding domains in the cavefish opsins. Furthermore, our results constitute strong evidence that Melanopsin and TMT-opsin indeed function as peripheral clock photoreceptors.
While Melanopsin represents a well-characterized photoreceptor [25],[26], until now TMT-opsin has been implicated as a photopigment based only upon sequence homology [9]. The cavefish represents a powerful model to confirm that TMT-opsin indeed functions as a photopigment. In all opsins a highly conserved lysine residue within the 7th transmembrane domain provides a Schiff base linkage with the chromophore retinal that is critical for phototransduction [27]. In zebrafish TMT-opsin, a lysine residue at position 295 would be predicted to fulfil this role [9]. To test this prediction, we mutated lysine 295 to an alanine (TMTK295A). Expression of TMTK295A failed to rescue light-induced zfper2-Luc expression in cavefish cells (Figure 7D, green trace), thus strongly indicating that TMT-opsin indeed functions as an opsin photopigment.
It is well established that the response of opsins to light is wavelength dependent. Melanopsin has been shown to respond preferentially to the blue region of the light spectrum [25], while the wavelength sensitivity of TMT-opsin is unknown. Are Melanopsin and TMT-opsin alone sufficient to account for the wavelength sensing properties of zebrafish peripheral clocks? As a first step towards addressing this key question, we repeated our cavefish zfper2 expression rescue assay under three different monochromatic light sources (blue, green, and red, Figure S8). We then compared these results with the response of the zfper2 promoter to the same light sources in zebrafish cells. Exposure of Melanopsin or TMT-opsin transfected cavefish cells to blue (468 nm) or green (530 nm) light is able to activate the zfper2 promoter (Figure 8A–D). In contrast, no rescue was observed under red (657 nm) light (Figure 8E). However, exposure of zebrafish cells to these same monochromatic light sources revealed activation by all three light sources, with the strongest induction by blue (Figure 8F). These results confirm the wavelength sensitivity of Melanopsin [28] and provide the first evidence, to our knowledge, that TMT-opsin can respond to blue and green but not red wavelengths of light. The differences in the red light response of the zfper2 promoter between the zebrafish cells and the rescued cavefish cell system point to the existence of additional peripheral photoreceptors. Importantly, comparable results were obtained in all our rescue experiments when we substituted zfper2-Luc for the minimal light-responsive promoter fragment of the zebrafish per2 gene −0.43per2:Luc (unpublished data) [5]. This predicts that light-driven gene expression mediated by Melanopsin and TMT-opsin is dependent upon the LRM of the per2 promoter.
In summary, during 1.4–2.6 million years of isolation from the day-night cycle, the evolution of the cavefish P. andruzzii has lead to an aberrant circadian clock. Contrary to the situation in most organisms, this clock is no longer entrained by light. Furthermore, upon exposure to alternative zeitgebers it cycles with a remarkably long infradian period and shows reduced temperature compensation. It is possible that this reflects progressive loss of a mechanism that provides no selective advantage for animals that live under constant darkness and temperature. In support of this hypothesis, one recent report has documented a severe attenuation of circadian clock rhythmicity in an arctic mammal naturally exposed to the extreme polar photic environment, the reindeer, Rangifer tarandus [29]. To further test this hypothesis, it will be fascinating to compare the circadian clock phenotype of P. andruzzii with that of other cavefish species representing the full range of troglomorphic phenotypes.
Our study reveals that a regular daily feeding time does entrain the cavefish clock. Interestingly, a comparison of the clock gene expression rhythms encountered in various tissues of the regularly fed cavefish reveals many differences in phase and amplitude between tissues. This situation is strongly reminiscent of the different patterns of cycling gene expression observed in individual light entrained peripheral clocks of zebrafish [30] and may reflect differences in the molecular regulatory networks associated with each tissue. Several reports point to the general importance of feeding entrainment for the circadian timing system in fish [17],[19]. It is tempting to speculate that food availability in the subterranean environment of this cavefish might indeed be periodic, and therefore a clock responding to and anticipating feeding time may confer a survival advantage. Interestingly, several lines of evidence point to the existence of a food-entrainable oscillator in vertebrates distinct from the light-entrainable oscillator, however the anatomical location and molecular mechanism remains unclear [18],[19],[31],[32]. In this regard, fish could emerge as powerful models for the investigation of food entrainment. A comparative study involving zebrafish that possess both light and food entrainable oscillators and P. andruzzii that retains only the food entrainable oscillator could provide important insight into the basis of the feeding entrainment mechanism in vertebrates.
By employing a comparative functional analysis involving zebrafish and P. andruzzii, we have been able to provide direct evidence that TMT-opsin and Melanopsin serve as peripheral tissue photoreceptors in teleosts. Furthermore, our results strongly indicate that the two opsins activate gene expression via the LRM promoter element that mediates light-driven per2 expression [5]. Previously, the identity of the widely expressed photoreceptors that mediate the direct entrainment of peripheral clocks was unclear. In addition to opsins, other candidates include Cry4, a member of the cryptochrome family in zebrafish, and flavin-containing oxidases [7],[8]. It will now be important to re-evaluate the relative contribution of these other candidate photoreceptors to peripheral clock entrainment. Based on our study of the effects of different wavelengths of light on the induction of clock gene expression we predict that Melanopsin and TMT-opsin are not the only photoreceptors for peripheral clocks. This finding begs the question as to why several different photoreceptors might contribute to the photic entrainment properties of peripheral tissues. Changes in the photoperiod, intensity, as well as the spectral composition of sunlight represent reliable indicators of day-night as well as seasonal changes in the environment [33]. The presence of multiple photoreceptors, each one differentially extracting timing information from sunlight, could enable the circadian system to more reliably indicate the timing of dawn and dusk.
Finally from a broader perspective, in addition to displaying a unique and fascinating collection of adaptations to its extreme environment, we have demonstrated that P. andruzzii serves as a powerful complementary model to dissect the molecular pathways that respond to light.
The animal handling procedures and research protocols were approved by the University of Ferrara (Italy), University of Firenze (Italy), University of Murcia (Spain), and Karlsruhe Institute of Technology (Germany) Institutional Animal Care and Use Committees.
P. andruzzii originally collected from Somalia (Figure S9) were maintained and bred at the University of Firenze, Italy. The cavefish were kept in darkness at a constant 27°C except during food administration and aquaria maintenance. Three times per week the fish were fed with frozen chironomid larvae. Fertilized eggs were collected every 30 min and aliquots of 10–20 eggs were transferred into 75 cm2 tissue culture flasks (BD GmbH). Flasks were sealed and submerged in a large volume, thermostatically controlled water bath (Tetra). From the third/fourth day after hatching, larvae were fed once a day.
Cavefish (CF) cell lines were derived from fin clips of adult fish and maintained using standard methods described elsewhere [34]. Cells were transiently transfected using FuGene HD reagent according to the manufacturer's recommendations (Roche) in the absence of serum. Rhythmic clock gene expression originally established by serum treatment during the seeding of the CF cells was resynchronized by a 30 min treatment with a range of dexamethasone (Sigma) concentrations from 50 nM to 1 µM (see Figure S5) [20]. Subsequently the dexamethasone-containing medium was replaced by fresh medium again lacking serum and containing luciferin.
To test for photic entrainment of rhythmic clock gene expression, cavefish and zebrafish adults and larvae as well as cell lines were maintained at 27°C under a 12∶12 LD cycle with a light intensity of 350 µW/cm2 (full-spectrum cool fluorescent tubes, Osram GmbH). For behavioural analysis, zebrafish and cavefish were maintained under full-spectrum cool fluorescent tube light sources with a light intensity of 20 µW/cm2. For monochromatic light sources, light-emitting diodes (LED, Kopa) sources were used (blue: λpeak = 468 nm, green: λpeak = 530 nm, red: λpeak = 657 nm; white: 450 nm<λ<700 nm) (Figure S8). The light intensity of each LED light source was adjusted to ensure an equivalent number of photons were emitted from each source (1.42×1018±0.04×1018 photons/s/m2).
All experiments using adult and larval zebrafish as well as the zebrafish cell lines AB9 [35], PAC-2, and a stable PAC-2 cell line expressing −1.7per2:Luc were performed using standard methods described elsewhere [5],[13]. Induction of rhythmic clock gene expression in zebrafish cells using transient dexamethasone treatment was performed as described for the cavefish CF cell line (see also Figure S5).
Cavefish and zebrafish locomotor activity was registered continuously by means of an infrared photocell (E3S-AD62, Omron) placed at the aquarium wall, in the corner where food was provided. The photocell was placed 5 cm from the water surface and 20 cm from the bottom. The number of light-beam interruptions was counted and stored every 10 min by a computer connected to the photocell. The analysis of locomotor activity records, representation of actograms, and calculations of mean waveforms and χ2 periodograms (Sokolove-Bushell test) were performed using the chronobiology software El Temps (version 1.228).
To obtain partial cDNA sequences, single-stranded cDNA was synthesized using SuperScript III Reverse Transcriptase (Invitrogen). Cavefish genes were amplified by PCR using Taq DNA Polymerase (Invitrogen) with primers designed by Primer3 software on the basis of sequence of the zebrafish homologs (Table S1). Bands of the predicted sizes were cloned into the pGEM-T Easy Vector (Promega). The cavefish gene cDNA fragments were sequenced (QIAGEN GmbH) and compared with the GenBank database by using the BLAST algorithm. Additional cDNA sequences were subsequently cloned using a 5′-3′SMART RACE cDNA amplification kit (BD Bioscience), and then coding sequences were deposited in GenBank (Table S2). By this approach, we cloned 13 clock genes (Per1, Per2, Per3, Cry1a, Cry1b, Cry2a, Cry2b, Cry3, Cry4, Cry5 Clk1a, Clk1b, Clk2) and 2 opsins (Opn4m2, TMT-opsin) from P. andruzzii.
Sequences from the cavefish PER and CLK protein families were aligned with homologs from other teleost species (Takifugu rubripes, Tetraodon nigroviridis, Danio rerio, Gasterosteus aculeatus, and Oryzias latipes) [36],[37] using ClustalW. Alignments were manually verified and phylogenetic trees were generated using Neighbour-joining methods [38] with a complete deletion mode. Bootstrap tests were performed with 1,000 replications. Poisson correction distance was adopted and rates among sites were set as uniform. Drosophila melanogaster PER and CLK sequences were used as an out-group to root the trees.
Single-stranded cDNA was synthesized using SuperScript III Reverse Transcriptase (Invitrogen). Quantitative PCR was performed for P. andruzzii and zebrafish clock genes using the pairs of primers shown in Table S1. The StepOnePlus Real-Time PCR System (Applied Biosystems) was employed using SYBR-green-primer-master mix according to the manufacturer's recommendations with the following cycle conditions: 15 min at 95°C, then 40 cycles of 15 s at 95°C, and 30 s at 60°C. The relative levels for each RNA were calculated by the 2−ΔΔCT method. Relative expression levels were normalized to β-actin. Each CT value is the mean of three biological replicates and each assay was performed a minimum of three times.
zfPer1b-Luc contains a promoter region extending 3.3 kb upstream of the 5′ end of the Period1b cDNA (equivalent to the zfperiod4 promoter construct in [13]). The zfPer2-Luc reporter has been described previously as −1.7per2:Luc that contains a fragment of 1,571 bp upstream of the transcription start site and 129 bp of the 5′UTR of the zebrafish Period2 gene [5]. In addition, the minimal light responsive per2 promoter reporter construct −0.43per2:Luc was also tested in the cavefish cell rescue experiments. This construct contains 431 bp upstream of the transcription start site and 164 bp of the 5′UTR of the zebrafish Period2 gene [5]. Both zfPer2-Luc reporters responded in an equivalent manner. cfPer2-Luc contains 876 bp upstream of the transcription start site and 112 bp of the 5′UTR of the cavefish Period2 gene. All in vivo bioluminescence assays were performed as described previously [13],[34].
The full-length zebrafish TMT-opsin (ENSDART00000081729) was amplified with the primers Fwd 5′-AATGGATTGCGGATTGGATCCATTGTGTCCAACTTG-3′ and Rev 5′-CTGCAGAATTCACTAGTGATTTCGCCTGTA-3′ resulting in the mutation of the ATG translation initiation codon sequence into TTC and the creation of BamHI and EcoRI restriction sites, respectively, for cloning into a modified pcDNA3.1(+) expression vector (Invitrogen) that incorporates an HA-Tag into the N-terminus of the expressed protein [39]. Similarly, the full length zebrafish Opn4m2 (ENSDART00000018501) was amplified with the Fwd 5′-GCTCGGATCCGCCTTGAGCCATCACTCTTCA-3′ and Rev 5′-GCCCTCTAGACTCTTAGTTCCCTCCAAGCAA-3′ primers, thus mutating the ATG initiation codon into TTG as well as creating BamHI and XbaI sites, respectively, for cloning into the pcDNA3.1(+)HA-tag modified expression vector. PCR reactions were performed using the Perkin Elmer Gene Amp XL PCR kit according to the manufacturer's instructions. Construction of the expression vectors for zfTMTK295A and the truncated forms zfOpn4m2K286X and zfTMTY224X involved site-directed mutagenesis using the QuikChange MultiSite-Directed Mutagenesis Kit (Stratagene) according to the manufacturer's instructions. For the truncated forms, in the case of zfTMT-opsin the codon at position 224 (TAT, tyrosine) was mutated to a stop codon (TAA) while for Melanopsin (zfOpn4m2), the codon at position 286 (AAA, lysine) was mutated to a Stop codon (TAA). For the mutation of a key lysine in the 7th transmembrane domain of zfTMT-opsin to an alanine residue, the codon sequence AAG was mutated to GCG. Following sequence analysis, the integrity and function of all opsin expression vectors was tested by transient transfection into a mammalian cell line and then Western blotting analysis using an anti-HA-tag specific antibody.
Hepa1-6 cells (1×105) (ATCC) were transfected with 2 µg of each HA-Tag opsin expression construct using the standard Promo-Fectin transfection protocol (PromoKine). Twenty-four hours after transfection, protein extracts were prepared according to a standard method [40], with some modifications. Cells were solubilized and incubated at 4°C in a mixture of equal volumes of TSA buffer (2 mM Tris-HCl, pH 8.0, 140 mM NaCl, 0.025% NaN3) and Lysis buffer (TSA buffer plus 2% Triton X-100, 5 mM iodoacetamide, 0.2 U/ml aprotinin, and 1 mM phenylmethylsulfonyl fluoride). After 1 h, 0.2 volumes of 5% sodium deoxycholate were added, and the mixture was incubated on ice for 10 min. The lysate was centrifuged at 2,800 g for 10 min at 4°C, and the supernatant was collected and stored at −80°C until use. Before electrophoresis in 10% or 15% SDS-polyacrylamide gels, the proteins were diluted in Laemmli buffer (final concentration 0.05 M Tris, pH 6.8, 2% SDS, 100 mM DTT, 10% glycerol, 0.05% bromophenol blue) without heating. Following transfer to nitrocellulose, the membranes were incubated with a high affinity anti-HA rat monoclonal antibody (clone 3F10, Roche) according to the manufacturer's instructions and the bound antibody was visualized using the ECL detection system (Amersham Biosciences) (Figure S10).
All the results were expressed as means ± SEM. Data were analyzed by one- or two-way analysis of variance (ANOVA) to determine significant differences using the software GraphPad Prism 4.0 (GraphPad Software Inc.). p values<0.05 were considered statistically significant. To evaluate the period length of gene expression, we measured the time span between two consecutive peaks. A trigonometric statistical model was applied to evaluate periodic phenomena. The single cosinor procedure [41] was used to define the main rhythmic parameters (circadian period and peak time).
Bioluminescence data were analyzed using Microsoft Excel or CHRONO software [13],[42]. Period estimates measured after 2 d in DD were made by linear regression following peak finder analysis with CHRONO. For Q10 temperature coefficient calculations, period length estimates for cells held at 22°C, 25°C, and 29°C were calculated as cycles per hour and then plotted against temperature. Linear regression analysis revealed a good fit to a straight line (cavefish R2 = 0.99; zebrafish R2 = 0.98). Mean period lengths at 22°C and 29°C were then substituted into the equation Q10 = (R2/R1)10/(T2−T1), where R is rate and T is temperature.
The sequences reported in this article are deposited in GenBank under accession numbers GQ404475–GQ404490 (see Table S2).
|
10.1371/journal.pntd.0005275 | Comparison of Stable and Transient Wolbachia Infection Models in Aedes aegypti to Block Dengue and West Nile Viruses | Pathogen replication and transmission in Wolbachia infected insects are currently studied using three Wolbachia infection systems: naturally infected Wolbachia hosts, hosts transinfected with Wolbachia (stably maintained and inherited infections) and hosts transiently infected with Wolbachia. All three systems have been used to test the effect of Wolbachia on mosquito transmitted pathogens such as dengue virus (DENV), West Nile virus (WNV) and Plasmodium. From these studies it is becoming increasingly clear that the interaction between a particular pathogen and Wolbachia is heavily influenced by the host-Wolbachia interaction and the model of infection. In particular, there is some evidence that under very specific conditions, Wolbachia can enhance pathogen infection in some hosts. In this study, we compared the effect of Wolbachia in two infection models (stable transinfected and transiently infected) on the replication, infection- and transmission rates of two flaviviruses, DENV and WNV (Kunjin strain). Our results indicate that Wolbachia had similar blocking effects in both stable and transient models of infection, however, the magnitude of the blocking effect was significantly lower in mosquitoes transiently infected with Wolbachia. More importantly, no evidence was found for any enhancement of either DENV or WNV (Kunjin strain) infection in Ae. aegypti infected with Wolbachia, supporting a role for Wolbachia as an effective and safe means for restricting transmission of these viruses.
| Wolbachia is a naturally occurring endosymbiotic bacterium that, when introduced into a naïve mosquito host, has been shown to effectively reduce the replication and transmission of pathogens such as dengue virus, West Nile virus, Chikungunya virus, yellow fever virus and Plasmodium. However, a recent study has indicated that, under certain conditions, transiently infected Wolbachia can enhance West Nile virus infection in Culex tarsalis mosquitoes. We wanted to investigate whether this enhancement effect could also be observed in Aedes aegypti mosquitoes and if so, whether it is specific to the nature of the Wolbachia infection model under study (transient vs stable). We compared the replication and transmission of dengue virus and WNV (Kunjin strain) in Aedes aegypti mosquitoes transiently infected with Wolbachia and mosquitoes stably infected with the identical Wolbachia strain. Contrary to the previous study, our results show no enhancement of replication or transmission for either dengue virus or WNV (Kunjin strain) in mosquitoes transiently or stably infected with Wolbachia.
| The potential for Wolbachia as a natural control method for mosquito-borne pathogens such as dengue virus (DENV), Chikungunya virus (CHIKV), West Nile virus (WNV), yellow fever virus (YFV) and malaria has been the focus of intense study in recent years [1–8]. The majority of these studies have focussed on DENV replication and have shown conclusively that Wolbachia effectively reduces DENV replication and transmission when introduced as a stable infection in the naturally uninfected host Aedes aegypti [8–11].
To date, only a few mosquito species have been successfully transinfected with Wolbachia. These include Ae. aegypti (transinfected with the Wolbachia strains wMel, wMelPop, wAlbB, and superinfection with wMelwAlbB [6, 8–10]), Ae. albopictus (cured of its natural Wolbachia infection and transinfected with the wMel Wolbachia strain [12]), and Ae. polynesienses and Anopheles stephensi (both transinfected with the wAlbB Wolbachia strain [13, 14]). These transinfected strains have shown excellent potential for the biocontrol of several important mosquito-transmitted diseases (for recent reviews see [15–17]). However, several disease transmitting mosquito species remain recalcitrant to Wolbachia transinfection, hampering efforts to better understand the interaction between Wolbachia, it’s host and disease causing pathogens [18].
Natural Wolbachia infection models have therefore also been examined to provide insight into Wolbachia-host-pathogen interactions. In this model, the naturally occurring Wolbachia infection is first cured from the host and pathogen replication is subsequently compared in cured and naturally infected hosts [19, 20].
Using this model Baton et al. found that wFlu infection in its natural host Ae. fluviatilis, enhanced oocyst infection with the avian malaria parasite P. gallinaceum [19, 20]. Zele et al. also showed that in the natural mosquito-Wolbachia-Plasmodium combination, Wolbachia increased the susceptibility of Culex pipiens mosquitoes to P. relictum [20]. Furthermore, a study by Mousson et al. using this model, found that Ae. albopictus naturally superinfected with the two Wolbachia strains (wAlbA and wAlbB) infection limited the transmission, but not replication of DENV. Here, both the naturally occurring Wolbachia strains were cured and the vector competence for DENV of the resulting uninfected line was compared to the superinfected line [21].
In addition to natural infection systems, transient infection systems have been used to investigate the effect of Wolbachia on Plasmodium and WNV infection. Here, Wolbachia is injected into an uninfected mosquito host and allowed to establish a transient somatic infection [22]. Using this model, the effect of the Wolbachia strains wAlbB and wMelPop on the malaria parasite P. berghei in A. gambiae was investigated [23]. Contrary to the results of [14], in this experimental setup, wAlbB was found to enhance P. berghei infection, whilst wMelPop only had a moderate blocking effect [23]. A more recent study utilised the same infection model to investigate the effect of the wAlbB Wolbachia strain on WNV infection in Culex tarsalis [24]. Contrary to previous studies that found wAlbB inhibited WNV infection in Ae. aegypti [5], C. tarsalis transiently infected with wAlbB enhanced WNV infection rates at 7 days post infection [24].
Together these results suggest that the degree of pathogen modulation from different host-Wolbachia combinations can differ considerably depending on the mode of infection, the host and the pathogen. Consequently, it is important not to base predictions of pathogen modulation in a particular host-Wolbachia strain combination on results obtained from divergent infection modes and host species. In this study we have compared the effect of wAlbB on replication and transmission of DENV and WNV (Kunjin strain) in Ae. aegypti infected through both transient somatic infection and stable transinfection. Our results showed significantly lower Wolbachia infection densities in transiently infected Ae. aegypti when compared to the stable infected line. More importantly, both Wolbachia infection models displayed similar effects, blocking replication and transmission of both DENV and WNV (Kunjin strain). These results conclusively show that neither DENV nor WNV (Kunjin strain) infection is enhanced in Ae. aegypti either transiently or stably infected with wAlbB.
Wolbachia density and distribution was analysed in female Ae. aegypti mosquitoes transiently infected with the wAlbB Wolbachia strain and compared to the stable infected wAlbB line. Wolbachia density was determined using qPCR and primers specific to the Wolbachia surface protein (wsp) in conjunction with the Ae. aegypti actin gene for normalisation. In our experiments, even when Wolbachia was injected at very high densities (~1011 bacteria/mL), there were significantly lower (Mann-Whitney test, p = 0.007) Wolbachia densities at 7 days post injection (dpi) in the transiently infected mosquitoes than densities observed in the stable wAlbB infected line (Fig 1A).
Wolbachia in transiently infected mosquitoes were predominantly located in the brain (Fig 1B), muscle tissue (Fig 1C), the midgut (Fig 1D) and the fat body (Fig 1E). In stark contrast to the stable wAlbB infected line, however, very little to no Wolbachia could be detected in the ovaries of transiently infected lines (Fig 1F & 1G). In the stable wAlbB infected line, the vast majority of Wolbachia are found in the ovaries and the lack of Wolbachia found in the ovaries of transiently infected mosquitoes could explain the significant differences found in Wolbachia density between stable and transiently infected mosquitoes in our qPCR results (Fig 1A). These results are also consistent with previous studies that showed only limited Wolbachia localisation in the ovaries of transiently infected Culex tarsalis [24] and comparatively low levels of Wolbachia in the ovaries of transiently infected Anopheles gambiae compared to the rest of the body [25].
We next investigated whether female Ae. aegypti mosquitoes transiently infected with wAlbB displayed the same DENV blocking phenotype as the stable infected wAlbB line [24]. Townsville wild type (W.T.), W.T. transiently infected with wAlbB and stable wAlbB infected females were provided with a DENV infected blood meal 7 dpi. The mosquitoes were incubated for a further 7 days as described in materials and methods and subsequently analysed for DENV replication (Fig 2A), DENV infection rate (Fig 2B), DENV transmission rate (Fig 2C), as well as wAlbB density (Fig 2D).
DENV copy number (as determined by positive strand genome copy number) in the bodies of transiently infected females was significantly reduced (Mann-Whitney, p = 0.0002) by ~1.5 logs when compared to DENV replication in W.T. mosquitoes. Ae. aegypti stably infected with wAlbB showed the greatest reduction in + strand DENV genome copies with a ~ 3 log reduction compared to W.T. (Mann-Whitney, p = 0.001) and ~ 2 log reduction compared to transiently infected females (Mann-Whitney, p = 0.004). Similarly, the DENV infection rate was significantly reduced (~2 fold, Fisher exact test, p = 0.004) in Ae. aegypti females transiently infected with wAlbB when compared with W.T. females (Fig 2B). Ae. aegypti stably infected with wAlbB again showed the greatest reduction in DENV transmission rates with an ~8 fold reduction compared to W.T. females (Fisher exact test, p = 0.0001) and ~4 fold reduction compared to Ae. aegypti females transiently infected with wAlbB (Fisher exact test, p = 0.04).
DENV transmission in transiently infected wAlbB mosquitoes was significantly reduced compared to W.T. mosquitoes. Saliva was collected 7 days post feeding from females fed with an infected blood meal and then injected into DENV-naïve W.T. females according to [26]. The mosquitoes were incubated for an additional 7 days before analysing DENV infection status by qRT-PCR (Fig 2C). No DENV infectious saliva was detected from female Ae. aegypti mosquitoes stably infected with wAlbB. In contrast, 22% of female Ae. aegypti mosquitoes transiently infected with wAlbB expectorated DENV infectious saliva.
Finally, we analysed the Wolbachia density in the same transiently and stably infected mosquitoes analysed for DENV replication and transmission. Our results indicate much lower Wolbachia densities in the transiently infected females compared to the stable wAlbB infected line (Fig 2D). Wolbachia density has been correlated with the degree of pathogen blocking in Wolbachia infected hosts [27] and the lower densities in transiently infected mosquitoes observed here provides a plausible explanation for the reduced DENV blocking phenotype we observed in the transiently infected mosquitoes compared to the stable infected line.
We repeated the infection experiments using WNV (Kunjin strain). W.T. and wAlbB infected females were provided with a virus infected blood meal at 7 dpi. The mosquitoes were incubated for a further 7 days as described and virus titres were determined in whole bodies (Fig 3A) and saliva (Fig 3B). We also compared the virus infection (Fig 3C) and transmission rates (Fig 3D) between Wolbachia infected and uninfected mosquitoes.
Virus titres were significantly reduced in the bodies of both the transient and stable infected mosquito lines compared to W.T. females. Transient Wolbachia infection resulted in an approximate one log reduction (Mann-Whitney, p > 0.0001) in virus PFU (Fig 3A) in whole mosquito bodies. A small but significant (Mann-Whitney, p = 0.001), 0.28 log reduction in virus PFU was observed in saliva from these mosquitoes compared to W.T. (Fig 3B). The greatest reduction in virus PFU was observed in the stable infected mosquito line with a more than 1.5 log reduction in wAlbB-infected mosquito’s bodies (Fig 3A). No infectious virus was detected in saliva from these mosquitoes (Fig 3B).
We observed a small (less than two-fold), non-significant (Fisher exact test, p = 0.13) reduction of virus infection rates between transiently infected and W.T. mosquitoes (Fig 3C). Similarly, a small (less than two-fold), non-significant (Fisher exact test, p = 0.1) reduction in the percentage infectious saliva was found between transiently infected and W.T. mosquitoes (Fig 3D). A significant reduction (Fisher exact test, p = 0.001) in infection rates was observed between the stable wAlbB infected mosquito line and W.T. mosquitoes (Fig 3C).
As with the DENV infected mosquitoes, we analysed the Wolbachia density in the same transiently and stably infected mosquitoes analysed for WNV (Kunjin strain) replication and transmission. Similar to the DENV infected mosquitoes, our results indicate much lower Wolbachia densities in the transiently infected females compared to the stable wAlbB infected line (Fig 3E).
Wolbachia, when stably transinfected into mosquito hosts, has been shown to inhibit a range of pathogens, in particular DENV, CHIKV, WNV, YFV and Plasmodium [1–8]. There are however, a few studies that have demonstrated infection with Wolbachia can lead to enhanced pathogen replication [23, 24, 28, 29]. In particular, a study by Dodson et al. showed that when wAlbB transiently infects C. tarsalis, WNV infection rates can be enhanced [24]. These results are in contrast to a previous study that showed two stable transinfected Ae. aegypti lines (wMel and wMelPop) both inhibited WNV transmission [5].
This would suggest that the interaction between Wolbachia and a particular pathogen is highly dependent on either the infection model, the Wolbachia strain or the host background. To determine whether the results obtained by Dodson et al. [24] could be a result of the Wolbachia infection model, we compared DENV and WNV (Kunjin strain) infection in both Ae. aegypti transiently infected with wAlbB as well as Ae. aegypti stably transinfected with wAlbB. In our experimental setup, DENV and WNV (Kunjin strain) replication was significantly reduced in both Wolbachia infection models. In addition, DENV infection rate and transmission rate was significantly reduced in both models. We also observed a small, but not significant reduction in WNV (Kunjin strain) infection and transmission rates in transiently infected mosquitoes.
These observations differ markedly from those described by Dodson et al. [24] and suggest that in Ae. aegypti mosquitoes, transient and stable Wolbachia infections have similar pathogen modulation effects. In Ae. aegypti, unlike the observations in C. tarsalis, transient infection with wAlbB led to lower virus transmission rates in transiently infected mosquitoes compared to Wolbachia naive wild type mosquitoes. We also observed decreased pathogen blocking in transient Wolbachia infections compared to stable Wolbachia infections. Results generated through the use of transient Wolbachia infection models should therefore be interpreted with caution, as they could potentially underestimate the degree of pathogen blocking compared to the stably infected systems typically used for field disease control programs. Most importantly, our results conclusively show no enhancement of either DENV or WNV (Kunjin strain) infection in Wolbachia infected Ae. aegypti.
Blood feeding by volunteers (Monash University human ethics permit no CF11/0766-2011000387) for this study was approved by the Monash University Human Research Ethics Committee (MUHREC). All adult volunteers provided informed written consent; no child participants were involved in the study.
The experimental design for this study is summarised in S1 Fig. We compared the replication and transmission of DENV and WNV (Kunjin strain) in Townsville wild type (W.T.) mosquitoes, W.T. mosquitoes injected with wAlbB and the stable wAlbB line described in [9]. To generate transient Wolbachia infections in female W.T. mosquitoes, Wolbachia was isolated from 200 wAlbB-infected ovaries and injected into 100 W.T. mosquitoes. For the W.T. and stable wAlbB controls, an extraction was done from 200 W.T. ovaries in the same fashion as the Wolbachia extraction.
This extract was used to inject 50 W.T. females and 50 wAlbB stable infected females. The injected females were incubated for 7 days as described and subsequently allowed to feed on virus infected blood. Fed females were separated from unfed females 24 h post feeding. Females that showed no evidence of feeding were used to analyse the Wolbachia infection 7 days post injection, using qPCR and FISH. Engorged females were incubated for a further 7 days. Seven days post feeding, saliva and carcasses (legs and wings were removed) were collected from all fed mosquitoes and assayed for Wolbachia density, DENV and WNV (Kunjin strain).
Wolbachia-uninfected Ae. aegypti eggs were collected from Townsville (Queensland, Australia) in 2015. The Wolbachia-infected wAlbB mosquito line has been described previously [9] and was a gift from Prof Zhiyong Xi (Michigan State University). All Ae. aegypti mosquitoes were reared and maintained as described in [6] with the following modification. For hatching, eggs were placed in hatching water (distilled H2O, boiled and supplemented with 50 mg/L fish food [Tetramin]) and allowed to hatch for 24 h. Larvae were subsequently reared at a set density of ~150 in 3 L of distilled water as described in [6]. wAlbB infected females were backcrossed for 2 generations with W.T. males prior to infection experiments.
Wolbachia was isolated from the ovaries of wAlbB-infected females according to [24] with the following modifications. Ovaries from 200 wAlbB-infected females were dissected on ice and suspended in 50 μL of ice-cold Schneiders media (Sigma-Aldrich) in a 1.5 mL eppendorf tube. The ovaries were crushed briefly using a small plastic pestle after which one 3 mm glass bead was added and the suspension vortexed for 2 min. One mL of ice-cold Schneiders media was added to the homogenised and the solution were centrifuged at 4°C for 5 min at 2000 x g. The supernatant was subsequently sequentially filtered through 5 μM and 1.2 μM syringe filters. The resulting filtrate was centrifuged for 4°C for 10 min at 12000 x g. The supernatant was discarded and the pellet resuspended in 50 μL of ice-cold Schneiders media until use. The extraction was repeated with ovaries from W.T. females for use in control injections. Total bacterial counts were estimated using the LIVE/DEAD staining kit (Thermofisher) and counting the live stained bacteria in a hemocytometer.
Female mosquitoes were injected intrathoracically with ~ 1 μL of Wolbachia suspension solution (~1011 bacteria/ml in Schneiders media) using a pulled glass capillary and a handheld microinjector (Nanoject II, Drummond Sci.). Injected mosquitoes were incubated for 7 days (40 mosquitoes per cup) at 26°C with 65% relative humidity and a 12h light/dark cycle. All injection experiments were conducted in duplicate.
Wolbachia density and distribution in the transient infected mosquitoes were compared 7 and 14 days post injection (dpi) to the wAlbB line using qPCR and fluorescence in situ hybridisation (FISH). DNA was extracted from stable and transiently Wolbachia infected mosquitoes using the DNeasy 96 Blood & Tissue kit (Qiagen) according to the manufacturer’s specifications. Quantitative PCR to determine the total relative Wolbachia densities of infected lines was performed as described by [30] using primers specific to the gene coding for the Wolbachia surface protein (wsp) (forward primer 5’-GCATTTGGTTAYAAAATGGACGA-3’, reverse primer 5’- GGAGTGATAGGCATATCTTCAAT-3’), as well as the Ae. aegypti actin gene (forward primer 5’- GACTACCTGATGAAGATCCTGAC-3’, reverse primer: 5’- GCACAGCTTCTCCTTAATGTCAC-3’) [24]. Statistical differences were determined using a Mann-Whitney (Graphpad Prism version 6.0f).
Wolbachia was localized in sections of paraffin-embedded 5–7 day old female mosquitoes by FISH, as described in [31], except that only one probe against 16S rRNA was used and its concentration was increased 10-fold to improve the signal. wAlbB was detected using AlbBW5: 5’-CTTAGGCTTGCGCACCTTGCAA-3’, labelled with Alexa 488 dye (green). DAPI was used to stain total DNA.
The propagation and maintenance of dengue virus serotype 2 (DENV-2) ET300 [32] was carried out as previously described [33]. WNV (Kunjin strain) was obtained as a gift from Prof Jason Mackenzie (Melbourne University). WNV (Kunjin strain) was propagated on C6/36 cells in a fashion similar to DENV-2.
Mosquitoes were infected with either DENV-2 (ET300) or WNV (Kunjin strain) (virus strains were grown fresh for each infection) through an infectious blood meal. For feeding experiments with virus infected blood, Ae. aegypti female mosquitoes were placed in 500 mL plastic containers (40/container), starved for 24 hours and allowed to feed on a 50:50 mixture of defibrinated sheep blood and tissue culture supernatant containing 107 genome copies/mL of DENV-2 or 108 pfu/mL of WNV (Kunjin strain). Feeding was done through a piece of desalted porcine intestine stretched over a water-jacketed membrane feeding apparatus preheated to 37°C for approximately three hours. Fully engorged mosquitoes were placed in 500 mL containers and incubated for 7 days at 26°C with 65% relative humidity and a 12h light/dark cycle. All infection experiments were conducted in duplicate.
Saliva from infected mosquitoes was collected 7 days post feeding (dpf) as described by [26]. Following saliva collection, the bodies of infected mosquitoes were collected in 100 μL serum free RPMI media (Sigma-Aldrich) and stored at -80°C until processing. For DENV-2, the collected saliva was re-injected into 3-day-old W.T. female mosquitoes as described by [26]. Female mosquitoes injected with saliva were incubated for 7 days at 26°C with 65% relative humidity and a 12h light/dark cycle after which they were collected in RPMI media as above. DENV-2 genome copies were subsequently determined in the blood fed and saliva injected mosquitoes using qRT-PCR as described.
For WNV (Kunjin strain), the mosquitoes and saliva were collected as described above. After collection the mosquito bodies were homogenised in a bead beater at 30 beats/min for 3 min using one 3 mm sterile glass bead. The suspension was briefly centrifuged at 2000 x g and 10 μL of the supernatant was used in plaque assays as described by [24]. Collected saliva was used directly in plaque assays.
To quantify DENV-2 genomic copies, total RNA was isolated from DENV-2 injected mosquitoes using the Nucleospin 96 RNA kit (Macherey-Nagel). DENV-2 qPCR analysis was done using cDNA prepared from individual mosquitoes according to [31] using forward primer 5’-AAGGACTAGAGGTTAGAGGAGACCC-3’ and reverse primer 5’-CGTTCTGTGCCTGGAATGATG-3’. Infectious virus titre of WNV (Kunjin strain) was quantified using plaque assays as described by [24].
|
10.1371/journal.pgen.1005868 | Gene Network Polymorphism Illuminates Loss and Retention of Novel RNAi Silencing Components in the Cryptococcus Pathogenic Species Complex | RNAi is a ubiquitous pathway that serves central functions throughout eukaryotes, including maintenance of genome stability and repression of transposon expression and movement. However, a number of organisms have lost their RNAi pathways, including the model yeast Saccharomyces cerevisiae, the maize pathogen Ustilago maydis, the human pathogen Cryptococcus deuterogattii, and some human parasite pathogens, suggesting there may be adaptive benefits associated with both retention and loss of RNAi. By comparing the RNAi-deficient genome of the Pacific Northwest Outbreak C. deuterogattii strain R265 with the RNAi-proficient genomes of the Cryptococcus pathogenic species complex, we identified a set of conserved genes that were lost in R265 and all other C. deuterogattii isolates examined. Genetic and molecular analyses reveal several of these lost genes play roles in RNAi pathways. Four novel components were examined further. Znf3 (a zinc finger protein) and Qip1 (a homolog of N. crassa Qip) were found to be essential for RNAi, while Cpr2 (a constitutive pheromone receptor) and Fzc28 (a transcription factor) are involved in sex-induced but not mitosis-induced silencing. Our results demonstrate that the mitotic and sex-induced RNAi pathways rely on the same core components, but sex-induced silencing may be a more specific, highly induced variant that involves additional specialized or regulatory components. Our studies further illustrate how gene network polymorphisms involving known components of key cellular pathways can inform identification of novel elements and suggest that RNAi loss may have been a core event in the speciation of C. deuterogattii and possibly contributed to its pathogenic trajectory.
| Genome instability and mutations provoked by transposon movement are counteracted by novel defense mechanisms in organisms as diverse as fungi, plants, and mammals. In the human fungal pathogen Cryptococcus neoformans, an RNAi silencing pathway operates to defend the genome against mobile elements and transgene repeats. RNAi silencing pathways are conserved in the Cryptococcus pathogenic species complex and are mediated by canonical RNAi components. Surprisingly, several of these components are missing from all analyzed C. deuterogattii VGII strains, the molecular type responsible for the North American Pacific Northwest outbreak. To identify novel components of the RNAi pathways, we surveyed the reference genomes of C. deuterogattii, C. gattii, C. neoformans, and C. deneoformans. We identified 14 otherwise conserved genes missing in R265, including the RDP1, AGO1, and DCR1 canonical RNAi components, and focused on four potentially novel RNAi components: ZNF3, QIP1, CPR2, and FZC28. We found that Znf3 and Qip1 are both required for mitotic- and sex-induced silencing, while Cpr2 and Fzc28 contribute to sex-induced but not mitosis-induced silencing. Our studies reveal elements of RNAi pathways that operate to defend the genome during sexual development and vegetative growth and illustrate the power of network polymorphisms to illuminate novel components of biological pathways.
| Genome reduction is a common adaptation among bacterial pathogens and commensals, and has been hypothesized to occur for a number of reasons, including increased specificity to a host or environmental range, or to increase virulence more directly through loss of an antivirulence gene or gene cluster. The former case can be explained primarily through loss of genes that play only accessory roles. These genes can become dispensable as an organism becomes obligately associated with a host, which then acts as an alternative source for these gene products, such as amino acids or metabolic intermediates [1–3]. In some cases, network polymorphisms can result from loss of one of the components, which then enables additional inactivating mutations to occur in other components of the crippled or disabled pathway, such as loss of the Gal80 repressor in Saccharomyces kudriavzevii [4]. Genes also can be lost as a result of an “antivirulence” function, as is seen in Shigella and E. coli, where the presence of the lysine decarboxylase cadA interferes with the synthesis of enterotoxins through production of cadaverine [5]. This model, termed the black hole hypothesis, suggests that gene losses can be the result of active interference with pathogenesis, likely as the result of gain of a new incompatible function. In either model, understanding the gene network polymorphism can elucidate the biology and evolution of the pathogen, facets that are particularly relevant for new and emerging pathogens.
Cryptococcus deuterogattii, previously C. gattii molecular type VGII [6], is an emerging human fungal pathogen in the Pacific Northwest (PNW) of the United States and southwest Canada [7–9]. While the sibling species C. neoformans predominantly infects immunocompromised individuals, many of the C. deuterogattii infected patients in the Pacific Northwest outbreak were otherwise healthy. Both species cause severe pulmonary and central nervous system infections, and are fatal if untreated. Surprisingly, whole genome sequencing revealed that the C. deuterogattii strain R265 is missing both of the Argonaute genes, essential components of the RNAi-induced silencing complex (RISC) [10,11]. Further examination revealed that in addition to the loss of both Argonaute genes, one of the two Dicers and the only RNA-dependent RNA polymerase have also undergone pseudogenization through large sequence losses similar to those of the Argonaute genes [12]. The loss of critical canonical components of the RNAi pathway raises a number of questions about the origins and biology of the C. deuterogattii species as well as the function of RNAi within the Cryptococcus pathogenic species complex as a whole.
RNA interference (RNAi) is a highly conserved mechanism among eukaryotes that facilitates homology-dependent gene silencing. This transcriptional regulatory strategy was initially observed in Caenorhabditis elegans where exogenously introduced double-stranded RNA (dsRNA) triggers silencing of the transcript complementary to the dsRNA sequence [13]. Since its discovery in C. elegans, numerous species of plants, animals, fungi, and protists have been found to employ similar strategies to either protect their genomes from foreign DNA or to orchestrate gene expression and diverse cellular, developmental, and physiological processes [14–17]. Repetitive sequences are often found in mobile genetic elements and previous studies found an association between RNAi and transposable elements, which are ubiquitous in eukaryotic organisms. Transposon activation and movement impairs genome stability and increases the mutational burden of the host. Therefore, eukaryotes employ different strategies to inhibit and limit transposon expansion. Arabidopsis thaliana, Drosophila melanogaster, Saccharomyces castellii, Neurospora crassa, and C. elegans all utilize RNAi strategies to control and inhibit transposon expression [18–22].
C. neoformans also employs an RNAi-related pathway to inhibit transposable elements. In previous studies, Wang et al. showed that the insertion of a tandem multicopy transgene triggered a homology-dependent gene silencing mechanism during sexual development and termed this process sex-induced silencing (SIS) [10]. This process was identified specifically with a SXI2a-URA5 transgene array inserted into the ura5 locus, resulting in the presence of three functional copies of URA5 and one nonfunctional copy. During mating, progeny that inherit the array silence the URA5 gene in an RNAi-dependent manner approximately 50% of the time. In addition, Wang et al. later found that transgene silencing can also occur during vegetative growth, named mitotic-induced silencing (MIS), but at a relatively lower frequency in mitotic (~0.2%) compared to meiotic progeny (~50%) [23]. Further analysis showed that SIS and MIS require the RNAi components Rdp1 (RNA-dependent RNA polymerase), Ago1 (Argonaute), and Dcr1/2 (dicer-like proteins) [10,23]. SIS and MIS function to inhibit transposon movement and thus serve as a genome defense mechanism during meiosis and mitosis. The initial observation of transposon silencing during sexual development was made in the highly virulent C. neoformans lineage. Later studies found that transgene-related SIS also occurs in C. deneoformans and that the RNAi components are required for transposon silencing during both bisexual and unisexual development [24].
The lack of the critical Argonaute, Dicer, and RdRp components of the RNAi pathway in C. deuterogattii suggests that the loss of RNAi may represent a gene network polymorphism. In fact, the RNAi pathway is intermittently conserved and lost across eukaryotes [12,25–27]. In Leishmania and trypanosomes, RNAi losses were previously taken advantage of in order to identify additional, previously unknown components of the RNAi pathway via comparative genomics [16]. To test the hypothesis that the RNAi pathway represents a gene network polymorphism, we surveyed the genomes of the R265 (C. deuterogattii), WM276 (C. gattii), H99 (C. neoformans), and JEC21 and B-3501A (C. deneoformans) strains and found 14 genes missing from C. deuterogattii, including the canonical components of the RNAi pathway RDP1, AGO1, and DCR1. Here we focus on four of these lost components: ZNF3, previously identified as a regulator of hyphal development during unisexual and bisexual reproduction [28]; CPR2, a G-protein coupled receptor (GPCR) previously studied for its role as an accessory constitutively active pheromone receptor [29]; QIP1, independently identified as an RNAi component via a mass spectrometry approach [30]; and FZC28, a putative transcription factor with no obvious phenotypes in a systematic genome-wide transcription factor deletion study [31].
Here we demonstrate that the loss of the RNAi components represents a bona fide system polymorphism, with several previously unknown RNAi components lost in C. deuterogattii. In addition, we show that mutants of these missing genes in C. neoformans fall into two classes: mutants that lose both vegetative silencing and sex-induced silencing, and mutants that are affected only in the frequency of sex-induced silencing. This suggests that sex-induced silencing may be a specialized, highly induced variant of the vegetative transgene-induced silencing pathway, rather than a separate pathway. Taken together, our results show that a substantial loss of genes contributing to two related RNAi pathways has occurred in C. deuterogattii. By using comparative genomics, these gene losses reveal key insights that aid in elucidating the functions of these RNAi-based genome conservation pathways.
The C. deuterogattii lineage (previously VGII C. gattii) is responsible for the recent, ongoing outbreak on Vancouver Island and its expansion into the Pacific Northwest of the United States. Initial analysis of the R265 C. deuterogattii reference genome revealed that both the key canonical RNAi components AGO1 and AGO2 are missing, indicating that the VGII lineage of C. deuterogattii may lack a functional RNAi pathway [10,11]. Upon further examination, we discovered that two of the other canonical components, DCR1 and RDP1, had both suffered truncations removing key functional domains and are therefore pseudogenes. Of the known RNAi canonical components, only DCR2 remains intact in C. deuterogattii (Fig 1A) [11,32–35]. We hypothesized that this loss of multiple RNAi components may represent a gene network polymorphism where all of the components of a pathway are intact in one species, but have been selectively lost in another closely related species. We further hypothesized that a whole genome comparison of C. deuterogattii with other related Cryptococcus species would reveal novel components of the RNAi pathway lost in C. deuterogattii but otherwise maintained throughout the pathogenic species complex.
We compared the publicly available reference genomes of JEC21 (C. deneoformans)[36], B-3501A (C. deneoformans)[36], H99 (C. neoformans)[37], and WM276 (C. gattii)[11] with R265 (C. deuterogattii)[11] to identify otherwise conserved genes that were missing or truncated in the C. deuterogattii lineage. We found seven conserved genes that were not annotated in R265 and seven others that were dramatically shortened (over 50% different in length) as a result of extensive deletions of genomic sequence (Table 1). All 14 genes were lost across the entire VGII group, based on 53 publicly available whole genome sequences from C. deuterogattii [32]. These genome sequences did reveal some diversity in these regions. Estimation of Tajima’s D in windows across the genome and within the regions left by the deletion events showed a highly negative value for the genome as a whole (mean of -1.122), and a slightly more positive (mean of -0.796), but not statistically significant value (p = 0.0901) for the deletion windows (S1 Fig). We did not identify any transposable elements or repeats that may have mediated the deletion events. One of the seven missing genes was the previously identified canonical RNAi component AGO1. In each case, localized deletions of sequence occurred, encompassing entire ORFs, start codons, and/or functional domains of the candidate genes (Fig 1B–1H and S2 Fig). Our screen identified two potential transcription factors, FZC27 and FZC48, and three genes, including GWC1, GWO1, and QIP1, which have been previously identified as participating in the degradation of unspliced mRNA through RNAi [30]. Two of the 14 missing or truncated genes, CPR2 and ZNF3, were previously shown to play roles in unisexual and bisexual reproduction, but were not described as having a role in RNAi [17,35]. We chose to focus on four genes as candidates to interrogate for a role in the SIS and MIS RNAi pathways: CPR2, FZC28, ZNF3, and QIP1.
CPR2 encodes a seven transmembrane domain GPCR closely related phylogenetically to the Ste3 family of pheromone receptors, but it is constitutively active and independent of pheromone ligand binding [29]. Cpr2 signals via the same G proteins as the pheromone receptor Ste3, and overexpression of CPR2 can rescue the sterility defect of ste3Δ mutants, although it may bias cells towards unisexual reproduction [29]. FZC28 is a transcription factor about which very little is known. It was identified and mutated as part of a genome-wide transcription factor deletion library, and experiments in that study identified no obvious phenotypes [31].
In previous studies we found that ZNF3 is required for hyphal development during unisexual and bisexual reproduction in C. deneoformans [28]. Deletion of the gene blocks hyphal development and impairs pheromone expression during mating. However, it does not play a direct role in the pheromone-signaling cascade. Surprisingly, microarray expression analysis revealed that deletion of Znf3 increased transposon and transposon-related gene expression during bisexual reproduction [28]. Znf3 is also somewhat rapidly diverging in amino acid sequence. While it is found in the Cryptococcus pathogenic species complex and the neighboring sensu stricto (including C. amylolentus) and sensu lato groups (including C. heveanensis), the sequence is not well conserved, and it shares only weak homology over a 211 amino acid stretch (23% identity and 38% positive) with the reciprocal best BLAST hit ortholog in Tremella mesenterica. The encoded protein in Cryptococcus neoformans contains three zinc finger domains, two predicted nuclear localization signals (NLS), and a conserved coiled coil region, often involved in protein-protein interactions, as well as a putative ribonuclease conserved domain indicating that it may be involved in cleavage of RNA.
QIP1 is named for N. crassa QIP, which functions during quelling and MSUD by binding to RISC and stimulating cleavage of the passenger strand of the duplex siRNA [38]. Moreover, a previous study directly implicated Qip1 in the transcriptional squelching of transposons and the degradation of mRNAs that have poorly spliced non-canonical introns [30]. Dumesic et al. localized Qip1 in the nucleus and showed that it physically interacts with Rdp1 as part of the Spliceosome-Coupled and Nuclear RNAi (SCANR) complex [30]. Analysis of N. crassa Qip revealed a conserved 3’-5’ exonuclease domain belonging to the DEDDh superfamily of nucleases, showing high similarity to the E. coli DNA polymerase III ε subunit [33]. Although, the C. neoformans Qip1 protein does not contain any detected conserved functional domains, it exhibits weak similarity to the helical domain of Class IIa histone deacetylases, which may suggest a role different than that of N. crassa Qip.
In previous studies, Wang et al. found that a tandem multicopy insertion of a SXI2a-URA5 transgene triggered silencing of the URA5 gene during bisexual reproduction and vegetative growth in C. neoformans [10,23]. When F1 progeny were isolated from a cross between WT MATα URA5 (H99α) and MATa SXI2a-URA5 (JF289), ~25% were found to be uracil-auxotrophic despite the fact that all of them had intact copies of the URA5 allele. Further analysis revealed that ~50% of the progeny that inherited the SXI2a-URA5 transgene were uracil auxotrophic. Recent studies showed that the transgene induced silencing mechanism is activated efficiently during bisexual and unisexual reproduction (SIS) and less efficiently during vegetative growth (MIS) [23,24].
Deletion of RNA-dependent RNA-polymerase Rdp1 abolished transgene induced silencing during SIS and MIS in both C. neoformans and C. deneoformans. To investigate the role of the missing genes from R265 in silencing we generated deletion mutants in the JF289a isolate bearing the SXI2a-URA5 transgene (derived from strain KN99a), and the congenic WT H99α strain. Two independent deletion mutants for each gene were isolated and analyzed.
To determine the silencing efficiency of the mutants during sexual reproduction, unilateral (one parent is mutant) and bilateral (both parents are mutants) crosses were performed on MS media. We dissected random F1 spore progeny from each cross and these were tested for growth in the absence of uracil and genotyped for the presence of the SXI2a-URA5 transgene (Fig 2A and S3 Fig). In unilateral matings with a deletion allele only present in one of the two parents, two meiotic progeny were ura- for qip1Δ (out of 14 inheriting the array, ~14%), none were ura- for znf3Δ (out of 18 inheriting the array, 0%), and three were ura- for cpr2Δ (out of 22 inheriting the array, ~13.6%) indicating significantly reduced silencing efficiency compared to WT (Fig 2A and S4 Fig). These results suggest that all three components play a role in RNAi during sexual development. In contrast, the silencing efficiency of the SXI2a-URA5 transgene in the fzc28Δ, and fzc47Δ unilateral mutant matings was similar to WT (~50%) (Fig 2C). All of the ura- progeny carry an intact copy of the SXI2a-URA5 transgene, as verified by PCR.
Previous studies showed that bilateral matings of all three canonical RNAi component mutants (ago1Δ, dcr1Δ, rdp1Δ) yielded ~20 fold fewer spores, with rdp1Δ mutants also demonstrating disorganized and atypical basidia, but with no effect on the sporulation efficiency of the spores that were produced [10]. Similarly, although deletion of ZNF3 severely impaired mating in C. deneoformans [28], hyphal development during bisexual reproduction was similar to WT in C. neoformans znf3Δ mutants, albeit somewhat delayed. In contrast, in bilateral qip1Δ x qip1Δ mutant crosses we found that spore production was severely impaired and the few spores that were isolated failed to germinate, indicating that Qip1 is required for completion of the sexual cycle and may play a role in meiosis (S4B Fig). On the other hand, deletion of RDP1 or ZNF3 did not affect sporulation efficiency. Deletion of ZNF3 in both parents completely abolished silencing, as none of the progeny that inherited the transgene were ura- (S4A Fig). These results indicate that Znf3 is required for silencing during mating and deletion of the gene causes a severe SIS silencing defect, similar to rdp1Δ. Silencing of the URA5 gene was also impaired in fzc28Δ and cpr2Δ bilateral matings (Fig 2C). However, fzc47Δ mutation in both parents did not impair silencing of the URA5 transgene and it was similar to WT, despite a modest increase in silencing rate in a unilateral cross (Fig 2C and S4 Fig).
We then examined the silencing frequency of the SXI2a-URA5 transgene in the mutant strains by measuring spontaneous 5-FOA resistance following mitotic growth in rich media. The strains bearing the qip1Δ and znf3Δ deletions failed to yield any colonies on 5-FOA media, indicating that these two genes are required for transgene-induced mitotic silencing (Fig 3). In contrast, deletion of two transcription factors, FZC28 and FZC47, obtained from a recently reported systematic transcription factor deletion collection and crossed into the JF289 background [31], and the GPCR CPR2, did not alter the mitotic silencing frequency of the SXI2a-URA5 transgene compared to WT.
In conclusion, we found that Znf3 and Qip1 are required for silencing during both MIS and SIS and deletion of the genes generates a phenotype similar to mutation of RDP1, whose gene product is essential for RNAi function in C. neoformans. These results suggest that Znf3 and Qip1 are novel regulators or components of the RNAi pathway. In addition we found that a new transcription factor Fzc28 and the GPCR Cpr2 influence transgene-induced silencing specifically during sexual development, possibly coupling the sexual cycle with the RNAi pathway but likely not acting mechanistically during silencing itself.
In a previous study we found that deletion of Znf3 in C. deneoformans activates transposon expression [28] and here we have shown that it is required for MIS and SIS. Recent studies revealed that transposable element expression increases during sexual reproduction and the components of the RNAi pathway maintain genome integrity through an efficient transposon silencing mechanism [10]. Deletion of RDP1 results in centromeric and telomeric retrotransposon overexpression during sexual development in C. neoformans [10]. We examined the transcript abundance of two transposons, Tcn1 and Tcn2, in znf3Δ mutant crosses and found that abundance was dramatically increased, similar to rdp1Δ and ago1Δ mutant crosses (Fig 4A). Deletion of QIP1 also yielded elevated levels of transposon transcript abundance, indicating that Qip1 also plays a major role in transposon quenching during sexual development (Fig 4A).
To further investigate the role of Znf3 in transposon silencing on a genome-wide scale, we performed a comparative transcriptome analysis of znf3Δ x znf3Δ and rdp1Δ x rdp1Δ crosses during sexual development and vegetative growth. Bilateral crosses of znf3Δ x znf3Δ and rdp1Δ x rdp1Δ mutants were incubated on solid V8 medium (pH = 5) for 24 hours, as well as H99α x JF289a wild type crosses. RNA was isolated from the mating cultures, transcribed to cDNA, and hybridized to a C. neoformans genome microarray.
Genome-wide expression analysis revealed that among the transcripts with altered expression level, the majority were increased in the znf3Δ mutant cross relative to WT during sexual development, indicating that Znf3 has a repressive role during sexual development. The few transcripts whose abundance was decreased in the znf3Δ and rdp1Δ crosses are involved in hypoxia, oxidation, ion channels, sugar transport, and possibly sporulation. During znf3Δ sexual development more than 80 independent microarray tags exhibited a twofold increase in abundance compared with the WT. Further analysis revealed that the majority of these tags correspond to sequences from hypothetical proteins or align to intergenic regions of the C. neoformans H99 genome. Alignment to a retrotransposon library [39] showed that almost all of the intergenic probes that were increased in znf3Δ mutants correspond to retrotransposon sequences found in multiple sites in the genome (S3 Table). We found that these retrotransposons have long terminal repeats (LTR) and reside in the centromeric and telomeric regions of the chromosomes. In addition, most of the upregulated hypothetical proteins in znf3Δ x znf3Δ crosses were found to be RNA and DNA helicases, RNA-dependent DNA polymerases, and other transposon-related proteins (S3 Table). During vegetative growth fewer transcripts were upregulated in znf3Δ mutants; however, the transcripts that exhibited differential abundance were also involved in transposon expression or activation. As was observed previously, the Tcn1, Tcn2, and Tcn3 elements were increased in znf3Δ × znf3Δ crosses, while their abundance was diminished during znf3Δ vegetative growth but remained significantly higher than the WT. We compared the transcriptional profile to the rdp1Δ x rdp1Δ mutant cross profile, and the whole genome transcript profiles between the two mutants were highly similar (Fig 4B). The highly correlated transcript profiles of upregulated genes suggests that Znf3 and Rdp1 have similar functions and may mediate retrotransposon silencing through the same RNAi pathway.
Interestingly, in spite of the loss of RNAi components in C. deuterogattii, transposon copy number does not appear to have dramatically increased in the genome (S5 Fig). The vast majority of transposable elements are present at substantially lower copy number in C. deuterogattii compared to C. gattii. However, several classes of transposable elements are present in approximately equal amounts (TCN3 and TCN6) or at even higher levels (TCN4 and LTR13) in C. deuterogattii (R265) than in C. gattii (WM276), based on a BLAST search using a C. neoformans library [39].
In previous studies we found that, although Znf3 regulates sexual development, ZNF3 expression remains stable during vegetative growth and mating in C. deneoformans [28]. In addition, mRNA levels for the RNAi components are relatively similar between mitotic growth and mating based on northern blot analysis; however, their protein abundance was significantly higher during sexual development suggesting that the RNAi components are translationally induced or stabilized during the sexual cycle [10]. Based on this evidence we hypothesized that ZNF3 and QIP1 expression might also remain the same between the two conditions in C. neoformans. RNA was isolated during mitotic growth and mating from WT and bilateral mutant crosses and the abundance of their transcripts was analyzed using quantitative RT-PCR. Unlike the canonical RNAi components, we found that both ZNF3 and QIP1 expression was significantly higher during mating compared to WT (Fig 5A). This was a surprising result given that the expression of the highly conserved ZNF3 gene in C. deneoformans remains the same and similar to WT during both conditions [28]. Moreover, Znf3 and Qip1 have similar roles with the RNAi components in SIS and MIS whose expression remains stable. This indicates that Znf3 and Qip1 expression may have a unique mode of regulation distinct from Rdp1 and Ago1.
We next assessed whether the RNA abundance during sexual development is correlated with the protein level between the two conditions. The C-termini of Znf3 and Qip1 were fused with mCherry at the endogenous genomic loci. MIS and SIS assays were conducted to test if the chimeric proteins retain their functional roles in silencing. Znf3 tagged with mCherry was completely defective in SIS and MIS, indicating that the mCherry tag interferes with function. On the other hand, Qip1 tagged with mCherry exhibited wild type levels of silencing during vegetative growth and sexual development. The protein levels were examined during both conditions and we found that, although the Qip1 protein was present during both vegetative growth and mating, it was significantly more abundant during sexual development, similar to the difference observed in RNA abundance (Fig 5B). These results indicate that Qip1, and possibly also Znf3, have a unique mode of regulation that is possibly distinct from that of other RNAi components.
The MIS and SIS silencing phenotypes of znf3Δ and qip1Δ mutants are very similar to rdp1Δ mutants. Previous studies have suggested that an unknown RNA-binding factor may govern translational regulation of the transcripts of the RNAi components to result in elevated protein levels specifically during sexual development [10]. We found that Znf3 bears both zinc fingers and an RNase domain and transcription of the gene is sexually induced. Considering that Znf3 has a similar phenotype to Rdp1, it could be involved in the translational regulation of the RNAi components, and the severe loss of silencing phenotype of znf3Δ mutants might be attributable to an absence of these factors.
It is unlikely that Znf3 regulates the transcription of the RNAi components based on microarray expression analysis. Nevertheless, we performed quantitative RT-PCR in the absence of each of the RNAi components during sexual development. Surprisingly, we observed a modest increase in the expression of the RNAi components during sexual development compared to vegetative growth (Fig 5C). In previous studies, northern blot analysis was employed to investigate the expression of these genes during vegetative growth and mating, and the modest 2- to 4-fold increase we observed using RT-PCR was possibly below the level of detection by northern blot. However, deletion of ZNF3 and QIP1 did not alter the expression of RDP1 or AGO1, suggesting that Znf3 and Qip1 do not act as transcriptional regulators of the canonical RNAi components or mediate the modest increase in expression we observed in mating conditions We also investigated the expression of ZNF3 and QIP1 in the absence of the canonical RNAi components during sexual development. Deletion of RDP1 did not affect the ZNF3 transcript levels during sexual development, indicating that RDP1 does not control the expression of this gene (Fig 5C). The expression of Znf3 was modestly but significantly increased in the ago1Δ mutants, which is the catalytic subunit of the RISC complex. Interestingly, expression of QIP1 during sexual development decreased to vegetative levels in the absence of RDP1.
To explore a possible role of Znf3 in the translational regulation of the RNAi components, we deleted ZNF3 and investigated the protein levels of Ago1 and Rdp1 fused with mCherry under the control of the endogenous promoter during sexual development. We detected a strong protein signal for both Ago1-mCherry and Rdp1-mCherry during sexual development with or without ZNF3 (Fig 5D). Western blot analysis revealed that deletion of ZNF3 resulted in a modest decrease in the protein abundance of Rdp1 under mating conditions (Fig 5E). It is possible that this decrease may not have been detectable via direct microscopy of cells expressing the Rdp1-mCherry fusion protein (Fig 5D). These results indicate that although Znf3, is not involved in transcriptional regulation of the canonical RNAi components, it could be involved in either translational regulation or in modulating protein stability via a role as a scaffolding protein.
RNAi silencing is a multifunctional pathway and different steps occur at different sites within the cell. The presence of tandem repeated genes or retrotransposons in the genome induces the transcription of aberrant ssRNA in the nucleus through an unknown mechanism and Rdp1 generates dsRNA from these sequences and evokes the RNAi pathway. The dsRNA travels to P-bodies, where processing and RNA silencing occurs. Dcr1/2 and Ago1, which localize to P-bodies, generate siRNAs that target mRNAs with complementary sequences for degradation [10]. These findings suggest that additional components of the pathway will localize either to the nucleus or to P-bodies.
Znf3 has two NLS signals, therefore we initially hypothesized that Znf3 might localize to the nucleus where it could act as a transcription factor, or bind and degrade dsRNAs generated by Rdp1. To investigate the localization of Znf3, and because endogenous C-terminal tagging had failed to produce functional protein, the N-terminus of the protein was fused to mCherry, and expressed from the constitutively active GPD1 promoter. The H99α and JF289a strains were transformed with the mCherry-Znf3 plasmid and evaluated by direct fluorescence microscopy. Surprisingly, we observed multiple bright foci in the cells indicating that the protein was present in more than one cellular compartment during sexual development (Fig 6A). To determine this cellular localization, we utilized two established marker components, one for P-bodies and the other for the nucleus. Dcp1, found in P-bodies, is responsible for decapping mRNAs during exonucleolytic degradation, while Nop1 is a component of the small subunit processome (a ribosome assembly intermediate) complex of the nucleolus [10,40]. GFP-Dcp1 and GFP-Nop1 were expressed from plasmids that were ectopically introduced into the genomes of strains expressing the mCherry-Znf3 protein and localization was observed during vegetative growth and sexual development. Surprisingly, we found that Znf3 localizes only in the P-bodies during both vegetative growth and sexual reproduction, despite the putative NLS signals (Fig 6A). These results suggest that Znf3 may participate directly in the RNAi silencing process and it may represent a novel element of the RNAi pathway.
Previous studies found that Qip1 localizes in the nucleus and that it physically interacts with Rdp1 and Ago1 during vegetative growth [30]. Although Ago1 is primarily localized in P-bodies during mating, where RNA silencing occurs, it has been also reported in the nucleus under vegetative growth conditions [30]. To further investigate the localization of Qip1, the protein was fused at the C-terminus with mCherry and expressed from the endogenous QIP1 promoter. The fluorescent signal was evaluated via microscopy during vegetative growth and sexual development. Co-localization of Qip1-mCherry with GFP-Dcp1 or GFP-Nop1 revealed surprising results. During sexual development, where the RNAi pathway is highly induced, Qip1 was localized exclusively in P-bodies, potentially reflecting a role in RNA degradation (Fig 6B). During vegetative growth we observed Qip1 in association with either the P-bodies or the nucleus (Fig 6B). These results suggest that Qip1 may interact with Rdp1 in the nucleus during vegetative growth, possibly as a component of the SCANR complex to participate in the processing of the stalled splicing intermediate [30]. During mating Qip1 migrates to the P-bodies where it may subserve its conserved role in the RISC complex.
We observed that both cpr2Δ and fzc28Δ mutants had defects in SIS but not in MIS. This suggests that these two pathways may differ in more than just their efficiency. As a result, we sought to test whether the role of Cpr2 in SIS was linked to its role in mating or independent of this function. We tested this by analyzing mutants lacking Ste3α, a pheromone receptor that shares the same G proteins as Cpr2. Ste3α mutants fail to mate, so a deletion was instead constructed in an a/α diploid (Fig 7A). FACS was used to verify that two independent ste3αΔ/a deletions remained stably diploid (Fig 7B). The ste3αΔ/a diploids were then sporulated and the progeny were dissected and tested for silencing of the URA5 transgene. Both independent mutants demonstrated a defect in SIS, with a silencing frequency of only ~21–22% compared to 50% silencing in WT crosses.
To test whether this effect was mediated by the shared downstream MAP kinase cascade (Fig 7D), we utilized a ste3αΔ mutant complemented with an overexpressed CPR2 gene under the control of the GPD1 promoter to test whether ectopic overexpression of Cpr2 could compensate for the ste3αΔ mutant defect in SIS. This strain was mated with the JF289a SXI2a-URA5 transgene array strain and spores were dissected, germinated, and phenotyped (Fig 7E and 7F). Overexpression of Cpr2 restored the SIS efficiency of the ste3αΔ/a mutant to ~ 67–68%. This suggests that both Cpr2 and Ste3 act coordinately to induce the RNAi pathway during the sexual cycle, and may not act as essential RNAi components themselves. Similarly, the transcription factor Fzc28 is a candidate to be the downstream effector of the Ste3/Cpr2 pathway, as it has no MIS defect but an absolute defect in SIS, suggesting it is essential for that arm of the pathway.
In this study we found that four novel proteins are required for silencing of the SXI2a-URA5 transgene during sexual development (SIS) and/or vegetative growth (MIS), and that they fall into two distinct classes: proteins essential for both RNAi pathways, and proteins influencing just the sex-induced arm of the pathway. Deletion analysis reveals that Znf3 and Qip1 are required for MIS and SIS RNAi silencing, similar to the canonical RNAi component Rdp1, while Cpr2 and the novel transcription factor Fzc28 influence only in SIS.
In further support of this model we found that rdp1Δ and znf3Δ mutants have very similar transcript profiles during mating characterized by increased abundance of messages from retrotransposons and other transposon-related genes. However, unlike RNAi components, whose expression remains largely stable during mitotic growth and mating, ZNF3 and QIP1 are transcriptionally induced during the sexual cycle. Although Znf3 has two NLS tags, we found that it localizes in P-bodies, where Dcr1/2 and Ago1 are also localized. Qip1 also localizes in P-bodies during sexual development; however, in some cases it migrates to the nucleus during vegetative growth. This may indicate that these two proteins act mechanistically in the silencing pathway, rather than as regulators of the canonical components.
RNA silencing is a highly conserved mechanism of transcriptional regulation. Since its discovery in C. elegans it has been identified in numerous species throughout the eukaryotic kingdom and it is hypothesized to be an ancestral feature of the last common eukaryotic ancestor [13,14,41]. An RNAi-related phenomenon was initially identified in plants and fungi, and later multiple species have been found to undergo RNA silencing mechanisms, including the fungi Neurospora crassa, Mucor circinelloides, and Schizosaccharomyzes pombe [20,42–44]. However, RNAi has been independently lost in some species, such as Saccharomyces cerevisiae and Ustilago maydis, which are missing all of the components of the RNAi pathway [18,25]. Nevertheless, the closely related species of S. castellii and C. albicans retain some of the RNAi components and substitute for the absence of others by employing noncanonical factors to produce dsRNA and shRNAs that map to transposable elements [45]. As a result, it is possible to learn more about the intact RNAi pathway of a species by comparing it to a related species that has lost some or all of its RNAi components [16,26]. This study shows that examining individual cases of RNAi loss with available sequenced genomes for closely related organisms is likely to be fruitful. The Ustilago clade is another basidiomycete example to which this approach could be applied [27].
In this study we identified a number of novel RNAi components, three of which, Cpr2, Znf3, and Fzc28, had no functional domains or similarity to a known RNAi component that would have suggested they might be involved in an RNAi pathway. Notably, there are two known canonical RNAi components that we did not identify using this approach: Ago2 and Dcr2. In the first case, this is because Ago2 is not conserved across all of the Cryptococcus RNAi-proficient genomes, as it is missing from the H99 Cryptococcus neoformans reference genome. In the second case, Dcr2 is retained in C. deuterogattii, which may suggest either that Dcr2 already has an additional non-RNAi role in Cryptococcus, that it has acquired a second role during the loss of RNAi and speciation of C. deuterogattii, or that the entire RNAi pathway has not been lost. As a precedent for the first two hypotheses, in C. albicans, a noncanonical Dicer plays a role in snRNA processing [46].
Also notable is that our screen identified components that play a role in only the sex-induced pathway but not in mitotic silencing. Loss of these components suggests one of two hypotheses regarding the RNAi loss: either loss of RNAi began with the inactivation of SIS, and without SIS the evolutionary pressure to maintain the MIS pathway was no longer strong enough to prevent loss of core components, or alternatively, loss began with the core mechanistic components involved in the vegetative arm, and the pressure to maintain the specialized regulatory machinery of SIS vanished, allowing loss of both CPR2 and FCZ28. The latter case seems potentially more likely, as the loss of canonical RNAi components can allow relatively robust transposon movement even without undergoing the sexual cycle [30,47]. In addition, the loss of these SIS-specific components also provides an opportunity to elucidate the signaling processes connecting mating to induction of the RNAi pathway.
It is also interesting that the Cryptococcus species that has lost RNAi, C. deuterogattii, is the species causing an ongoing outbreak in the Pacific Northwest. The loss of RNA silencing is possibly associated with higher virulence in this strain, but because C. neoformans or C. deneoformans strains missing RNAi elements are not altered in virulence in a murine host [10], the loss of RNAi in the C. deuterogattii lineage may have instead had a longer-term impact on virulence trajectory. Indeed, loss of RNAi liberates transposons in Cryptococcus [10,30,47], which could provide adaptive benefits through the generation of increased genetic and phenotypic diversity.
We showed herein that Znf3 and Qip1 influence transgene induced silencing during mitotic growth and sexual reproduction. Deletion of these genes severely impaired silencing efficiency, even in unilateral crosses where only one parent was mutant. This phenotype is similar to rdp1Δ mutations, which abolish silencing during unisexual mating and impact silencing efficiency in unilateral bisexual crosses [10]. Given that Rdp1 is a major component of RNAi silencing, and that it is responsible for the initial steps generating dsRNAs, we propose that Znf3 also plays an important role in the pathway. Although the RNAi components Dcr1/2 and Ago1 are required for SIS, their deletion in a unilateral cross only lowers the silencing efficiency, indicating that their role may be redundant or largely complemented when one wild type nucleus is still present. Therefore, it is possible that Znf3 interacts with these proteins and may participate in the formation of the RISC complex. Znf3 is a large protein (~1515 aa), and so it could act as a scaffold to bring components of the RNAi pathway together in complex with the dsRNA substrate. Localization of Znf3 in P-bodies, where Dcr1/2 and Ago1 process the dsRNAs, may further support this hypothesis. Further, we provided evidence that Znf3 may play a role in protein translation or protein stability, which could be through a role as a scaffold in the P-bodies, or even as an RNA-binding protein through its zinc finger motifs.
Dumesic et al. showed that Qip1 localizes in both the cytoplasm and the nucleus during growth in rich media, and that it physically interacts with both Rdp1 and Ago1 [30]. We further confirmed that Qip1 localizes in the nucleus during vegetative growth; however, we showed that during sexual development Qip1 expression is highly induced and it is localized in P-bodies during sexual development. Possibly, Qip1 enhances the function of Rdp1 in the nucleus during vegetative growth, by participating in the generation of aberrant dsRNAs from repetitive loci. During sexual development, where transposon movement is highly induced, Qip1 may resume its conserved role in RNA processing in P-bodies, where it may interact with Ago1 and participate in the cleavage of the passenger strand. Interestingly, Qip1 appears to play a role in meiosis, as bilateral mutant crosses failed to yield recombinant progeny. Lee et al. showed that Qip is also essential for meiotic silencing and meiosis in N. crassa [48]. However, the two proteins are significantly different and C. neoformans Qip1 does not contain a canonical exonuclease domain, unlike Qip. This may suggest that either Qip1 plays a different mechanistic role in C. neoformans than Qip in N. crassa, despite a similar phenotypic outcome, or that Qip1 may contain an unrecognized exonuclease domain.
Sex-induced silencing is an efficient mechanism that protects the genome against mobile elements. Previous studies showed that ~5% of the Cryptococcus genome consists of transposons that cluster together in blocks and reside in both telomeric and centromeric regions on the chromosomes [36]. Transposon activation and movement may drive genome instability and phenotypic variation. Wang et al. found that transposons are transcriptionally induced specifically during sexual development, but they are silenced post-transcriptionally by the RNAi pathway [10]. These results suggest that transposons are derepressed during mating, which could increase the mutational burden of the progeny unless counteracted by the SIS RNAi pathway. During unisexual reproduction this mechanism could generate de novo genotypic and phenotypic variation in clonal populations and enable rapid adaptation to new environments. Thus, loss of the RNAi components may confer a beneficial advantage in clonal mitotic or sexual populations.
The strains and plasmids used in this study are listed in S1 Table. The strains were maintained in glycerol stocks at -80°C and grown on rich YPD media at 30°C (Yeast extract Peptone Dextrose). Strains with selectable markers were grown on YPD containing nourseothricin (NAT) and/or G418 (NEO). Uracil auxotrophic isolates were tested on both SD medium lacking uracil and synthetic medium containing 5-FOA (1 g/l). Mating assays were performed on 5% V8 juice agar medium (pH = 5 for C. neoformans or pH = 7 for C. deneoformans) or on MS (Murashige and Skoog) medium minus sucrose (Sigma-Aldrich). The mating cultures were incubated in the dark at room temperature for 1 week.
To visualize and isolate spores, strains of interest were co-cultured on solid V8 medium for 2 weeks at room temperature in the dark without parafilm. Basidiospores from the edges of the colonies were randomly isolated using a microdissection microscope equipped with a 25-μm microneedle (Cora Styles Needles ‘N Blocks, Dissection Needle Kit) as previously described [49]. Following germination the colonies were tested on YPD, YPD + NAT or NEO, SD-ura, and 5-FOA media. Genomic DNA was isolated using a CTAB protocol as previously described [50]. The presence of the SXI2a-URA5 transgene in the progeny was assessed by PCR using the primer pair JOHE16835/JOHE16836. Example gel images can be found in S3 Fig for the unilateral crosses depicted in Fig 2.
The gene of interest was disrupted using a standard overlap PCR approach described previously [17]. Briefly, the 5’ and 3’ flanking regions of the ZNF3, QIP1, and CPR2 genes were amplified from H99α genomic DNA, and the selectable markers NAT and NEO were amplified from plasmids pAI3 and pJAF1, respectively. The flanking sequences and the selectable markers were used to generate a full-length deletion cassette in an overlap PCR reaction with the flanking primers. The deletion cassettes were introduced into the H99α and JF289a strains by biolistic transformation [51]. Gene replacement via homologous recombination was confirmed by PCR and Southern hybridization. The primers used to generate the deletion mutants are listed in S2 Table. All deletions were constructed from at least two independent cultures, inoculated from different single isolated colonies of the parent strain, and independent mutants were isolated from different transformations.
For the fzc47Δ and fzc28Δ mutants, two independent deletions were available in the KN99 background from a recent deletion collection [31]. They were crossed with JF289a and spores were dissected to isolate segregants that inherited the deletion, the transgene array, and were MATa. These segregants were named strains SEC5, SEC6, SEC7, and SEC8.
To determine the cellular localization of the Znf3 protein, the mCherry protein was fused to the C-terminus of the protein under the control of the endogenous promoter using a standard overlap PCR approach. Briefly, 1 kb of sequence upstream of the start codon and 1 kb downstream of the stop codon were amplified from the wild type strain H99 genomic DNA using primers listed in S2 Table. The mCherry sequence fused with the NEO marker was amplified from plasmid pLKB25. The flanking sequences and the fluorescence marker were combined as a template for an overlap PCR reaction. The overlap PCR products were introduced into the H99α and JF289a strains by biolistic transformation. Transformants were analyzed by PCR and Southern hybridization.
To construct a plasmid encoding mCherry-Znf3, the mCherry protein was fused to the N-terminus of Znf3 under the control of the constitutively active GPD1 promoter. The genomic sequence of ZNF3 was amplified from H99α genomic DNA using primer pair JOHE37890/JOHE37891 and cloned into plasmid pLKB49 [52] digested with XbaI and PacI. Plasmid pMF81 was introduced into the H99α and JF289a strains by biolistic transformation and the transformants were screened by PCR and direct fluorescence microscopy. Stable transformants expressing mCherry-ZNF3 and QIP1-mCherry were also transformed via ectopic insertion with the pXW11 (GFP-DCP1) and pSL04 (GFP-NOP1) plasmids to visualize P-bodies and the nucleus, respectively.
To visualize mCherry-Znf3 and Qip1-mCherry together with GFP-Dcp1 and GFP-Nop1, the strains of interest were grown on YPD medium to determine localization during vegetative growth or mixed with the opposite mating-type strain on V8 medium for 24 hours to visualize the proteins during sexual reproduction. Briefly, cells were grown overnight in liquid YPD and washed with PBS. Cells were then counted and mixed in equal proportions of MATa and MATα and spotted on V8 pH5 medium. These plates were incubated in the dark at room temperature for 24 hours. Cells were then scraped from the plates and mixed into sterile water and placed on prepared slides covered with 1.5% water agar. Imaging was performed with a Zeiss Axio Imager widefield fluorescence microscope at the Light Microscopy Core Facility at Duke University. Analysis was performed using the Metamorph Premier software package.
For vegetative growth the cells were grown in 5 ml liquid YPD overnight at 30°C. The following day the cells were harvested, washed, frozen in liquid nitrogen, and lyophilized. Samples were kept at -80°C until analysis. To isolate RNA from mating assays, the desired α and a strains were grown in YPD liquid, washed with sterile water, mixed in equal amounts in eppendorf tubes, and a 1 ml suspension was spotted on V8 agar pH = 5 and incubated in the dark at room temperature for 24 hrs. The next day the mating cultures were harvested, washed with sterile water, frozen, lyophilized, and stored at -80°C. Total RNA was extracted using the RiboPure-Yeast Kit (Ambion) following the manufacturer’s instructions (Life Technologies #AM1926). Denaturing agarose gel electrophoresis and NanoDrop were used to assess quality and concentration of the RNA samples. The RNA was amplified using the Ambion® MessageAmp™ Premier RNA Amplification kit following the manufacturer’s instructions. cDNA was synthesized using AffinityScript reverse transcriptase (Stratagene), Cy3/Cy5 labeled, and hybridized to a C. neoformans microarray slide (Cryptococcus Community Microarray Consortium, Washington University, St. Louis, MO). Labeling and hybridization were conducted in the DNA Microarray Core Facility at Duke University. The slides were washed, scanned with a GenePix 4000B scanner (Axon Instruments), and processed with GenePix Pro (version 4.0). All microarrays were conducted in triplicate. GeneSpring software was used for statistical analysis employing Lowess normalization, reliable gene filtering, and ANOVA analysis (significance p<0.05).
Published reference genomes from H99, B3501A, JEC21, and WM276 were compared with R265 using FungiDB (http://fungidb.org/fungidb/) [11,36,37,53]. Genes were selected that were present in all four of the non-C. deuterogattii reference genomes but absent or greater than 50% different in length in the R265 genome. Positive hits were manually examined using synteny maps produced by FungiDB in order to confirm that orthologs had been correctly identified. The majority of the initial hits were false positives attributable to sequencing gaps or incorrect gene annotation. For the remaining hits, sequences from all five reference genomes were manually aligned using Clustal [54] to verify that sequence deletions had occurred. In addition, de novo assemblies of 53 VGII genomes from previously published data were used to validate that deletions of all of the components were conserved in the VGII lineage and not restricted to the R265 genome [32,55].
Estimation of Tajima’s D was performed using a custom Perl script and the Bio::PopGen:Statistics package of BioPerl [56]. Briefly, genomes were aligned and SNPs were called as previously described [32]. To avoid sampling bias, a representative genome was chosen from each clonal expansion in the sequencing dataset for a total of 17 individual lineages out of the original 53 genomes. SNPs were sampled over a range using VCFTools [57], alternate references were constructed using GATK [58], and regions were aligned using ClustalW [54]. These alignments were imported by Bio::AlignIO and sampled using Bio::PopGen:Statistics [59] within BioPerl. To sample the genome as a whole, 2 kb windows every 10 kb throughout the genome were chosen. Locations with missing data for an individual were discarded, resulting in a total of 1159 data points.
|
10.1371/journal.ppat.1005592 | Legionella pneumophila-Derived Outer Membrane Vesicles Promote Bacterial Replication in Macrophages | The formation and release of outer membrane vesicles (OMVs) is a phenomenon of Gram-negative bacteria. This includes Legionella pneumophila (L. pneumophila), a causative agent of severe pneumonia. Upon its transmission into the lung, L. pneumophila primarily infects and replicates within macrophages. Here, we analyzed the influence of L. pneumophila OMVs on macrophages. To this end, differentiated THP-1 cells were incubated with increasing doses of Legionella OMVs, leading to a TLR2-dependent classical activation of macrophages with the release of pro-inflammatory cytokines. Inhibition of TLR2 and NF-κB signaling reduced the induction of pro-inflammatory cytokines. Furthermore, treatment of THP-1 cells with OMVs prior to infection reduced replication of L. pneumophila in THP-1 cells. Blocking of TLR2 activation or heat denaturation of OMVs restored bacterial replication in the first 24 h of infection. With prolonged infection-time, OMV pre-treated macrophages became more permissive for bacterial replication than untreated cells and showed increased numbers of Legionella-containing vacuoles and reduced pro-inflammatory cytokine induction. Additionally, miRNA-146a was found to be transcriptionally induced by OMVs and to facilitate bacterial replication. Accordingly, IRAK-1, one of miRNA-146a’s targets, showed prolonged activation-dependent degradation, which rendered THP-1 cells more permissive for Legionella replication. In conclusion, L. pneumophila OMVs are initially potent pro-inflammatory stimulators of macrophages, acting via TLR2, IRAK-1, and NF-κB, while at later time points, OMVs facilitate L. pneumophila replication by miR-146a-dependent IRAK-1 suppression. OMVs might thereby promote spreading of L. pneumophila in the host.
| Of all intracellular pathogens, Legionella pneumophila show the highest genetic adaptation to the host. They are important human pathogens causing mainly pneumonia, but are also highly successful pathogens of water amoeba. To manipulate host cells for optimal intracellular replication, they express several hundred specific virulence factors, which they apply to their host cells by two different secretion systems. Another intriguing avenue that leads to the transfer of virulence factors to host cells—even over distance—is the transport via outer membrane vesicles (OMV) that are released by Legionella. We report here that after exposure to these vesicles, macrophages become permissive hosts for bacterial replication incapable of efficient pathogen defense. In addition, markers of inflammation that are induced by OMV pre-treatment do not further increase upon subsequent bacterial infection. This goes along with improved survival of infected macrophages despite the increased intracellular bacterial growth. We could attribute this effect to TLR- and NF-κB-dependent mechanisms of microRNA-146a induction and prolonged IRAK-1 depletion.
| Bacteria have developed numerous strategies to deliver virulence factors into their eukaryotic host cells. In close proximity, the transfer of virulence factors can take place by direct translocation into the host cytosol. Distant cells can be reached by the secretion of soluble proteases, lipases or toxins to the extracellular environment [1]. Additionally, Gram-negative bacteria developed the strategy of outer membrane vesicle (OMV) formation. OMVs are small, spheroid membrane vesicles of 10–300 nm in diameter, secreted during all phases of growth as well as in a variety of growth environments (liquid culture, solid culture, biofilms) [2, 3]. They transport diverse virulence factors, including proteins, adhesins, toxins and enzymes as well as non-protein antigens such as lipopolysaccharide (LPS), which is present on the outer leaflet of the OMV membrane [4]. They serve as a means of communication among bacteria, but can also be recognized and taken up by eukaryotic cells [4]. OMVs may influence the course of infection and the host immune response by presenting pathogen-associated molecular patters (PAMPs) and antigens to their respective host receptors [5]. For example OMVs derived from Clostridium perfringens induce cytokine secretion in macrophages, Borellia burgdorferi OMVs activate B cells, and vesicles secreted by Helicobacter pylori act on gastric epithelial cells [6–8]. In addition, OMVs transport active virulence factors of Gram-negative bacteria which gain access to the extracellular environment and can act over long distances, since the vesicular membrane protects the luminal cargo from extracellular host proteases and facilitates penetration into tissue [9–11]. OMVs have been found not only in close proximity to the site of bacterial colonization, but also in body fluids and distant organs [12]. Furthermore, they can mediate bacterial binding and invasion into host cells and cause cytotoxicity [4].
Legionella pneumophila (L. pneumophila) is a Gram-negative bacterium that replicates in freshwater amoebae [13]. When it enters the human lung, it primarily infects alveolar macrophages and can cause legionnaires’ disease, an acute fibrinopurulent pneumonia [14]. Macrophages have been described to respond to L. pneumophila infection by up-regulation of TNF-α, IL-6 and IL-1β [15]. These factors potently contribute to the establishment of a pro-inflammatory activation state, which is commonly referred to as classical macrophage activation (M1). In contrast, alternatively activated macrophages (M2), as characterized e.g. by CD206 (MRC1), do not show increased bactericidal potential. Upon infection, macrophages establish a transient activation state in the spectrum between these two canonical states, and this balance is decisive for the disposal of intracellular bacteria. L. pneumophila manipulates the host by secreting effector proteins into the cytoplasm via its type IV secretion system, leading to the blockage of phagosome-lysosome fusion, thus enabling L. pneumophila to recruit host organelles and to form its replication niche, the Legionella containing vacuole (LCV) [16, 17]. The inhibition of phagosome-lysosome fusion is achieved by effector protein secretion, and also by secretion of OMVs [18]. A proteomic analysis of L. pneumophila OMVs revealed about 70 proteins, many of them exclusively secreted via OMVs and associated with virulence function, e.g. persistence and spreading in the lung (fliC [19]), invasion (IcmK [20]) or intracellular survival and replication (ProA1 [21]). Additionally, L. pneumophila OMVs display proteolytic and lipolytic activity in vitro [22]. Experiments with human lung tissue explants (HLTE) demonstrated that OMVs can cause tissue damage [23]. Immune histological analysis of OMV-treated HLTE showed that OMVs were mainly bound to alveolar macrophages [24]. Since it is known that L. pneumophila OMVs are pro-inflammatory activators of macrophages and epithelial cells [22, 23], we aimed to analyze the impact of L. pneumophila OMVs on macrophage activation and their influence on a subsequent infection with L. pneumophila. As the innate immune response largely depends on the recognition of PAMPs, e.g. LPS, in their natural context, we speculate that L. pneumophila OMVs could have immunomodulatory functions, as already described for Brucella abortus OMVs [25] and OMVs derived from Porphyromonas gingivalis [26]. We investigated for the first time the influence of L. pneumophila OMV pre-treatment on a subsequent infection with L. pneumophila and the role of PAMP receptors and associated signaling pathways in OMV sensing.
As macrophages are the main target of L. pneumophila infection and replication [14], we tested the effect of L. pneumophila OMVs on macrophages. We incubated PMA-differentiated macrophage-like THP-1 cells with increasing doses of OMVs (0.01–25 μg/mL according to the protein concentration) for 24 and 48 h, respectively. The lowest OMV dose (0.01 μg/mL) was sufficient to induce a significant pro-inflammatory IL-8 release in THP-1 cells (Fig 1A) that increased time- and dose-dependently. Moreover, we observed a time- and dose-dependent release of IL-6 (Fig 1B) and IL-10 (Fig 1E) and a dose-dependent release of TNF-α (Fig 1C) and IL-1β (Fig 1D).
To gain further insight into the OMV amounts present under infection conditions, we generated bacteria-free cell culture supernatant of infected THP-1 cells (MOI 0.5, 24 h) by sterile filtration. The remaining membrane bodies were pelleted by differential centrifugation and probed for their LPS content. We detected LPS amounts which equal 0.04–0.08 μg of free OMVs. Given the fact that L. pneumophila OMVs are rapidly internalized by macrophages [24] and that we could only measure free OMVs after the incubation time, 0.1, 1, and 10 μg/mL OMVs were used for the following experiments.
No decrease in cell viability was observed by MTT assay for up to 96 h of OMV incubation (S1 Fig) in accordance to previous studies [22, 24].
Since macrophages sense L. pneumophila via LPS on the bacterial surface mainly by TLR2 [27, 28], we hypothesized that OMVs might stimulate macrophages in a similar way. We used murine bone marrow-derived macrophages (mBMDM) from wildytpe and TLR2/4-/- mice, which were treated with 0.1 or 1 μg/mL OMVs for 24 and 48 h, respectively. CXCL1, the murine functional homologue of IL-8, served as a measure for pro-inflammatory activation of mBMDM. Wildtype mBMDM secreted significant amounts of CXCL1 dose-dependently when incubated with OMVs (Fig 1F). Contrary to wildtype cells, TLR2/4-/- cells showed a rigorously reduced response to OMVs, as CXCL1 secretion was only slightly above the detection limit (0.2 ng/mL for both OMV concentrations). These results demonstrate that L. pneumophila OMVs are pro-inflammatory stimulators of macrophages and that macrophage activation occurs in a TLR2-dependent manner.
We investigated whether cellular contact with L. pneumophila OMVs has an impact on a following encounter with L. pneumophila. To this end, cells were pre-incubated with OMVs for 20 h, as they already responded with a pro-inflammatory response at this time point, and then infected with L. pneumophila. Cells were lysed at different time points to monitor bacterial uptake (2 h post infection (p.i.)) and replication (24 and 48 h p.i.) by CFU (colony forming unit) count. Infected cells without OMV pre-treatment served as a reference, and L. pneumophila replication in not pre-treated THP-1 cells is shown in S2A Fig. LPS/IFN-γ stimulation of THP-1 cells was used to induce a classical (M1) phenotype, which enables macrophages to efficiently kill L. pneumophila [29]. Lysis of cells 2 h p.i. did not reveal differences in bacterial uptake capacity in any of the tested conditions (Fig 2A and S2B Fig). Bacterial replication was significantly reduced at 24 h p.i. by 42% after LPS/IFN-γ pre-treatment. OMV pre-incubation showed similar tendencies; 0.1 μg/mL OMVs only modestly reduced L. pneumophila replication (10% reduction), whereas 1 μg/mL and 10 μg/mL of OMVs significantly reduced the bacterial load 24 h p.i. (23% and 33% reduction). At 48 h p.i., LPS/IFN-γ treated cells had progressively killed L. pneumophila (96% reduction). In contrast to LPS/IFN-γ, OMV pre-treatment had an enhancing effect on bacterial load (0.1 μg/mL: 88%, 1 μg/mL: 119%, 10 μg/mL: 118% more as compared to control). Additionally, we tested whether this mechanism also holds true with an L. pneumophila mutant which is not able to establish its replication vacuole in macrophages as it lacks a functional dot/icm system [17, 30]. To this extent, THP-1 cells were pre-stimulated with OMVs from WT L. pneumophila and infected with L. pneumophila ΔdotA mutant. Bacterial replication was analyzed by CFU assay. Interestingly, the highest dose of OMVs could enhance the replication of L. pneumophila ΔdotA by a factor of more than 14 at 48 h p.i. (S2C Fig).
Overall, pre-incubation of macrophages with OMVs leads to an altered bacterial replication.
As we detected a doubling in L. pneumophila replication after OMV pre-treatment, we aimed to gain insight into the fate of intracellular L. pneumophila. Immunofluorescence microscopy analysis was performed with not pre-treated and OMV pre-treated THP-1 cells which were subsequently infected with L. pneumophila for 48 h. After exposure of THP-1 cells to OMVs, an increased maximal cell diameter was observed (33.2 μm vs 19.1 μm). Not pre-treated THP-1 cells had one or two vacuoles per cell (Fig 2B and 2C), while in THP-1 cells with OMV pre-treatment, several LPS positive vacuoles were detected within the cytoplasm. Eighty percent of the not pre-treated cells had one or two vacuoles per cell. In contrast, only 56% of OMV pre-treated cells had one or two vacuoles (Fig 2D).
Contact with OMVs before an infection with L. pneumophila seems to increase the vacuole amount per cell in macrophages.
We further examined the expression of established markers of classically activated macrophages (M1; IL-1β, TNF-α, and IL-6) and alternatively activated macrophages (M2; CD206) [31, 32]. After infection of not pre-treated THP-1 cells, an increase in IL-1β expression was observed (Fig 3A). In contrast, while both pre-treatments (OMVs or LPS/IFN-γ) increased basal levels of IL-1β mRNA, they inhibited a further transcriptional induction above the initially induced level. Only the lowest dose of OMVs (0.1 μg/mL) that triggered a weak induction of IL-1β transcript allowed a minor additional increase after 48 h of infection. Similar results were obtained for TNF-α (Fig 3B) and IL-6 (Fig 3C). CD206, a marker for alternatively activated macrophages, was down-regulated after L. pneumophila infection of not pre-treated cells (Fig 3D). OMV pre-treatment resulted in a significantly reduced expression of CD206, which was comparable to infected and not pre-treated cells at 48 h p.i.. These experiments demonstrate a reduced M1 response of macrophages to an infection with L. pneumophila when pre-incubated with OMVs in comparison to naïve macrophages.
We then aimed to identify the OMV component mediating the observed differences in bacterial replication and M1 activation. OMVs were incubated for 5 min at 60°C, denaturing proteins but preserving OMV shape and membrane integrity [33]. Cells were pre-treated with heated OMVs and infected with L. pneumophila, and bacterial replication was determined by CFU assay. Here, the effect of 1 μg/mL OMVs was compared to 1 μg/mL heated OMVs. Interestingly, the heated OMVs did not significantly alter the bacterial replication at 24 h p.i. in comparison to not pre-treated cells (8% reduction; S3A Fig). At 48 h p.i., cells pre-treated with heat-denaturated OMVs were equally permissive to L. pneumophila replication as cells pre-treated with non-heated OMVs.
OMVs display LPS on their surface, which is a potent activator of macrophages. The LPS concentration of the OMV preparations was determined by Limulus amebocyte lysate test. OMVs contained 0.22 μg LPS per 1 μg protein. L. pneumophila has a unique LPS structure, which is sensed via TLR2 on the cell surface [27, 34]. Thus, a lipoteichoic acid and lipoprotein containing cell wall preparation (LTA), activating TLR2, or a Salmonella minnesota LPS preparation, activating TLR4, were used for alternative pre-treatment of THP-1 cells to test whether TLR activation alone is able to alter the response of macrophages to an infection with L. pneumophila. The bacterial load was significantly reduced by TLR2 or TLR4 activation before the infection (24 h p.i.: 19% or 12% reduction; 48 h p.i.: 39% or 60% reduction; S3B Fig). Furthermore, a TLR2 blocking antibody was used alone or in combination with OMVs to analyze the influence of TLR2 signaling—presumably activated by LPS—in the response of macrophages to L. pneumophila OMVs. Blockage of TLR2 signaling before infection did not significantly alter L. pneumophila replication, as determined by CFU assay. The combination of OMVs with the blocking antibody led to significantly more bacteria at 24 h p.i. (22% increase), but did not alter the bacterial load at 48 h p.i.. Incubation with a control antibody in combination with OMVs did not reveal significant differences.
To validate blocking efficiency of the TLR2 antibody we analyzed expression of macrophage activation markers. The combination of OMVs with the TLR2 blocking antibody led to a reduced mRNA induction of the M1 markers IL-1β (S3C Fig), TNA-α (S3D Fig) and IL-6 (S3E Fig), whereas the expression of the M2 marker CD206 was increased (S3F Fig).
Furthermore, L. pneumophila replication after OMV pre-incubation was analyzed in mBMDM from different genetic backgrounds. In WT mBMDM, OMV pre-treatment before infection induced the bacterial replication by a factor of 15 (S3G Fig). In mBMDM lacking TLR2 receptor bacterial replication was significantly reduced in comparison to WT cells. Additionally, we used TRIF/MyD88 double knockouts to block any TLR signaling in CFU assays. TRIF/MyD88-/- cells showed a response to OMV pre-treatment similar to TLR2-/- cells, as the replication was 7.5 fold reduced compared to WT mBMDM.
Taken together, inhibition of TLR2 activation blocked the initial bacterial killing in the macrophages, and bacterial replication was reduced at later time points. Therefore, TLR2 activation seems to be of equal importance for the initial stimulation of macrophages and the bacterial replication.
Since it has been shown for L. pneumophila OMVs that they may contain fliC [22], which is able to induce pro-inflammatory cell activation via TLR5 [35], we tested whether flagellin was inducing the observed effect. For this, cells were pre-exposed to OMVs obtained from ΔflaA L. pneumophila which were generated as the wildtype OMVs, but no differences could be observed in comparison to OMVs from wildtype L. pneumophila in the ability to render THP-1 cells more permissive for bacterial replication (S4A Fig).
As OMVs do not only contain LPS and proteins but also nucleic acids [36, 37], we assessed their influence on L. pneumophila replication in macrophages in more detail. OMVs were treated with RNases (RNase A and RNase III) or DNase in the presence or absence of Triton X-100. In the following CFU assay, disruption of OMV membrane prevented the increase in bacterial replication in THP-1 macrophages (S4B Fig). At 24 h p.i., Triton X-100 treatment of OMVs did not change bacterial replication in comparison to untreated control cells. This correlation was still observed upon 48 h p.i. while untreated OMVs increased the replication 4 fold. The exposure of OMVs to RNases or DNase alone did not alter L. pneumophila replication. Addition of Triton X-100 to RNases/DNases reduced L. pneumophila replication comparable to Triton X-100 alone.
Therefore we conclude that OMVs membrane integrity is essential for the response of macrophages to a following infection with L. pneumophila.
A key molecule downstream of TLR signaling is the transcription factor NF-κB [38]. As NF-κB activation is rapidly seen after L. pneumophila infection [39], we asked whether NF-κB is equally important in the OMV context. Cell fractionation experiments were performed after short time OMV incubation (30 min) and we could demonstrate that NF-κB subunit p65 is rapidly shuttling into the nucleus after OMV stimulation of THP-1 cells (S5 Fig).
We then inhibited the IKK complex with a small molecule inhibitor, which was used alone or in combination with OMV pre-treatment before THP-1 cells were infected with L. pneumophila, after testing that the IKK inhibitor does not reduce the cell viability (S6 Fig). CFU count revealed that NF-κB inhibition did not affect the early phase of L. pneumophila replication. Interestingly, it influenced the long-term replication, as observed by a reduction in bacterial replication by 68% at 48 h p.i. (Fig 4A). Combination of OMVs with the IKK inhibitor reduced the bacterial load to 50% at 48 h p.i., compared to DMSO control. To test whether NF-κB inhibition also affected the expression of M1/M2 markers, qPCR analysis was performed. For all analyzed M1 markers, the expression was reduced at the 0 h time point, i.e. by OMV pre-treatment (Fig 4B–4D). This reduction was not maintained throughout the course of infection in any of the observed cases. CD206, an M2 marker, showed an increased expression when NF-κB was inhibited (Fig 4E).
This implies that the canonical NF-κB pathway is necessary for early M1 activation of macrophages and intracellular long-term replication of L. pneumophila.
Since we observed a strong dependence on NF-κB signaling in the response of macrophages to L. pneumophila OMVs and since NF-κB has also anti-apoptotic targets [40], we analyzed the influence of OMVs on THP-1 cell viability in the course of infection with L. pneumophila. The amount of viable cells decreased during infection in not pre-treated cells (24 h p.i.: 16% reduction; 48 h p.i.: 45% reduction in comparison to untreated control cells) as a replication cycle of L. pneumophila in eukaryotic cells ends with the induction of host cell apoptosis concomitant with bacterial egress [41] (S7A Fig). However, in OMV pre-treated cells, the amount of viable cells was doubled at the lowest dose of OMVs and tripled at the highest dose at 48 h p.i.. We conclude that L. pneumophila OMVs improve cell viability throughout a following infection with L. pneumophila.
We investigated the expression of BCL2A1, an anti-apoptotic NF-κB target gene that has been demonstrated to be expressed in L. pneumophila infection [42]. We observed increased expression of BCL2A1 in OMV pre-treated cells compared to unstimulated cells (S7B Fig). At 24 h p.i., cells treated with higher doses of OMVs still had significantly more BCL2A1 compared to not pre-treated cells. The gain of OMV-treated cells over untreated cells was lost at 48 h p.i., when both pre-treated and not pre-treated cells had a high BCL2A1 expression.
Next we examined whether the activated anti-apoptotic signaling is responsible for the increase in bacterial replication after OMV pre-stimulation of macrophages. To this end, THP-1 cells were pre-treated with the pan-caspase inhibitor zVAD-fmk and then infected with L. pneumophila. The CFU assay did not reveal significant differences in bacterial replication after caspase inhibition (S7C Fig).
The expression of BCL2A1 together with a higher level of cell viability suggests that OMV exposure activates anti-apoptotic signaling via NF-κB in macrophages.
The strong dependence on intact NF-κB signaling for L. pneumophila replication and the involvement of TLR2 in OMV recognition led us to test for expression of the anti-inflammatory microRNA 146a (miR-146a), as it has been shown to be involved in the response to infections [43]. After OMV treatment of THP-1 cells, we observed a dose-dependent increase in miR-146a expression (Fig 5A). LPS/IFN-γ treatment also led to an induction of miR-146a, albeit to a weaker extent. When cells were additionally infected with L. pneumophila, the expression of miR-146a further increased over the time. Of note, L. pneumophila infected cells, which were not pre-treated with OMVs, showed a lower miR-146a expression. The induction of miR-146a could be reduced by blocking TLR2- and NF-κB-signaling (S8A Fig; 69% reduction and 48% reduction). Furthermore, WT mBMDM showed an induction of miR-146a after OMV stimulation (S8B Fig), which was significantly lower in TLR2-/- (35% less miR-146a). TRIF/MyD88 double knockouts did not express miR-146a after OMV stimulation.
The expression of the primary transcript of miR-146a (pri-mir-146a) is NF-κB dependent [44]. As we did not observe a further transcriptional induction of the M1 markers following Legionella infection of OMV pre-treated THP-1 cells (Fig 3), we asked whether the strong induction of miR-146a was due to transcriptional induction or processing of pri-mir-146a. OMV treatment of THP-1 cells strongly induced the primary transcript, which could not further be induced by L. pneumophila infection (Fig 5B). The not pre-treated cells showed a transcriptional induction of pri-mir-146a in response to L. pneumophila infection. Thus, transcriptional activation, rather than processing, seems to be responsible for the observed elevated miR-146a levels upon infection. OMV pre-treatment already induced the primary transcript of miR-146a that is then processed following L. pneumophila infection, leading to further increase of the mature miRNA.
The strong expression of miR-146a after OMV stimulation, concomitant with the dependence on active NF-κB signaling, suggest that the TLR-NF-κB pathway may play a role in the macrophage response to L. pneumophila OMVs. It has been shown that mycobacterial replication in macrophages is facilitated by miR-146a [45]. As we observed an increase in L. pneumophila replication following OMV stimulation together with an increase in miR-146a, we transfected miR-146a mimic/inhibitor or a corresponding control and infected the cells with L. pneumophila. Overexpression of miR-146a (Fig 5C) resulted in a 20% enhanced L. pneumophila number 24 h p.i. which was maintained throughout the experiment (Fig 5D). On the contrary, THP-1 cells transfected with miR-146a inhibitor had significantly less L. pneumophila replication than miR-146a overexpressing cells (24 h p.i.: 40%, 48 h p.i.: 41%; Fig 5D), arguing for an involvement of miR-146a in the response of macrophages to L. pneumophila OMVs.
One well-described target of miR-146a is the kinase IRAK-1 [43, 46]. Upon TLR2 activation, IRAK-1 mediates downstream signaling, undergoing phosphorylation, polyubiquitination and subsequent degradation [47]. Its 3’ UTR can be bound by miR-146a, suppressing its translation. We analyzed IRAK-1 protein expression in OMV stimulated THP-1 cells and observed a strong reduction in IRAK-1 expression after incubation with L. pneumophila OMVs (Fig 6A). The infection of OMV pre-treated cells further decreased IRAK-1 protein levels more than L. pneumophila infection alone. In contrast, combination of LPS and IFN-γ led to an increase of IRAK-1 on protein (Fig 6A). Furthermore, LPS/IFN-γ pre-treated cells showed a higher IRAK-1 protein expression than OMV pre-treated cells following infection. Quantification of western blots from three independent experiments is shown in S9A and S9B Fig.
As THP-1 cells responded differently to LTA or LPS as a first stimulus before infection with L. pneumophila, we analyzed IRAK-1 protein expression after treatment with these TLR ligands. The stimulation of THP-1 cells with LPS for 20 h led to the degradation of 19% of IRAK-1 protein and LTA reduced the IRAK-1 protein level by 17% (S9C and S9D Fig). In contrast, 10 μg/mL OMVs led to the degradation of 70% IRAK-1 protein, which was maintained throughout the incubation.
As a prolonged absence of IRAK-1 protein seems to be favorable for L. pneumophila replication, knockdown experiments for IRAK-1 were performed. The siRNA-mediated knockdown of IRAK-1 led to 50% reduced mRNA and protein levels in comparison to scramble (scr) transfected control cells at the time point of infection (Fig 6B), which remained stable during the time course of the infection (24 h p.i.: 57% reduction; 48 h: 57% reduction). At 24 h p.i., there was no difference in L. pneumophila replication compared to scramble control, whereas at 48 h p.i., there was 75% more L. pneumophila replication in IRAK-1 silenced cells (Fig 6C).
As IRAK-1 is an essential signaling molecule in TLR/IL-1R signaling, the expression levels of macrophage markers were measured after IRAK-1 silencing. THP-1 cells transfected with siIRAK-1 and infected with L. pneumophila showed less expression of markers for classically activated macrophages (IL-1β, IL-6, and TNF-α) in comparison to scramble control at 48 h p.i. (Fig 6D–6F), whereas the STAT6-dependent CD206, a marker for alternative activation, remained unchanged (Fig 6G).
These experiments show an involvement of IRAK-1 in the response of macrophages to L. pneumophila OMVs and that a decrease of IRAK-1 protein by OMV pre-treatment was associated with enhanced bacterial replication.
OMVs are potent pro-inflammatory stimulators of different cell types, carrying endotoxin and additional bacterial antigens [4]. OMVs from Acinetobacter baumannii induce cytokine secretion in epithelial cells [48], Clostridium perfringens OMVs stimulate the murine macrophage cell line RAW264.7 to produce G-CSF, TNF-α, and IL-6 [6]. L. pneumophila also produces OMVs that can be taken up by macrophages and furthermore induce tissue damage in human lung tissue explants [23]. Until now, the majority of OMV studies focused on direct OMV-mediated changes in their target cell/tissue. In this study, we investigated the impact of OMV pre-treatment on a following encounter of macrophages with L. pneumophila. THP-1 cells responded upwards of 0.01 μg/mL OMV with IL-8 secretion in a time- and dose-dependent manner. Additionally, the cells secreted IL-1β, IL-6, IL-10, and TNF-α dose-dependently. In contrast to THP-1 cells, stimulation of an alveolar type II epithelial cell line with L. pneumophila OMVs was reported to require much higher doses to induce pro-inflammatory activation (50 μg in contrast to 0.01 μg/mL) [22]. Human primary macrophages induce TNF-α release when exposed to 0.3 μg/mL L. pneumophila OMVs [24]. Here, we observed a response intensity to OMVs similar to what others observed when administering whole L. pneumophila [42]. Based on LPS-measurements, we assume that the experimentally used OMV doses resemble those that occur under infection conditions.
It has been described that LPS is present on the OMV surface [27, 28]. Therefore, we analyzed the importance of TLR signaling in macrophage activation by OMVs. mBMDMs from TLR2/4-/- mice showed a reduced CXCL1 secretion upon OMV exposure in comparison to cells from wildtype mice. Our results are in line with the results obtained by Jäger et al., who performed experiments with HEK293 cells overexpressing either TLR2 or TLR4 and observed a stronger response to OMVs when TLR2 was overexpressed [24]. Hence, LPS might be the most important stimulant on OMVs derived from L. pneumophila.
The two most important OMV components, LPS and proteins, are both essential for the initial activation of macrophages. The combination of both and the physiological context in which they are recognized seem to determine their effect on the host. This has recently been demonstrated for Pseudomonas aeruginosa OMVs [5]. Likewise, knockout mice for either TLR2 or TLR4 still show induction of inflammation in the lungs after being treated with P. aeruginosa OMVs [49].
Since L. pneumophila OMVs can contain the TLR5-activating fliC [22, 35], we pre-exposed THP-1 cells to OMVs obtained from ΔflaA L. pneumophila, but no differences could be observed in comparison to OMVs from wildtype L. pneumophila.
The infection of THP-1 with L. pneumophila led to an increased expression of classical (M1) macrophage activation markers. However, upon pre-exposure to OMVs, virtually no further increase was observed upon infection with viable bacteria. A reduced macrophage response to classical TLR stimuli has also been observed in a study with Brucella abortus OMVs, where THP-1 cells secreted less TNF-α and IL-8 in response to LPS, Pam3Cys and flagellin after OMV challenge [25]. Furthermore, Porphyromonas gingivalis OMVs are also reducing the TNF-α response to a second LPS stimulus [26]. We studied the impact of OMVs on a subsequent infection. We did not observe any differences in bacterial uptake capacity. However, 24 h post infection, OMV pre-treatment led to a reduction of bacterial replication in a similar range as LPS/IFN-γ pre-exposure. LPS/IFN-γ is an inducer of the M1 phenotype, which is capable of enhanced bacterial killing [29]. This restrictive effect on Legionella replication seems to depend in part on both OMV-transported proteins and TLR2 ligands, as it could be neutralized by heat exposure or treatment with a TLR2 blocking antibody, and mimicked by TLR2 activation. In addition, the early growth restriction could be reverted by inhibition of the canonical NF-κB pathway, because L. pneumophila needs the transcriptional induction of anti-apoptotic NF-κB target genes [42, 50]. In contrast to the early time point of infection (24 h p.i.), we observed a boosting effect of OMV pre-treatment on bacterial replication at 48 h time point, both in THP-1 cells and primary mBMDM of wildtype mice. Induction of bacterial replication after OMV treatment has also been observed in a study with Mycobacterium bovis in mice [51]. There, challenging mice with OMVs prior to an infection led to a doubling in bacterial load in the lungs and more spreading of the bacteria to the spleen. In contrast to the early restrictive effect on Legionella replication, the subsequently observed boost in replication seems not to rely on proteins or nucleic acids transported by OMVs and could not be mimicked by LPS or LTA preparations. However, the OMV-induced increase in L. pneumophila replication at the later time point was completely abolished by TLR2 knockout, TRIF/MyD88 double knockout, or inhibition of the canonical NF-κB pathway. Interestingly, OMV pre-treatment of macrophages before the infection with a dot/icm mutant of Legionella (ΔdotA), which is not capable of forming a LCV under normal conditions [17, 30], enables this mutant to replicate in THP-1 cells. Furthermore, the increased L. pneumophila replication was lost when OMVs were incubated with a membrane-permeabilizing agent, implicating that not only LPS on the OMV surface is critical for the observed effect, but also the context in which it is presented. Ellis et al. made similar observations with OMVs from P. aeruginosa [5]. LPS, which is important for binding of the OMV to the macrophage surface, or proteins, which are required for the internalization, are not sufficient alone. The observation that LPS aggregate size influences its internalization [52] and that vesicular LPS has a higher potency to activate macrophages [5], supports the concept that the three-dimensional structure of the vesicles impacts the response. Our results support the model that the response of inflammatory cells depends of the context in which LPS is presented [53].
L. pneumophila critically depends on establishing a fine balance between avoiding the immune response and successful infection. Pro-inflammatory cytokines and anti-apoptotic proteins are equally induced after L. pneumophila infection and are both dependent on p65 translocation into the nucleus [35, 42], which we observed after short time of OMV incubation. L. pneumophila itself can establish a phosphorylation of IκBα at ser-52 and ser-36 by LnaB and LegK1 which mimic eukaryotic serine/threonine kinases; this again leads to an IKK independent and robust induction of apoptosis antagonist genes as well as pro-inflammatory genes [54, 55]. As L. pneumophila activates NF-κB signaling itself and critically depends on this signaling as demonstrated by CFU assay with the IKK inhibitor, we conclude that macrophages might be a better replication niche for L. pneumophila when NF-κB signaling has already been induced by a first stimulus. There was no complete loss of bacterial replication observed upon IKK-inhibitor treatment, since SAPK/JNK and p38 signaling are still intact and also support L. pneumophila replication [56].
At a late time point during infection, OMV pre-treated cells showed better survival than not pre-treated macrophages. This might be a result of the activation of anti-apoptotic signaling via NF-κB, as we found increased BCL2A1 expression after OMV exposure. This BCL2 family member can be induced by GM-CSF, LPS, or TNF-α [57–59] via the transcription factor NF-κB [60]. The expression of anti-apoptotic BCL2A1 is in line with the survival advantage of OMV-treated cells when infected with L. pneumophila. However, caspase inhibition did not alter bacterial replication.
The L. pneumophila replication cycle in eukaryotic cells ends with host cell apoptosis and bacterial egress [41, 61], but here we observed an increase in bacterial replication after OMV incubation concomitant with a higher rate of cell viability. In addition to this, we found that OMV pre-treatment led to an increased number of vacuoles per cell, suggesting that exposure to OMVs rendered human macrophages more susceptible to intracellular L. pneumophila replication.
TLR and NF-κB signaling are tightly controlled not only by posttranslational modifications of involved proteins, but also by miRNAs that act on the 3’ UTR of mRNAs and lead to their translational repression [62]. L. pneumophila OMVs increase the levels of miR-146a by induction of its primary transcript, pri-mir-146a via TLR2 and NF-κB. This anti-inflammatory miRNA is involved in the intracellular replication of Mycobacterium bovis in macrophages [45]. We show that the artificial overexpression of miR-146a prior to L. pneumophila infection results in enhanced bacterial replication whereas the knockdown of this miRNA decreases the replication. This loss-of-function effect was not very strong in CFU assay, presumably because a functional knockdown of a microRNA as highly abundant as miR-146a is difficult to achieve. In addition, we are measuring a sum signal of only partly transfected cells. Moreover, the transfection of microRNAs can alter the loading of endogenous miRNAs into the microRNA processing machinery [63] and this could influence mRNA targets that we did not investigate.
The effect on bacterial replication of miR-146a in M. bovis infection has been linked to IRAK-1, which is a well-studied target of this miRNA [45]. IRAK-1 is essential in TLR and IL-1R signaling, as it mediates the downstream activation of NF-κB [47]. As we observed a long lasting IRAK-1 simultaneously with p65 nuclear translocation, we hypothesized that it is involved in the replication enhancing effect observed by OMV pre-treatment of macrophages. Knockdown of IRAK-1 by siRNA increased L. pneumophila replication and mimicked the effect observed with OMVs. Exposure of THP-1 cells to OMVs led to stronger effects on L. pneumophila replication than IRAK-1 knockdown, probably due to the moderate knockdown efficiency. Markers for classical macrophage activation were also reduced in siIRAK-1 transfected cells, similarly to OMV pre-treatment. It has been previously shown that LPS exposure can desensitize cells for a second LPS challenge by miR-146a upregulation, IRAK-1 degradation and translational repression of IRAK1 mRNA, which then results in a reduced cytokine response to the second LPS stimulus [64]. TLR2 or TLR4 activation alone did not mimic the full repertoire of L. pneumophila OMV exposure, probably because it results in much higher remaining IRAK-1 protein levels than OMV pre-treatment, even though the LPS doses were comparable (200 ng/ml S. minnesota LPS was used and 1 μg/mL OMV contains 220 ng/mL LPS). Contrasting OMV treatment, LPS/IFN-γ even induced IRAK-1 on protein level and led to reduced L. pneumophila replication in THP-1 cells.
Our results and the cited studies point towards an immunomodulatory action of OMVs, which might help the bacteria evading the host immune system. To our knowledge, our report is the first to describe the phenomenon of an OMV-induced replication advantage of L. pneumophila. Legionella has developed several sophisticated strategies to ensure its replication and spreading by evading the host immune system. They hijack key cellular processes to avoid lysosomal degradation after phagocytosis, and they interfere with host vesicular trafficking, ubiquitination and autophagy [17, 65, 66]. In doing so, L. pneumophila gains control over the host innate immune response. The activation of NF-κB signaling, by either OMVs or whole bacteria, leads to secretion of pro-inflammatory cytokines, which likely results in the recruitment of new potential host cells to the site of infection, which then can be infected. The increased infection of macrophages in the presence of OMVs might therefore be an important pathogenic strategy of L. pneumophila.
Taken together, this study has demonstrated for the first time that OMVs directly influence the course of Legionella infection by activating macrophages in a pro-inflammatory way and at the same time promoting bacterial replication (Fig 7). This property seems to depend on TLR2- and NF-κB-dependent miR-146a upregulation and consequently prolonged IRAK-1 depletion. Thereby, OMVs could facilitate replication and spreading of L. pneumophila in the human lung.
RPMI-1640, FCS, phalloidin Alexa Fluor 546 and goat anti-mouse Alex Fluor 488 were obtained from LifeTechnologies (Darmstadt, Germany). PBS was from Biochrom GmbH (Berlin, Germany). PMA was purchased from Sigma-Aldrich Chemie GmbH (Taufkirchen, Germany). LTA was from InvivoGen (Toulouse, France). DAPI was from AAT Bioquest (Sunnyvale, USA). IKK inhibitor XIII was obtained from EMD Millipore Corporation (Billerica, USA). zVAD-fmk was obtained from AdipoGen (Liestal, Switzerland) Salmonella minnesota LPS was from Enzo Life Sciences (Lörrach, Germany). Human recombinant IFN-γ was obtained from PromoKine (Heidelberg, Germany). TLR2 blocking antibody was acquired from eBioscience (T2.5, San Diego, USA). Anti-LPS antibody (L. pneumophila specific; ABIN235748) was obtained from antikörper-online.de (Aachen, Germany). Anti-IRAK-1 antibody was purchased from Cell Signaling (4359S), anti-tubulin antibody was from Santa Cruz (sc-5286) as well as p65 antibody (sc-372) and β-actin antibody (sc-1616). All chemicals used were of analytical grade and obtained from commercial sources.
Legionella pneumophila strain Corby wildtype was grown on buffered charcoal-yeast extract (BCYE) agar at 37°C for three days. For OMV preparation, L. pneumophila were transferred from BCYE agar plates into sterile yeast extract broth (YEB) medium at a density of 1x109 bacteria per mL. Bacteria were incubated in a shaking incubator (MaxQ 6000, Thermo Fisher Scientific, Karlsruhe, Germany) at 37°C until they reached early stationary phase. Pure L. pneumophila cultures were then spun down three times (4,500 x g, 15 min, 4°C; Multifuge X3R, Thermo Fisher Scientific). To remove remaining bacteria, the supernatant was passed twice through a 0.22 μm sterile filter. The supernatant was then ultracentrifuged (100,000 x g, 3 h, 4°C). After washing the obtained OMV pellet with sterile PBS to remove contaminating free protein, it was again ultracentrifuged and then resuspended in sterile PBS and stored at -20°C. The protein content of the preparations was determined by Pierce BCA protein assay kit (Thermo Fisher Scientific), and equal protein amounts were used for cell stimulation. For protein denaturation, OMVs were incubated at 60°C for 5 min as previously described [33].
The human monocytic cell line THP-1 was obtained from American Type Culture Collection (Rockville, MD, USA) and cultivated in RPMI-1640 with supplements and 10% FCS in a humidified incubator at 37°C and 5% CO2. THP-1 cells were differentiated into a macrophage-like phenotype using 20 nM PMA for 24 h and then plated at the desired density. For all infection experiments, differentiated THP-1 cells were infected with WT L. pneumophila with a multiplicity of infection (MOI) of 0.5. For infection experiments of mBMDM L. pneumophila ΔflaA were used at an MOI of 0.5.
Murine bone marrow derived monocytes were isolated from tibiae and femora of wildytpe, TLR2-/-, TLR2/4-/- or TRIF/MyD88-/- mice (C57BL/6N). For each experiment, cells were differentiated into macrophages in the presence of M-CSF (macrophage colony-stimulating factor). After 72 h of incubation, M-CSF was added again, and cells were incubated for additional 48 h. Differentiated cells were detached and re-plated at the desired density in the presence of GM-CSF. After 24 h, cells were used for experiments.
Transfection of THP-1 cells with miRNA mimics or inhibitors was performed with siPORT NeoFX (Invitrogen) according to the manufacturer’s instructions. mirVana mimics and inhibitors for miR-146a and corresponding controls were purchased from Thermo Fisher Scientific. Following the manufacturer’s protocol, siRNA transfection was performed by using Lipofectamine 2000 (Thermo Fisher Scientific). Silencer select siRNA targeting IRAK-1 mRNA and a corresponding scramble control were purchased from Thermo Fisher Scientific.
Differentiated THP-1 or mBMDM were stimulated with different doses of OMVs (0.01–25 μg/mL as determined by the protein concentration of the OMV preparation). Cells were incubated with OMVs for 24 or 48 h. In all infection experiments OMV pre-incubation was carried out for 20 h and then the infection followed for up to 48 h. LPS (200 ng/mL) alone or in combination with IFN-γ (200 ng/mL) or LTA (1 μg/mL) were alternatively used to pre-stimulate THP-1 cells for 20 h before the infection with L. pneumophila. To assess the influence of TLR2 signaling, a TLR2 blocking antibody was added to the cells 90 min before OMV stimulation at a final concentration of 20 μg/mL. NF-κB inhibition was achieved by incubation with 1 μM IKK XIII inhibitor prior to treatment. The influence of nucleic acids in the OMVs on L. pneumophila replication was analyzed by digestion of OMVs with RNase A (0.2 μg/μL) and RNase III (0.02 U/μL) in combination or DNase I (0.004 U/μL). OMVs were incubated for 1 h at 37°C with the enzymes alone or in combination with 0.3% Triton X-100 to permeabilize the OMV membrane (final concentration in the cell culture well: 0.75 ‰).
To analyze bacterial replication in THP-1 cells, cells were pre-stimulated with rising doses of OMVs or left untreated and subsequently infected with L. pneumophila. To quantify the bacterial load, cells were lysed at indicated time points with saponin (0.1% in the supernatant) and different dilutions of the lysates were streaked on BCYE agar plates. After three days of incubation at 37°C, L. pneumophila colonies were counted and the bacterial load was calculated.
IL-8 or CXCL1 from the cell-free supernatant of THP-1 or mBMDM was analyzed with commercial ELISA kits (IL-8: OptEIA, BD Biosciences, Heidelberg, Germany; CXCL1: DuoSet, R&D, Minneapolis, USA). All other secreted cytokines (IL-1β, IL-6, IL-10, TNF-α, MCP-1, GM-CSF) were measured with a Luminex Assay (R&D) in a Bio-Plex Magpix (Luminex Corporation, Austin, USA) according to the manufacturer’s instructions.
For analysis of gene expression, total RNA isolation was carried out by phenol-chloroform extraction with Isol RNA lysis reagent (5Prime, Hamburg, Germany). For the detection of pri-mirnas, purified RNA was DNase I digested (Roche, Mannheim, Germany). After reverse transcription (High-Capacity RNA-to-cDNA kit or TaqMan miRNA reverse transcription kit, both Thermo Fisher Scientific), quantitative real-time PCR was performed in a ViiA7 (Thermo Fisher Scientific) with Fast SYBR Green Master Mix (Thermo Fisher Scientific) and specific primer pairs. miRNA expression was analyzed with TaqMan assays detecting miR-146a, pri-mir-146a, pri-mir-16-2 and RNU48 (Thermo Fisher Scientific).
BCL2A1: fwd: 5’-GGCCCACAAGAAGAGGAAAATG-3’, rev: 5’-GGAGTGTCCTTTCTGGTCAACA-3’
CD206: fwd: 5‘-CAGCGCTTGTGATCTTCATT-3‘, rev: 5‘-TACCCCTGCTCCTGGTTTTT-3‘
IL-1β: fwd: 5‘-AGCTCGCCAGTGAAATGATGG-3‘, rev: 5‘-CAGGTCCTGGAAGGAGCACTTC-3‘
IL-6: fwd: 5‘-AATTCGGTACATCCTCGACGG-3‘, rev: 5‘-TTGGAAGGTTCAGGTTGTTTTCT-3‘
IRAK-1: fwd: 5’-TGAGGAACACGGTGTATGCTG-3‘, rev: 5‘-GTTTGGGTGACGAAACCTGGA-3’
RPS18: fwd: 5’-GCGGCGGAAAATAGCCTTTG-3‘, rev: 5‘-GATCACACGTTCCACCTCATC-3‘
TNF-α: fwd: 5‘-GCTGCACTTTGGAGTGATCG-3‘, rev: 5’-TCACTCGGGGTTCGAGAAGA-3‘
Cells were seeded into microtiter plates and incubated with rising amounts of OMVs before infection with L. pneumophila. At indicated time points, 10 μL thiazolyl blue tetrazolium bromide (MTT, 5 mg/mL; Sigma-Aldrich Chemie GmbH) was added to the cells and incubated for another 2 h at 37°C. Medium was removed and replaced with EtOH:DMSO (1:2). After shaking, absorption was measured at 570 nm.
THP-1 cells were seeded on coverslips and infected with L. pneumophila for 48 h with or without OMV pre-incubation. After 15 min fixation with 4% paraformaldehyde, slides were washed three times with PBS and permeabilized with 0.2% Triton X-100 in TBS. Blocking was carried out with 10% FCS for 90 min. Incubation with α-LPS (1:500, in blocking solution) was followed by α-mouse (1:1000, in blocking solution) in combination with DAPI (1:2000) and phalloidin Alexa Fluor 546 (1:40). Stainings were analyzed on a fluorescence microscope (Axio Vision, Zeiss, Jena, Germany).
All animals were handled according to national and European legislation, namely the EU council directive 86/609/EEC for the protection of animals. The performed protocols were approved by the responsible animal ethics committee (Philipps-University Marburg; permit number: EX-22-2013).
Data are shown as mean values ± SEM for at least three biologically independent experiments. Prism 5 (GraphPad, La Jolla, USA) was used. The non-parametric Mann-Whitney test was performed for unpaired samples. P-values ≤ 0.05 were considered statistically significant. If not indicated otherwise, tests were performed vs. corresponding control (*).
|
10.1371/journal.pbio.1001261 | Superhelical Architecture of the Myosin Filament-Linking Protein Myomesin with Unusual Elastic Properties | Active muscles generate substantial mechanical forces by the contraction/relaxation cycle, and, to maintain an ordered state, they require molecular structures of extraordinary stability. These forces are sensed and buffered by unusually long and elastic filament proteins with highly repetitive domain arrays. Members of the myomesin protein family function as molecular bridges that connect major filament systems in the central M-band of muscle sarcomeres, which is a central locus of passive stress sensing. To unravel the mechanism of molecular elasticity in such filament-connecting proteins, we have determined the overall architecture of the complete C-terminal immunoglobulin domain array of myomesin by X-ray crystallography, electron microscopy, solution X-ray scattering, and atomic force microscopy. Our data reveal a dimeric tail-to-tail filament structure of about 360 Å in length, which is folded into an irregular superhelical coil arrangement of almost identical α-helix/domain modules. The myomesin filament can be stretched to about 2.5-fold its original length by reversible unfolding of these linkers, a mechanism that to our knowledge has not been observed previously. Our data explain how myomesin could act as a highly elastic ribbon to maintain the overall structural organization of the sarcomeric M-band. In general terms, our data demonstrate how repetitive domain modules such as those found in myomesin could generate highly elastic protein structures in highly organized cell systems such as muscle sarcomeres.
| The contraction and relaxation cycles of active muscles generate substantial mechanical forces, both axially and radially, that place extraordinary stress on the molecular structures within the muscle fibers. These forces are sensed and buffered by unusually long and elastic filament proteins with highly repetitive domain structures. Myomesin is one such repetitive filament protein that is thought to form bridges between the main contractile filaments of the muscle, providing the muscle structure with resistance in the radial dimension. To investigate how the repetitive structure of myomesin contributes to muscle elasticity, we determined the overall architecture of its complete repetitive domain array using a combination of four complementary structural biology methods. Our study reveals a long, dimeric tail-to-tail filament structure folded into an irregular superhelical coil arrangement of almost identical domain modules separated by short linkers. When we applied tension to these myomesin filaments, we found they could stretch to about 2.5 times their original length by unfolding these linkers, and then return to their original state when the tension was removed. Our findings explain how myomesin might adapt its overall length in response to the changing dimensions of the contracting and relaxing muscle, so acting as a highly elastic ribbon that maintains the overall structural organization of the muscle fibers. More generally, these findings demonstrate how repetitive domain modules, such as those in myomesin, can provide elasticity to highly organized biological structures.
| Striated myofibrils are found in skeletal and cardiac muscle cells and represent a highly organized cellular system for studying how active force can be generated while the overall structural organization of the underlying sarcomeric units is maintained. The principal protein components of myofibrils are large longitudinal filaments that include actin (thin filament), myosin (thick filament), titin, and nebulin [1]. These filaments form a well-established striated pattern of distinct zones, with the M-band at the center [2]. On activation, both substantial axial and radial forces are generated within the overall sarcomere structure [3]. To maintain a constant sarcomere volume under defined physiological conditions, these forces can lead to changes in both radial and longitudinal contour dimensions of the sarcomere. Under typical tension conditions, M-band-associated thick filaments can substantially move away from the sarcomeric center by 0.1 µm or more, which can lead to M-band-induced instability of the sarcomere [4]. Because of the presence of a stiff Z-disk architecture at the sarcomeric periphery, the amount of movement decreases with the overall sarcomere length so that the resting tension stays constant. In cardiac muscles, elastic M-band motions are thought to correlate with heart beat rate [5], rendering investigations of the underlying molecular parameters highly relevant to heart and skeletal muscle research.
To ensure the restoration of sarcomere integrity on activation, there are two principal structural compartments with elastic properties. The first section is defined by the I-band segment, which is situated between the stiff and highly interconnected Z-disk at the sarcomere periphery and the more dynamic central A-band and M-band [1],[6]–[9]. The second site for molecular elasticity is within the M-band, in which so-called M-bridges transversely connect thick filaments with each other and with titin filaments [2],. At the molecular level, M-bridges are thought to be primarily composed of myomesin (MYOM1), which is universally expressed, and two related isoforms, MYOM2 and MYOM3, which show tissue-specific expression [12]. The three proteins share a common domain topology that is characterized by a unique N-terminal myosin-binding domain, followed by an array of fibronectin type III (Fn-III) domains and immunoglobulin-like (Ig) domains. In addition, they are capable of forming C-terminal tail-to-tail homodimers, as shown for MYOM1 and MYOM3 [12],[13]. Correct M-band localization of myomesin depends on the presence of the C-terminal M-band region of the titin filament [14]; myomesin, titin, and the filament protein obscurin localize to the same region [14], assigning myomesin a central role in maintaining the M-band architecture. The crucial importance of myomesin for sarcomere integrity has been shown by studies suppressing MYOM1, leading to disintegration of obscurin in the M-band [15]. However, in the absence of ultrastructural data at a molecular resolution, the overall organization of M-bridges, and any associated requirements for molecular elasticity, has remained largely unknown to date.
To address these questions we have made use of a previous prediction suggesting that the entire C-terminal part of the myomesin filament consists of an array of repetitive Ig domains followed by exposed α-helical linkers [16]. Here, we report the complete structure and extent of longitudinal elasticity of the entire C-terminal tail-to-tail myomesin filament My9–My10–My11–My12–(My13)2–My12′–My11′–My10′–My9′ (My9–My13). It folds into a superhelical architecture with almost identical Ig domain/helix modules and an estimated overall length of 360 Å in the absence of external tension. When stretched by low molecular forces <30 pN, myomesin can be extended reversibly by about 2.5-fold its original length, demonstrating that this filament can adapt its overall length to the changes in dynamic M-band dimensions that have been observed in operating myofibrils [2],[3].
First, we determined the overall architecture of the complete myomesin domain array, including Ig domains My9–My10–My11–My12–My13, from crystal structures of a total of three filament fragments: the My9–My11 triplet, determined at 2.5 Å resolution; the My10–My11 doublet, at 1.9 Å resolution; and the homodimeric My11–My13 triplet at 3.5 Å resolution (Figures 1 and S1; Table 1). For further structural analysis and comparison, we also used the published crystal structure of the My12–My13 fragment [16]. Although the structure-based sequence similarity of the five Ig domains is generally below 25%, all Ig domains except My13 belong to different sub-classes of the C-set type Ig domain topology (Figures 2A and S2). A detailed comparison of the individual Ig domains My9, My10, My11, My12, and My13 is provided in Text S1.
As the available independent structures of helix-connected Ig domain doublets show only limited variation in terms of domain/domain arrangements (Figure 3), a reliable overall structural model of the complete C-terminal myomesin Ig domain array My9–My13 can be generated (Figure 1). This reveals a tail-to-tail filament structure with an overall length of 360 Å and a pronounced zigzag-type arrangement within the central, C-terminal myomesin dimerization module (My13)2, followed by a superhelical coil arrangement towards the two symmetrical distal ends. The right-handed superhelix is defined by almost constant twist angles of neighboring My domains of 26–27°, except the arrangement of My12–My13, which has an average twist angle of 68° and thus appears to be less regular than preceding parts My9–My10–My11–My12 of the filament (Figures 1, 3, and S3C).
Remarkably, all five Ig domains of the myomesin filament are connected by α-helices that are identical in terms of orientation with respect to the preceding Ig domain, and they form a substantial, structurally highly conserved Ig domain/helix interface with an area of 350–600 Å2 (Figure 2B). These interfaces involve loops that form the C-terminal tip region in each Ig domain, and residues from the first three helical turns form various specific interactions that shield the N-terminal part of each α-helix from being completely exposed. This conserved Ig domain/helix module, found in all four myomesin My domains that are followed by α-helices (My9, My10, My11, and My12), defines a new type of Ig domain topology (Figures 2B and S3). We refer hereafter to this as the “IgH” segment, which to our knowledge has not been found in any other muscle filament protein with repetitive Ig or Fn-III domains.
By contrast, we have observed neither a common interface nor a similar orientation for any of the My–My domain-connecting helices and subsequent Ig domains My10, My11, My12, and My13 (Figures 1 and S3B). As the overall geometry of the IgH segments is rigid, the limited variability in terms of arrangements of neighboring My domains originates from the diverse helix/My domain connections. Whereas long six-turn helices are found in the My10–My11 and My12–My13 connecting segments, the corresponding helices in the other two connecting segments, My9–My10 and My11–My12, are slightly shorter, with about four turns each (Figures 1 and 3). Because of the smaller length of their linkers, My9–My10 and My11–My12 show limited direct interactions in their respective Ig doublets, whereas in the other two doublets, My10–My11 and My12–My13, the neighboring My domains are too far away from each other to form an interface.
The complete filament structure of the My9–My13 dimer therefore is defined by an arrangement of nine rigid bodies, comprising eight Ig domains and the central My13 dimer. A systematic distance analysis of the respective centers of gravity reveals rather narrow average distance windows, specifically for next neighbors (n, n+1, 50 Å) as well as for third and fourth next neighbors (n, n+3, 116 Å; n, n+4, 153 Å), regardless of whether they are intramolecular (within one My9–My13 monomer) or intermolecular (generated from Ig domains across the My9–My13 dimer) (Figure 4A and 4B). This distinct distribution demonstrates the highly regular arrangement of domains within the complete My9–My13 filament dimer.
To independently validate the crystallographic model of the dimeric C-terminal myomesin filament, we used negative stain transmission electron microscopy (EM) (Figures 5A and S4). As it was difficult to define the precise filament ends when using a native My9–My13 myomesin fragment—probably because of limited distal flexibility coupled with the small size of each Ig domain—we fused the N-termini of each of the two myomesin My9–My13 filament protomers with maltose binding protein (MBP), which has about four times the molecular mass of a single Ig domain. This engineered version of the C-terminal myomesin filament revealed elongated molecular images about 500 Å long and 50 Å wide (Figure S4). The ends of the molecular images feature large globular densities about 50 Å in diameter, consistent with the molecular structure of MBP. Individual molecular images were aligned, classified, and averaged. A number of the resulting class averages showed 2-fold rotational symmetry, allowing the generation of averaged images in which the terminal MBP fusion and the myomesin My9–My13 filament could be unambiguously identified (Figure 5A). In these averages, the dimeric myomesin My9–My13 filament has a length that is similar to that derived from the X-ray composite model. Moreover, we could not detect any side-to-side oligomerization. This finding is further supported by complementary biophysical data for My9–My13, which do not indicate any further oligomerization (Figure S5).
The central region of the averaged EM image (Figure 5A) associated with the myomesin filament is in excellent agreement with the crystallographic My9–My13 model. The comparison suggests that the five areas of highest protein density are associated with the central C-terminal (My13)2 dimerization module and two flanking Ig doublets, My11–My12 and My9–My10, on both sides of the filament. In the crystal structures used to generate the crystallographic My9–My13 filament model, these two doublets are connected by shorter helices than the other myomesin/Ig domain doublets (Figure 1), which leads to an appearance of Ig domain tandems when the respective low-resolution-filtered projections are displayed (Figure 5B).
To further characterize the overall structure of the complete My9–My13 myomesin filament in solution, we used small angle X-ray scattering (SAXS) (Figure S6). The overall parameters indicate that the particle is dimeric and that the overall shape of My9–My13, in terms of simple bodies, is best represented by a long cylinder 370 Å in length (Table 2). The resulting low-resolution shape of wild-type My9–My13 reconstructed ab initio reveals an extended coiled conformation, in agreement with both the EM and crystallographic models (Figure 5).
The computed distance distribution p(r) displays a series of maxima at distances of 60 Å, 118 Å, and 165 Å (Figure 4C), which is characteristic of elongated particles with periodic domain arrangements [17]. A comparison with the domain/domain arrangement analysis of the composite My9–My13 structure (Figure 4B) indicates that the additional peaks in the experimental p(r) most likely arise from first, third, and fourth Ig domain neighbors. We further computed a p(r) function for this composite My9–My13 crystallographic model, and indeed all maxima positions agree well with the experimental SAXS data (Figure 4C). However, we observed reduced peak sharpness in the experimental SAXS distance distribution p(r) compared with the function computed from the crystallographic model (Figure 4C). This suggests a limited flexibility in the Ig domain/domain arrangements in the dimeric My9–My13 filament, in agreement with our comparison of those Ig domain tandems for which multiple crystal structures are available (Figure 3). To take this properly into account, we applied the Ensemble Optimization Method (EOM) [18] to generate an optimized ensemble that yielded an improved fit, and the resulting p(r) function displayed peak heights proportional to the experimental p(r) (Table 2; Figure S6B).
When using My9–My13 variants in which single proline residues were introduced in two helical linkers, My11–My12 (K1457P) and My12–My13 (Y1551P), to disrupt their helical conformation (Figure 2), most of the peaks in the p(r) function were lost (Figure 4C). This was accompanied by a substantial decrease in the maximum particle size (Table 2). The data from the myomesin mutants thus show that the structural integrity of these helical linkers is essential to establish the observed defined and rigid myomesin filament architecture.
The repeated structural pattern of at least four IgH modules in the C-terminal part of myomesin strongly suggests an elastic role for this segment. To test this hypothesis, we designed atomic force microscopy (AFM) experiments using an approach that was established to measure the level of molecular elasticity in the C-terminal myomesin tandem My12–My13 [19]. We created a modified version of the myomesin My9–My13 filament dimer by fusing polyhistidine tags to each of the two distal N-terminal My9 domains, and adsorbed this to a Ni-NTA-coated surface. We then probed the elasticity of the myomesin filament by adsorbing the protein chain to an AFM tip and recording the applied force versus the molecular extension. In the two sample traces shown (Figure 6A), a consistent saw-tooth pattern with a regular spacing can be observed, reflecting the sequential unfolding of individual Ig domains in the myomesin filament. The increase in contour length (ΔL) on unfolding can be measured with nanometer precision as 29.7 nm (Figure 6B, left) [20], which is in good agreement with the values expected for the unfolding of individual Ig domains.
At low extension and low force, however, a plateau in the force–extension curves can be observed (Figure 6A, arrows). No such plateau has been observed for Ig domain arrays in other systems investigated to date, suggesting a novel mechanism of unfolding. At high pulling velocities (typically 1 µm s−1), further substructures within these plateaus could not be resolved. Therefore, we performed experiments with decreased pulling velocities (10 nm s−1) in the critical extension range up to 50 nm, followed by an accelerated pulling velocity of 100 nm s−1 above an extension of 50 nm to reduce the overall experimental time. A sample trace shows that the unfolding of four Ig domains can be observed (Figure 6C, lower panel). Owing to the very low pulling velocity, the unfolding force of the first myomesin Ig domain is low and almost coincides with the plateau force. The plateau preceding this first unfolding event contains substructures, which can be seen when this region is enlarged (Figure 6C, central panel), and four distinct peaks can be resolved. A contour length histogram (Figure 6C, upper panel) reveals an average peak-to-peak distance of 6 nm. These values are in good agreement with the expected contour length increase upon unfolding of individual α-helical segments between the Ig domains. Rapid transitions can be observed between the peaks, indicating that α-helix unfolding is a rapid process close to equilibrium. This is further supported by repeated stretch (red) and relax (blue) curves within an individual plateau (Figure 6D). Both stretching and relaxing cycles exhibit the characteristic plateau. As soon as the molecule is relaxed, the helices contract back and thus act as truly elastic springs at forces around 30 pN. In summary, our AFM experiments suggest that important molecular elasticity components of myomesin originate from the helical connecting linkers between the Ig domains.
As the axial translational components for polypeptides in an α-helical conformation (1.5 Å per residue) and extended conformation (3.6 Å per residue) are known, the increase in length of a 20-residue helix on unfolding should be about 40 Å. The observed peaks at intervals of up to 60 Å when applying low pulling velocities in AFM experiments therefore match our calculations, when the straightening of substantially tilted helix orientations, as observed in the composite My9–My13 X-ray model (Figures 1 and 4), is also taken into account. Extending these calculations to the complete dimeric C-terminal myomesin filament, its estimated length with unfolded helical linkers is about 860 Å, which is almost 2.5-fold its original length, with an estimated increase in length of about 500 Å under low external forces (Movie S1).
The M-band is believed to be the key strain sensor in muscle sarcomeres, and important signaling events at or near the M-band may be regulated by mechanical forces originating from this region [1],[2],[21]. Members of the myomesin protein family have been identified as key bridging molecules that connect the major sarcomeric filaments in this central sarcomere segment [2]. Previous immuno-EM studies indicate that the overall orientation of myomesin has both longitudinal and transversal components, the latter being associated with the so-called M-bridges [10],[11],[22]. Although the precise orientation of different filament parts of myomesin in the sarcomere is unknown, available experimental data lead to a model in which the dimeric link via the C-terminal My13 domain in myomesin filaments is across the central M-line [13]. These data also assign myomesin an additional role in compensating the observed unbalanced filament movements with respect to the central M-line, which consequently requires a large potential of molecular elasticity.
To quantify the amount of molecular elasticity in myomesin, we applied AFM, an approach that has already been extensively used to estimate the elastic properties of the giant filament protein titin [23]. Titin filaments, however, assemble in parallel oligomeric bundles, propagating both from the Z-disk and from the M-band, and observed changes in persistence length have been estimated to be proportional to the square of the number of filaments [7],[24]. Therefore, single molecule stretching data may deviate from titin elasticity mechanisms in vivo. Accordingly, Ig domain unfolding/refolding, a process that is probably not generally reversible, has been dismissed as a mechanism for molecular elasticity under physiological conditions [23],[25],[26]. However, as there is no evidence for parallel filament bundling in myomesin (Figure S5), single molecule studies are suitable for estimating its level of molecular elasticity. In reference to the conclusions on titin, our model does not require additional unfolding of Ig domains, and, to the best of our knowledge, such reversible molecular elasticity is without precedent in any other filament system.
Our data, indicating the ability of myomesin to longitudinally extend by at about 500 Å under low external forces (Movie S1), provide a molecular model as to how myomesin could act as a highly elastic molecular strain sensor. The data presented here demonstrate that the established mechanism of molecular elasticity—reversible unfolding/refolding of highly exposed α-helical inter-domain inkers—is applicable for the complete repetitive array of IgH modules along the C-terminal part of the myomesin filament. These new findings are in agreement with previous proof-of-principle data, in which we established suitable structural biology and AFM protocols to investigate the C-terminal My12–My13 tandem [16],[19]. As the remaining domain structure of the N-terminal part of myomesin is markedly different, it is plausible that this mechanism of molecular elasticity is confined to the C-terminal My9–My13 segment of myomesin.
The length of the C-terminal myomesin filament of approximately 36 nm (360 Å) accounts for 85% of the previously estimated distance of about 44 nm between the pronounced M4 and M4′ lines in the sarcomeric M-band, which are thought to primarily consist of myomesin [2],[11],[22]. However, without more precise knowledge of the composition of these bands it remains uncertain whether bridging of the remaining distance requires some level of myomesin stretching, which would be well within our estimates of the range of myomesin extendability, or whether the part of myomesin associated with M4/M4′ is beyond the C-terminal myomesin filament My9–My13 that we have investigated. Our AFM measurements of myomesin elasticity also match the level of distance variability in the A-band observed in previous X-ray diffraction studies [3],[5].
Moreover, it will be of great interest to provide insight into how the predicted elastic properties of the C-terminal myomesin filament could functionally influence the nearby regions in myomesin that are involved in the assembly of interacting protein ligands, such as obscurin [15] and creatine kinase [27]. It remains speculative at this stage whether the coupling of mechanical forces, as they exist within the M-band, with molecular elasticity, as shown for the C-terminal part of myomesin, or with the potential regulation of kinase activity, as recently advocated for titin kinase at the M-band/A-band transition zone [21], has functional implications for interacting myomesin protein partners. Indeed, the possibility of cross-talk between mechanical stress sensing and signaling in muscle sarcomeres is an important question to be addressed. Mapping our findings on physiological stretching processes in the sarcomeric M-band is a challenging task for the future that will ultimately require phenotypic animal model studies with genetically modified versions of myomesin. Finally, it remains to be seen whether reversible filament elasticity by repetitive domain-connecting helix unfolding is unique to this M-bridge protein, or whether it also occurs in other filament systems.
The DNA sequences (MYOM1_HUMAN) encoding myomesin domains My9–My13 (residues 1141–1666), My9–My11 (residues 1141–1447), My10–My11 (residues 1247–1447), and My11–My13 (residues 1352–1666) were amplified by PCR from existing constructs [13]. The PCR products were cloned into the pET151/D-TOPO vector (Invitrogen). The two My9–My13 single residue mutants (K1457P and Y1551P) were prepared by standard mutagenesis protocols. The MBP–My9–My13 construct was prepared by recloning the My9–My13 construct into a pETM41 vector (European Molecular Biology Laboratory). All vectors used in this study carry an N-terminal hexa-histidine tag and a tobacco etch virus cleavage site N-terminal to the myomesin encoding sequence. The constructs were expressed in Escherichia coli strain BL21 (DE3) CodonPlus-RP. The purification protocol included two steps: Ni-NTA affinity chromatography, followed by size exclusion chromatography (GE, Superdex 200 16/60). When required, the hexa-histidine tag was removed by 6–8 h of incubation with tobacco etch virus protease. All purified proteins were dialyzed into 25 mM Tris/HCl (pH 7.5), 150 mM NaCl, and 5 mM β-mercaptoethanol.
Crystals of myomesin protein fragments were grown by vapor diffusion, by mixing equal volumes of protein solution (8 mg ml−1 for My10–My11; 8 mg ml−1 for My9–My11; 10 mg ml−1 for My11–My13) and precipitant solution (0.2 M sodium nitrate, 18% [w/v] polyethylene glycol 3350, and 5% [w/w] ethylene glycol for My10–My11; 0.18 M magnesium acetate and 20% [w/v] polyethylene glycol 3350 for My9–My11; 0.1 M 2-(N-morpholino)ethanesulfonic acid [pH 6], 0.22 M lithium sulfate, and 13% [w/v] polyethylene glycol 8000 for My11–My13). X-ray data were collected on the tunable wiggler beamline BW6 (My10–My11 seleno-L-methionine [SeMet], MPG/DESY, Hamburg, Germany), on X12 (My9–My11 EMBL/DESY, Hamburg, Germany), and on BM14 (My10–My11 native, European Synchrotron Radiation Facility, Grenoble, France), and ID23-1 (My11–My13, European Synchrotron Radiation Facility). All datasets were collected at 100 K using 2-methyl-2,4-pentanediol as cryo-protectant and were integrated, scaled, and merged using the HKL suite [28]. The My10–My11 structure was determined by using phases calculated from the anomalous signal of SeMet-incorporated protein. The My9–My11 structure was solved by molecular replacement using the individual Ig domains from My10–My11 as models in MOLREP [29]. The My11–My13 structure was solved by molecular replacement using the (My13)2 dimer module from the My12–My13 structure (2R15) and manually fitting the remaining My11 and My12 domains. The My9–My11 structure was refined by maximum likelihood including TLS refinement, as implemented in REFMAC5 [30]. Only two residues (0.7%) are found in generously allowed regions of the Remachandran plot. The native My10–My11 and My11–My13 structures were refined using the PHENIX suite [31], implementing maximum likelihood, simulated annealing, and TLS refinement protocols and, in addition, non-crystallographic symmetry restraints for My11–My13. For the My10–My11 structure all residues are found in the most favored or additionally allowed regions of the Ramachandran plot, whereas for My11–My13, 33 residues (3.0%) are located in generously allowed regions. Details of X-ray data collection and refinement are listed in Table 1. The composite dimeric My9–My13 structure was built by superimposing the My11 domain of the My11–My13 structure (chain A) and the My11 from the My9–My11 structure. Based on the observed limited tilt/twist angle variation (Figure 3C), additional composite My9–My13 models were generated. These models led to the same conclusions when compared with the EM and SAXS data.
Pure MBP–My9–My13 was diluted to 10 µg ml−1 and applied for 1 min onto a glow-discharged carbon-coated grid and subsequently stained with 2% uranyl acetate. Micrographs were recorded using a Tecnai G2 Spirit electron microscope (FEI electron optics) at a calibrated magnification of 41,400× and an accelerating voltage of 120 kV onto SO163 Kodak films. The micrographs were digitized in a SuperCoolscan 9000 Nikon scanner at a pixel spacing of 6.35 µm, and the images were binned by a factor of three, resulting in a sampling of 4.6 Å/pixel at the specimen level. 2,075 particles were selected manually using the program BOXER in EMAN [32]. Particles were Fourier filtered using a high pass of 100 Å and a low pass of 15 Å. They were initially aligned to a reference corresponding to a streak of density with width and length similar to that of the particles. The dataset was submitted to multivariate statistical analysis using the image processing software IMAGIC [33] followed by classification and averaging. A selection of class averages acted in turn as references for alignment of the particles using the image processing software SPIDER [34]. Successive iterations of alignment, multivariate statistical analysis, classification, and averaging were carried out, and a final selection was made from a set of 100 class averages. The coordinates from the X-ray composite model were converted into density map representations using IMAGIC [33]. This map was filtered to 30 Å resolution, and a model projection image for comparison with the electron microscope data was generated by projection of the map over the full range of possible orientations.
Scattering data from purified myomesin fragments My9–My13, My9–My13(K1457P), and My9–My13(Y1551P) were measured at a concentration range of 4–15 mg ml−1, each with intermittent buffer solution (25 mM Tris/HCl [pH 7.5] and 150 mM NaCl), at beamline X33 (EMBL/DESY, Hamburg, Germany). The measurements were carried out at 290 K, using a sample–detector distance of 2.7 m, covering the momentum transfer range 0.10 nm−1<s<4.5 nm−1 (s = 4π sin(ϑ)/λ, where 2ϑ is the scattering angle). The data were processed using standard procedures, corrected for buffer contribution, and extrapolated to infinite dilution using the program PRIMUS [35]. The radius of gyration Rg and forward scattering I(0), the maximum particle dimension Dmax, and the distance distribution function p(r) were evaluated using the program GNOM [36]. The molecular masses of the different constructs were calculated by comparison with reference bovine serum albumin samples. The scattering patterns from the high-resolution models were calculated by the program CRYSOL [37]. Seventeen ab initio models were reconstructed from the My9–My13wild-type scattering data using the simulated annealing program DAMMIN [38]. The SAXS data statistics are summarized in Table 2. To analyze flexibility, we generated 500 models deviating from the crystal structure within root mean square deviation up to 10 Å using the low frequency normal modes [39]. EOM [18] was employed to find the mixtures of the modified structures that best fit the experimental data (Table 2), and a modified p(r) function was computed (Figure S6B).
Single molecule force spectroscopy of My9–My13 dimers was performed on a custom-built atomic force microscope. Protein solution was adsorbed to freshly activated Ni-NTA-coated glass slides for 5–10 min and then rinsed with buffer (25 mM Tris/HCl [pH 7.5] and 150 mM NaCl) before starting the experiment. For all experiments, gold-coated cantilevers (Biolever type B, Olympus) with typical spring constants of 6 pN nm−1 were used. Cantilever deflection and piezo stage movement were recorded at 20 kHz. Data acquisition and analysis were performed using custom software within Igor Pro (Wavemetrics). Expected contour length increases were estimated from the difference between the contour length contribution of the end-to-end distance of the folded structure and the contour length of the corresponding sequence after unfolding. The latter is the number of amino acids multiplied by the average length contribution per amino acid, which has been determined to be 0.365±0.002 nm for our instrument using various proteins for calibration and a fixed persistence length of 0.5 nm [20].
The coordinates and experimental structure factors of the X-ray structures determined are deposited in the RCSB Protein Data Bank (http://www.rcsb.org; 2Y23, 2Y25, and 3RBS).
|
10.1371/journal.pbio.1001614 | A Temperature-Responsive Network Links Cell Shape and Virulence Traits in a Primary Fungal Pathogen | Survival at host temperature is a critical trait for pathogenic microbes of humans. Thermally dimorphic fungal pathogens, including Histoplasma capsulatum, are soil fungi that undergo dramatic changes in cell shape and virulence gene expression in response to host temperature. How these organisms link changes in temperature to both morphologic development and expression of virulence traits is unknown. Here we elucidate a temperature-responsive transcriptional network in H. capsulatum, which switches from a filamentous form in the environment to a pathogenic yeast form at body temperature. The circuit is driven by three highly conserved factors, Ryp1, Ryp2, and Ryp3, that are required for yeast-phase growth at 37°C. Ryp factors belong to distinct families of proteins that control developmental transitions in fungi: Ryp1 is a member of the WOPR family of transcription factors, and Ryp2 and Ryp3 are both members of the Velvet family of proteins whose molecular function is unknown. Here we provide the first evidence that these WOPR and Velvet proteins interact, and that Velvet proteins associate with DNA to drive gene expression. Using genome-wide chromatin immunoprecipitation studies, we determine that Ryp1, Ryp2, and Ryp3 associate with a large common set of genomic loci that includes known virulence genes, indicating that the Ryp factors directly control genes required for pathogenicity in addition to their role in regulating cell morphology. We further dissect the Ryp regulatory circuit by determining that a fourth transcription factor, which we name Ryp4, is required for yeast-phase growth and gene expression, associates with DNA, and displays interdependent regulation with Ryp1, Ryp2, and Ryp3. Finally, we define cis-acting motifs that recruit the Ryp factors to their interwoven network of temperature-responsive target genes. Taken together, our results reveal a positive feedback circuit that directs a broad transcriptional switch between environmental and pathogenic states in response to temperature.
| Microbial pathogens of humans display the ability to thrive at host temperature. So-called “thermally dimorphic” fungal pathogens, which include Histoplasma capsulatum, are a class of soil fungi that upon being inhaled into the human lung, undergo dramatic changes in cell shape and virulence gene expression in response to host temperature. The ability of these pathogens to cause disease is exquisitely coupled to temperature response. Here we elucidate the regulatory network that governs the ability of H. capsulatum to switch from a filamentous form in the soil environment to a pathogenic yeast form at body temperature. The circuit is driven by three transcription regulators (Ryp1, Ryp2, and Ryp3) that control yeast-phase growth. We show that these factors, which include two highly conserved proteins of the Velvet family of unknown function, bind to specific regulatory DNA elements and directly regulate expression of virulence genes. We identify and characterize Ryp4, a fourth regulator of this pathway, and define DNA motifs that recruit these transcription factors to their temperature-responsive target genes. Our results provide a molecular understanding of how changes in cell shape are linked to expression of virulence genes in thermally dimorphic fungi.
| Cells adapt to their environment by responding to specific environmental stimuli such as light, temperature, and nutrients. For microbial pathogens, mammalian body temperature can signal the induction of pathways required for host colonization and pathogenesis [1]. One such group of organisms is the thermally dimorphic fungal pathogens, which include Coccidioides, Paracoccidioides, Blastomyces, and Histoplasma species. These evolutionarily related fungi are notable among fungal pathogens in that they all cause disease in healthy individuals [2]. Each of these organisms grows in a mold form in the soil, forming long, connected filaments that produce vegetative spores [3]. When the soil is aerosolized, filamentous cells and spores can be inhaled by mammalian hosts and converted into a parasitic form within the host lung. Conversion entails a dramatic change in cell shape to a budding yeast form for the majority of these pathogens, as well as the transcriptional induction of virulence genes required to cause disease in the host [3]. For all thermally dimorphic fungi, host temperature is the key signal that triggers this developmental switch, but little is known about the coordinated induction of morphologic changes and virulence gene expression by temperature.
Histoplasma capsulatum, which is endemic to the Ohio and Mississippi River Valleys of the United States, can cause life-threatening respiratory and/or systemic disease (histoplasmosis) [2],[4]. It is estimated that up to 25,000 people develop life-threatening infections in endemic regions each year, with at least 10-fold more mild or asymptomatic infections [2],[4]. Although the pathogen propagates as spores and in a filamentous form in the environment, H. capsulatum is found almost exclusively in the yeast form within mammalian hosts. Despite the prevalence of H. capsulatum and its threat to human health, we have a limited understanding of the transcriptional regulatory network that governs pathogenic yeast-phase growth. Previously, we identified three regulators, Ryp1, Ryp2, and Ryp3, and showed that they are required for yeast-phase growth [5],[6]. Whereas wild-type cells grow in the yeast form at 37°C, ryp1, ryp2, and ryp3 mutants grow constitutively in the filamentous form independent of temperature. In wild-type cells, RYP1, RYP2, and RYP3 transcripts and proteins accumulate preferentially at 37°C and each Ryp protein is required for the wild-type expression levels of the others [5],[6].
RYP1 encodes a fungal-specific transcriptional regulator that is required for modifying the transcriptional program of H. capsulatum in response to temperature [5]. Ryp1 belongs to a conserved family of fungal proteins that regulate cellular differentiation in response to environmental signals. The best-studied member of this family of proteins is Wor1, which was identified as a master transcriptional regulator that controls a morphological switch required for mating in Candida albicans [7]–[9]. In the model yeast Saccharomyces cerevisiae, the Ryp1 ortholog, Mit1, is required for a morphologic switch that occurs under nutrient limitation [10]. Ryp1 orthologs in the plant pathogens Fusarium oxysporum (Sge1), Fusarium graminerium (Fgp1), and Botrytis cinerae (Reg1) are required for full pathogenicity and conidiation [11]–[13]. All of these observations signify the importance of Ryp1 orthologs for transduction of environmental cues to regulate cell morphology and virulence. Furthermore, it was recently demonstrated that Wor1 contains a DNA-binding domain that is conserved throughout the WOPR (Wor1, Pac2, Ryp1) family of proteins [14], suggesting that these regulators respond to specific signals by triggering a transcriptional program.
In contrast, Ryp2 and Ryp3 belong to the Velvet family of regulatory proteins [6], whose molecular function is unknown. This family is typified by Velvet A (VeA), which was initially characterized as a regulator of sexual spore production in Aspergillus nidulans [15],[16], but is now known to also regulate secondary metabolism and development in many fungi including Aspergillus species, Fusarium species, Neurospora crassa, and Acremonium chrysogenum (reviewed in [17]). In H. capsulatum, the VeA ortholog Vea1 has a role in sexual development but is dispensable for yeast-phase growth [18]. Additionally, many fungi have multiple Velvet family proteins that collaborate to serve regulatory functions. For example, in A. nidulans, three Velvet family proteins (the Ryp2 ortholog VosA, the Ryp3 ortholog VelB, and VeA itself) act together to regulate asexual and sexual development and secondary metabolism [19]. Notably, since Velvet family proteins do not contain canonical DNA binding domains or other domains of known function, their mechanistic role in regulation of developmental processes is unclear.
As noted above, both WOPR and Velvet family proteins are widely distributed among fungi, although the Hemiascomycetes, including Saccharomyces and Candida species, lack Velvet family proteins. Since both families of proteins are required for yeast-phase growth in H. capsulatum, we explored if and how these two distinct classes of fungal regulators work together to govern temperature-responsive traits by dissecting the Ryp regulatory network in H. capsulatum. To this end, we performed whole-genome transcriptional profiling and chromatin immunoprecipitation experiments to determine the shared and unique roles of Ryp1, Ryp2, and Ryp3 in regulating yeast-phase growth. We show that 96% of yeast-phase enriched transcripts are dependent on Ryp1, Ryp2, and Ryp3 for their enhanced expression in response to temperature, whereas 66% of filamentous-phase enriched transcripts require Ryp1, Ryp2, and Ryp3 to prevent their inappropriate expression at 37°C. We demonstrate that all three Ryp factors physically interact and associate with the upstream regions of a core set of target genes, including those required for yeast-phase growth and virulence. Additionally, we identify a fourth transcriptional regulator, Ryp4, to be a component of the Ryp regulatory network required for temperature-responsive yeast-phase growth. Finally, the identification of two distinct cis-acting regulatory sequences that are bound and utilized by Ryp proteins provides the first evidence that highly conserved Velvet family proteins can directly bind to DNA and activate gene expression using a unique cis-acting element. Overall, our results provide a molecular understanding of how regulation of cell morphology and virulence gene expression is coordinated in response to temperature in H. capsulatum.
Our previous studies showed that there are marked differences in the transcriptional profiles of wild-type yeast-form cells grown at 37°C and filamentous cells grown at room temperature [5],[20]. Cells lacking RYP1, RYP2, or RYP3 grow constitutively as filaments independent of temperature [5],[6], and Ryp1 is required for the expression of the majority of the transcripts enriched during yeast-phase growth at 37°C [5]. Here we sought to understand whether Ryp2 and Ryp3 are also involved in regulating expression of genes required for yeast-phase growth. To this end, we performed whole-genome expression profiling experiments comparing the transcriptional profiles of multiple biological replicates of ryp1, ryp2, ryp3 mutants and wild-type strains grown at room temperature (RT) and 37°C. We identified 388 genes with significantly increased transcript levels and 376 genes with significantly decreased transcript levels in wild-type yeast cells grown at 37°C compared to wild-type filaments grown at RT (Figure 1A and Table S1). These gene sets were referred to as yeast-phase–specific (YPS) and filamentous-phase–specific (FPS) genes, respectively.
Gene Ontology (GO) analysis showed that YPS genes were enriched in terms such as cell cycle regulation, chromosome segregation, and DNA recombination, all of which are characteristics of metabolically active, budding yeast cells (Figure 1C). In contrast, FPS genes were enriched in terms such as cell wall, extracellular region, response to oxidative stress, hydrolase, and lipase activity, consistent with the idea that filamentous cells have a distinct cell wall structure and produce enzymes related to saprophytic activity (Figure 1C). Additionally, YPS genes included the previously identified virulence genes of H. capsulatum: Genes encoding for calcium-binding protein 1 (CBP1) [21], yeast-phase specific protein 3 (YPS3) [22], super-oxide dismutase 3 (SOD3) [23], alpha-(1,4)-amylase (AMY1) [24], and heat-shock protein 90 (HSP82) [25] were among the most differentially regulated YPS genes. Furthermore, YPS genes included M antigen/catalase B (CATB), which has been shown to display yeast-specific expression in the H. capsulatum strain used here and encodes a secreted catalase that may help cope with oxidative stress in the host [26]. Transcript levels of these virulence genes were significantly down-regulated in ryp mutants, showing that Ryp factors are important for expression of virulence genes in H. capsulatum in response to temperature (Figure 1B).
Global comparison of the gene expression profile of wild-type cells to that of each ryp mutant revealed that the ryp mutants had a strongly diminished transcriptional response to temperature. The transcriptome of each ryp mutant grown at 37°C strongly resembled the transcriptome of wild-type cells grown at RT (Figure 1D). Additionally, transcriptional profiles of the individual ryp mutants were strikingly similar to each other (Figure 1A), indicating that they may act in the same temperature-responsive pathway. Transcript levels of the overwhelming majority of the YPS genes (96%) were decreased (>1.5-fold) in ryp mutants at 37°C, indicating that Ryp proteins were required for their wild-type expression level. Similarly, the Ryp proteins were required to prevent inappropriate expression of the majority of the FPS genes at host temperature: Transcript levels of 66% of the FPS genes were increased (>1.5-fold) in ryp mutants compared to the wild-type strain at 37°C. These results showed that Ryp1, Ryp2, and Ryp3 are master regulators required for the appropriate temperature-responsive transcriptional program in H. capsulatum.
As observed previously, our set of YPS genes included RYP1, RYP2, and RYP3, and each of the RYPs depended on the other two for its temperature-regulated expression (Figure S1B) [6]. Since RYP transcript levels are low at RT under laboratory conditions, we expected that the Ryp proteins might play only a minor role in regulation of gene expression at RT. Consistent with this idea, the Ryp factors are not required for the normal transcriptional profile of filaments grown at RT (Figure S1A). In sum, our transcriptional profiling experiments revealed that Ryp1, Ryp2, and Ryp3 are major regulators of yeast-phase growth at 37°C; are dispensable for filamentous-phase growth at RT; can either facilitate or repress transcript accumulation; and may act in the same pathway to regulate gene expression.
Previous studies reported that Wor1, an ortholog of Ryp1, associates with DNA at hundreds of intergenic regions to regulate gene expression in C. albicans [14]. However, genome-wide DNA associations of Ryp1 have not been investigated in H. capsulatum. Additionally, Velvet family proteins do not contain a known DNA-binding domain and the ability of Ryp2 and Ryp3 orthologs to associate with DNA has not been explored. To establish the genome-wide association of Ryp factors with DNA and to distinguish between direct and indirect targets in the Ryp regulons, we performed chromatin immunoprecipitation-on-chip (ChIP-chip) using antibodies raised against Ryp1, Ryp2, and Ryp3. Experiments were performed in either wild-type yeast cells grown at 37°C, or in the respective ryp mutant control grown at 37°C. We observed 361 ChIP events throughout the genome of wild-type cells (Figure 2, Figure S2A, and Table S2). Most notably, there were a large number of targets (182 loci) that associated with at least two Ryp factors, and 94 loci associated with all three Ryp factors, suggesting that Ryp1, Ryp2, and Ryp3 can act together to regulate gene expression. Interestingly, only Ryp1 had a large number of events (161 loci) that were not shared with other Ryp factors, indicating that Ryp1 has a broader regulon than Ryp2 and Ryp3 (Figure S2A). Further characterization of the ChIP events revealed that intergenic lengths corresponding to Ryp association events were significantly longer than the average intergenic length in the genome (Figure S2B). A similar trend was noted previously for Wor1 association events in C. albicans, but its biological significance is unknown [27]. Notably, shared ChIP events that involved all three Ryp factors showed the most drastic shift in intergenic length distribution.
To map the genomic regions defined by ChIP-chip events to specific genes, we used a validated gene set that was defined previously based on gene expression and sequence conservation [28] and identified genes that lie downstream of each ChIP event (Table S2). Our first notable observation was that Ryp1, Ryp2, and Ryp3 showed interdependent regulation. All three Ryp factors associated upstream of RYP1 and RYP2, although none of the three factors associated upstream of RYP3 (Figure 3A and Table S2). To further explore the relationship between DNA association and gene expression, we overlaid gene expression data onto all ChIP-chip target genes (Figure 2). Targets that associated only with Ryp1 were not significantly enriched in YPS genes compared to the whole genome. In contrast, we observed significant enrichment of YPS genes for DNA-association events that were shared by Ryp1, Ryp2, and Ryp3, as well as events that were shared only by Ryp1 and Ryp3. This analysis revealed a correlation between shared association events and genes whose enhanced expression was induced by growth at host temperature. Strikingly, many of the known virulence genes (CBP1, SOD3, and YPS3) were shared Ryp targets (Figure 3B and Table S2). These results indicate that the known core virulence genes are direct targets of the Ryp factors, and suggest that the remaining shared Ryp targets are interesting candidates for potential virulence factors. Notably, shared Ryp targets and all Ryp targets were enriched for predicted signal peptides (p = 1.4e-06 and p = 2.0e-06, respectively) compared to the whole genome, which is of interest since secreted proteins produced by intracellular pathogens are often involved in virulence.
In addition to virulence factors, the shared Ryp targets included a single transcription factor with a known DNA-binding domain and yeast-phase–specific expression. The corresponding gene, designated HISTO_DM.Contig933.eannot.1650.final_new, encodes a Zn(II)2Cys6 zinc binuclear cluster domain protein (Figure 4A). As shown by microarray and qRT-PCR experiments, transcript levels of this gene were 6.5- to 35-fold higher at 37°C compared to RT, indicating that it displays enhanced expression in wild-type yeast-phase cells as compared to filaments (Figure 4B and Table S1). Moreover, Ryp1, Ryp2, and Ryp3 were required for expression of this gene at 37°C (Figure 4B and Table S1). We investigated the possibility that, similar to the RYP factors, HISTO_DM.Contig933.eannot.1650.final_new is essential for growth in the pathogenic yeast form. Since targeted gene disruption in H. capsulatum is inefficient, we generated knockdown strains using RNA interference (RNAi). We were able to deplete mRNA levels of HISTO_DM.Contig933.eannot.1650.final_new by 72%–88% (Figure 4C). Additionally, these RNAi strains were unable to grow in the yeast form and instead exhibited robust filamentous growth at 37°C, similar to the ryp mutants (Figure 4D). Loss of RNAi plasmids resulted in reversion to yeast-phase growth at 37°C (Figure S3A), indicating that the phenotype was dependent on the RNAi plasmids. Thus, we renamed this gene RYP4 (Required for Yeast-Phase Growth). Although these morphologic studies revealed that ryp4 knockdown strains appeared drastically different from wild-type cells grown at 37°C, they were indistinguishable from wild-type filaments in appearance when grown at RT (Figure 4D).
BLASTP analysis indicated that the closest homolog of Ryp4 is FacB of A. nidulans (E-value <1e-180), and phylogenetic analysis (see Materials and Methods) revealed that Ryp4 is an ortholog of FacB (Figure S3B and S3C). FacB is a transcriptional regulator of genes involved in acetate utilization in Aspergillus species and N. crassa [29]–[33]. Considering this conserved role in multiple fungal species, we investigated whether Ryp4 has a role in acetate utilization in H. capsulatum. However, unlike facB mutants in other organisms, ryp4 knockdown strains were able to grow in acetate as a major carbon source (unpublished data), leading us to favor the hypothesis that Ryp4 has been rewired to regulate morphology in H. capsulatum.
To assess whether Ryp4 is required for normal yeast-phase–specific gene expression in response to temperature, we performed whole-genome expression profiling experiments using wild-type cells carrying control vectors (hereafter referred to as wild-type) and ryp1, ryp2, ryp3, and ryp4 knockdown strains grown either at RT or at 37°C. In contrast to the gene expression studies performed in Figure 1 using insertion mutants, here we used ryp1, ryp2, and ryp3 knockdown strains to match the ryp4 knockdown strain (hereafter referred to as “mutant”). Statistical analyses revealed 441 YPS genes and 362 FPS genes in this dataset (Table S3). Similar to the aforementioned microarray dataset, regulation of the expression of the majority of YPS and FPS genes was dependent on Ryp1, Ryp2, Ryp3, and Ryp4 (Figure 4E and Table S3). Notably, the transcriptional profile of the ryp4 mutant was very similar to that of ryp1, ryp2, and ryp3 mutants (Figure 4E). The transcriptional profile of ryp4 mutants grown at 37°C was similar to that of wild-type filamentous cells grown at RT (Figure S4A). Additionally, both microarray analysis and qRT-PCR experiments showed that Ryp4 was required for the expression of RYP1, RYP2, and RYP3 (Figure S3D). In contrast to its critical role at 37°C, Ryp4 was not required for the normal transcriptional profile of filaments grown at RT (Figure S4B). Overall, these results indicate that Ryp4, like Ryp1, Ryp2, and Ryp3, is critical for temperature-regulated gene expression at 37°C.
To further explore the Ryp4 regulon, we performed ChIP-chip experiments in wild-type cells using antibodies raised against Ryp4. We identified 61 Ryp4 association events that occur in cells grown at 37°C (Table S2). Interestingly, the majority (74%) of these events were shared with other Ryp factors, indicating that Ryp4 acts in concert with other Ryp factors to regulate gene expression. Next, we identified genes downstream of each Ryp4 ChIP event (Table S2), and visualized these data together with Ryp1, Ryp2, and Ryp3 events (Figure S5). We found that about a quarter (26%) of Ryp1, Ryp2, and Ryp3 shared targets were also occupied by Ryp4 (Figure 4F). Furthermore, these common Ryp targets were even more significantly enriched for YPS genes than targets that were shared by Ryp1, Ryp2, and Ryp3, but not Ryp4 (Figure 4F). These YPS genes included core regulators of morphology (RYP1, RYP2, and RYP4) and genes required for virulence (CBP1 and SOD3), further emphasizing the role of Ryp4 as a fundamental regulator of yeast-phase growth and an essential component of the temperature-responsive Ryp regulatory network (Figures 3A,B and 4A).
The ChIP data described above revealed that Ryp1, Ryp2, Ryp3, and Ryp4 share a large number of overlapping targets, suggesting that these proteins may physically interact. To test this hypothesis, we performed co-immunoprecipitation (co-IP) experiments in wild-type cells grown at 37°C using each of the Ryp antibodies. Of all possible co-IP combinations, we were able to reliably and reproducibly detect Ryp1, Ryp2, and Ryp3 in Ryp2 and Ryp3 IPs, indicating that at least these three Ryp proteins physically interact (Figure 5A). Ryp4 IPs did not yield reproducible interactions with other Ryp proteins. No Ryp protein signal was present in control IPs performed in the corresponding ryp mutant grown at 37°C (unpublished data).
In addition to these biochemical experiments, we used yeast two-hybrid experiments to assess Ryp1–Ryp2–Ryp3–Ryp4 interactions. We observed a reciprocal interaction between Ryp2 and Ryp3, confirming the co-IP results (Figure 5B). Our results also revealed that the Ryp2 N-terminus (which contains the Velvet domain), but not the Ryp2 C-terminus, is required for interaction with Ryp3 (Figure 5C). On the other hand, the Ryp2 C-terminus mediates Ryp2–Ryp2 interactions (Figure 5B and 5C). No interactions were observed for Ryp4 (unpublished data). Despite numerous attempts at transformation and analysis of multiple clones, we were unable to express Ryp1 bait or prey fusion proteins in the S. cerevisiae strains used for yeast-two-hybrid analysis. As a result, yeast-two-hybrid interactions with Ryp1 could not be assessed. Similarly, a previous study reported difficulty expressing yeast-two-hybrid constructs made with the Ryp1 ortholog from F. oxysporum [13], suggesting that expression of these Ryp1 two-hybrid fusion proteins is toxic in standard laboratory strains of S. cerevisiae.
To further explore the Ryp regulatory network, we analyzed each Ryp ChIP event set for the nonrandom occurrence of conserved cis-acting regulatory sequences (hereafter referred to as DNA motifs). This de novo motif analysis was especially interesting for Ryp2 and Ryp3, since there has been no prior evidence that Velvet family proteins associate with DNA, and thus, no DNA motifs have been identified for Velvet family proteins. Through the analysis of Ryp1, Ryp2, and Ryp3 ChIP events, we identified two distinct DNA motifs, which we named Motif A and Motif B. Specifically, different variants of Motifs A and B were identified through the analysis of Ryp ChIP events (Table S4, Figure S6A and S6B), and motifs that had the best predictive characteristics using receiver operating characteristic (ROC) plots (as described in [34]) were selected as Motif A and Motif B (Figure 6A). Motif A is very similar to the Wor1 motif, which was previously identified using biochemical approaches [14]. Identification of Motif A by a completely independent approach validates our ChIP-chip data and our motif analysis pipeline. Motif B, which is quite distinct from Motif A, did not resemble any previously identified motifs according to searches of the motif databases JASPAR (http://jaspar.genereg.net/) and YETFASCO (http://yetfasco.ccbr.utoronto.ca/). ROC plots using all ChIP events from each of the regulators demonstrated that Motif A and B were enriched in the entire ChIP set. Shuffled versions of each motif resulted in loss of specificity (Figure 6B). Of note, motif-finding algorithms did not yield a meaningful result for Ryp4, which is not surprising as the complex binding sites of zinc binuclear cluster transcriptional regulators can be difficult to predict [35],[36].
Further analysis revealed that Motif A was associated with Ryp1 ChIP events, whereas Motif B was associated with Ryp2 and Ryp3 ChIP events: Motif A specificity was dependent only on the inclusion of Ryp1 ChIP events (Figure S6C), whereas Motif B specificity was dependent only on the inclusion of Ryp2 and Ryp3 events (Figure S6D). Furthermore, Motif A enrichment was similar in all Ryp1 targets regardless of whether they were shared targets with the other Ryp factors or whether they associated only with Ryp1 (Figure S6E). In contrast, Motif B was enriched in Ryp1 targets that were shared with Ryp2 and Ryp3, but had no specificity in targets that were unique to Ryp1 (Figure S6E). These results corroborate the model that association of Ryp1 with DNA correlates with the presence of Motif A, and association of Ryp2 and Ryp3 with DNA correlates with the presence of Motif B.
Motifs A and B are distributed throughout the H. capsulatum genome and are present in many of the Ryp targets (Tables S5). Specific examples are shown in Figures 3 and 4 where Motif A and B distribution is shown in the upstream regulatory regions of RYP1, RYP2, RYP4, CBP1, SOD3, and CATB. Neither motif was found in the RYP3 upstream region (Figure 3), which is in agreement with our ChIP results that the Ryp factors do not associate upstream of RYP3.
Our motif analyses suggested that Ryp1 associates with DNA via Motif A, whereas Velvet family proteins (Ryp2 and Ryp3) associate with DNA via Motif B. To test whether Ryp1, Ryp2, or Ryp3 can directly bind to Motif A or Motif B, we performed electrophoretic mobility shift assays (EMSAs). The promoter region of the CBP1 gene, which associates with the Ryp proteins by ChIP-chip and contains both Motif A and Motif B (Figure 3B), was used to design two distinct 60 bp probes encompassing either Motif A or Motif B (Figure 3B). His-tagged versions of Ryp1, Ryp2, and Ryp3 were synthesized in coupled in vitro transcription and translation reactions and then subjected to purification. Additionally, it was already known that the N-terminus of C. albicans Wor1, which is highly homologous to the N-terminus of Ryp1, contains two globular domains that are sufficient to bind the Wor1 motif [14]. Therefore, we also synthesized and purified a His-tagged version of the N-terminus of Ryp1 (Ryp1(N), 1–267 aa) that harbors only the globular domains.
EMSA revealed that full-length Ryp1 binds directly to Motif A, whereas control extracts contained no binding activity (Figure 6C). In analogy to Wor1, we also observed that Ryp1(N) binds directly to Motif A (Figure 6C). To explore the ability of Ryp2 and Ryp3 to bind Motif B, we performed mobility shift assays with either Velvet protein alone or with a combination of Ryp2 and Ryp3. Whereas mobility shift assays performed with either Ryp2 or Ryp3 showed no binding activity, addition of both of these Velvet proteins resulted in binding of the Motif B probe (Figure 6D). All observed band shifts were diminished upon addition of unlabeled competitor probe into the binding reactions. Notably, these experiments provide the first evidence that Velvet family proteins bind DNA directly.
Since Ryp1 is sufficient to bind Motif A and Ryp2–Ryp3 is sufficient to bind Motif B, we next used an in vivo transcriptional activation assay [14] to explore whether each motif was sufficient to drive gene expression in a heterologous system when the appropriate Ryp proteins were expressed. We cloned a single copy of Motif A or B upstream of the UAS-less CYC1 promoter fused to the lacZ gene. We introduced these reporter plasmids along with constructs expressing the Ryp proteins, either individually or in combination, into S. cerevisiae and monitored β-galactosidase activity. S. cerevisiae provides an ideal heterologous expression system for our experiments, since the reporter strain we used lacks Ryp1 orthologs, and Velvet family proteins are not present in S. cerevisiae.
In the strains that contained the Motif A reporter construct, β-galactosidase activity was detected only when Ryp1 was heterologously expressed (Figure 7A). This activity was severely diminished when the conserved residues of Motif A were mutated. Co-expression of Ryp2, Ryp3, or Ryp4 individually with Ryp1, or Ryp1, Ryp2, and Ryp3 together, did not lead to an increase in β-galactosidase activity (Figure 7A). These results indicate that Ryp1 is necessary and sufficient to drive gene expression via Motif A. In contrast, in strains containing Motif B, β-galactosidase activity was detected only when Ryp2 and Ryp3 were expressed together, but not when Ryp2 or Ryp3 were expressed singly (Figure 7B). This activity was also dependent on the conserved nucleotides of Motif B. Co-expression of Ryp1 or Ryp4 along with Ryp2 and Ryp3 did not enhance Ryp2-Ryp3–dependent β-galactosidase activity (Figure 7B). Additionally, even though the Ryp2 N-terminus (Ryp2(N)), which contains the Velvet domain, is sufficient to interact with Ryp3 (Figure 5C), co-expression of Ryp2(N) and Ryp3 was not sufficient to drive gene expression using Motif B (Figure 7C). For all these experiments, β-galactosidase activity was dependent on the presence of the appropriate motif (Figure S7A), but independent of motif orientation (Figure S7B and S7C). These results indicate that Ryp2 and Ryp3 together are necessary and sufficient to drive gene expression via Motif B and, taken together with the EMSA studies, provide the first evidence that Velvet proteins bind DNA via a conserved motif to direct transcriptional activation.
Cells have derived complex regulatory networks to reprogram their transcriptional response upon changes in the environment. In thermally dimorphic fungi, regulation of cell morphology and virulence traits is coupled: they respond to host temperature by altering their cell shape and inducing virulence gene expression. This study is the first elucidation of the transcriptional circuitry underlying this dramatic change in any of these organisms. In this article, we have shown that an interlocking network of transcription factors regulate each other and common target genes to trigger a transcriptional program that is required for cell shape changes and virulence gene expression in response to host temperature in H. capsulatum.
Each of the Ryp transcription factors described here, including the newly discovered Ryp4, is absolutely required for the normal transcriptional profile of cells grown at 37°C. The structure of the corresponding transcription factor network is diagrammed in Figure 8, which illustrates the connectivity between the four proteins required to program the switch from filamentous to yeast-form growth in response to temperature. Each Ryp factor is required for the expression of the others, and each associates upstream of the RYP1, RYP2, and RYP4 genes—but none of the factors, including Ryp3 itself, associates upstream of the RYP3 gene. Thus we propose that at 37°C, each of the factors acts in a positive-feedback loop to regulate itself and each of the others. The exception is Ryp3, which regulates the other factors but is only indirectly regulated by them and does not directly regulate itself. Perhaps the interlocking nature of the network promotes a robust response by amplifying and stabilizing the signal generated by increased temperature. Additionally, the absolute requirement for activation of all four Ryp factors may provide specificity by requiring a sustained increase in temperature (as experienced within a mammalian host) to trigger the appropriate developmental program. Moreover, the requirement for multiple factors may allow unknown host signals other than temperature to influence the transition from filaments to yeast-form cells. The nature of the molecular signal that induces an initial increase in Ryp expression in response to temperature is unknown, but others have identified a histidine kinase, DRK1, that is required for yeast-phase growth in H. capsulatum and the related thermal dimorph Blastomyces dermatitidis [37]. A possible relationship between DRK1 and regulation of Ryp factors has not been explored.
Ryp factors are required both to promote transcription of genes with enhanced expression at 37°C and to prevent the inappropriate expression of filament-associated genes at this temperature. Transcriptional profiling of wild-type cells grown either at 37°C (in the yeast form) or at room temperature (in the filamentous form) yielded 764 genes (about 8.5% of the genome) that passed our criteria for differential expression in one of the two growth phases. Remarkably, 96% of YPS genes required each of the four Ryp proteins to achieve differential expression in response to temperature. The majority of this regulon is a consequence of indirect regulation by the Ryp factors, since only a fraction of YPS genes are downstream of Ryp association events. The most significant enrichment for YPS genes was observed for shared targets of multiple Ryp factors: about 21% of shared targets were YPS genes as opposed to 4% of the whole genome. Interestingly, Ryp association events that are unique to an individual Ryp factor, such as the large cohort of ChIP-chip events that associate only with Ryp1 and not the other three Ryp factors, showed no enrichment for YPS genes. Additionally, Ryp targets that were not YPS genes did not depend on Ryp factors for their expression under standard laboratory conditions. It is possible that Ryp factors are poised upstream of these genes to regulate their expression in response to environmental signals other than temperature, such as light intensity, nutrient availability or exposure to reactive radicals in the host.
Recruitment of the Ryp transcription factors to their direct targets is driven, at least in part, by DNA sequence motifs that are sufficient to recruit either Ryp1 (Motif A) or the Ryp2–Ryp3 complex (Motif B) (Figure 8). Motif A, which is highly similar to the DNA motif that was defined for the Ryp1 ortholog Wor1, is enriched in Ryp1 targets, whereas Motif B is enriched in Ryp2 and Ryp3 targets. Figure 8 illustrates the major categories of motif distribution for YPS gene regulatory regions that associate with Ryp factors. Interestingly, some YPS genes that contain only Motif A or only Motif B show association with Ryp1, Ryp2, and Ryp3. Although there are several models that could explain this multifactorial association, we favor the idea that there are different subcomplexes of Ryp proteins in the cell. For example, we propose that Ryp1 associates with DNA directly via Motif A in the absence of Ryp2 and Ryp3, thereby accounting for the many Ryp1 ChIP-chip events that are not shared by the other Ryp factors. Supporting this model is the finding that Ryp1 is sufficient to drive gene expression via Motif A in S. cerevisiae, and the biochemical experiments that show that the Ryp1 ortholog Wor1 associates directly with DNA via a motif that is highly similar to Motif A [14]. Alternatively, Ryp1 can be present in a complex with Ryp2 and Ryp3, and then could be recruited to the DNA either directly, via Motif A, or indirectly, via interaction of Ryp2–Ryp3 with Motif B. This model also predicts that Ryp2–Ryp3 can be recruited to the DNA either directly or via interaction with Ryp1. Interestingly, the existence of ChIP-chip targets that contain Motif A but associate only with Ryp1 and not with Ryp2–Ryp3 suggests that there is a genomic feature, such as chromosomal context for Motif A, that distinguishes the set of Ryp1-only targets from the shared targets. For example, perhaps chromosomal context causes the association of Ryp1 with Motif A in the shared targets to require nonspecific interactions of Ryp2–Ryp3 with the DNA. Interestingly, we identified a number of ChIP-chip targets that associate only with Ryp3, and found that these targets are enriched for Motif B. Since binding of Ryp3 to Motif B, at least in vitro, requires the presence of Ryp2, it is possible that these orphan events are actually shared with Ryp2 but fall below our level of detection for Ryp2 association. Alternatively, Ryp3 might be recruited to these targets via association with one of the other two Velvet domain proteins in H. capsulatum. Finally, the newly identified yeast-phase regulator Ryp4 seems to play no role in recognition of Motif A or Motif B, at least in vitro, and is incapable of driving gene expression via these motifs in a heterologous transcriptional activation assay. Additionally, we did not observe any physical interactions between Ryp4 and other Ryp factors. Taken together, these observations suggest that a third cis-regulatory element might recruit Ryp4 to its targets.
In this study, we provide the first evidence that Velvet family proteins such as Ryp2 and Ryp3 can bind DNA directly. This family of proteins is well studied in environmental fungi, fungal plant pathogens, and an opportunistic fungal pathogen of humans; however, although physical interactions between multiple Velvet family proteins have been defined, the molecular mechanism of Velvet family proteins in general is unknown. In addition to showing binding of Ryp2 and Ryp3 to the DNA via mobility shift assays, we show that Motif B is sufficient to drive gene expression in the model yeast S. cerevisiae when Ryp2 and Ryp3 are co-expressed. The N-terminus of Ryp2, which contains the Velvet domain, is required for Ryp2–Ryp3 interaction in the yeast two-hybrid assay, whereas the C-terminus of Ryp2 mediates multimerization of Ryp2 (Figure 5C). In a previous study, the C-terminal region of the Ryp2 ortholog VosA was predicted to be a transcriptional activation domain [38]. Since co-expression of the Ryp2 N-terminus with Ryp3 is not sufficient to activate gene expression through Motif B (Figure 7C), we conclude that the Ryp2 C-terminus is required for activation of gene expression in our S. cerevisiae transcriptional activation assays. Whether a Ryp2 homodimer or multimer has a role in gene expression in H. capsulatum or is important for the higher order formation of a Ryp complex remains to be investigated.
One of the most interesting findings to arise from this network analysis is that the Ryp factors, which themselves are YPS genes that are required for morphology, directly regulate YPS genes that are required for pathogenesis (e.g., CBP1, SOD3, and YPS3). A hallmark of thermally dimorphic fungal pathogens is that temperature regulates cell shape and other factors required for pathogenesis. In the case of H. capsulatum, yeast-phase morphology is thought to be critical for the lifestyle of this fungus in the host, especially since replication of the fungus within host immune cells is likely to be incompatible with filamentous growth. Thus, Ryp factors catalyze both the cell shape change and the increased expression of known virulence factors, resulting in a molecular link between temperature and the expression of traits required to cause disease in a mammalian host. A link between morphology and virulence has been explored for many of the major human fungal pathogens, perhaps most extensively for C. albicans (reviewed in [39]). Unlike the thermally dimorphic fungi, which are found almost exclusively in a hyphal form in the soil and a yeast or spherule form in the host, C. albicans is present in multiple morphologies in the host, and filamentous cells are thought to play a major role in disease. We observed co-regulation of yeast-phase morphology and known virulence genes in H. capsulatum. Similarly, numerous studies in C. albicans have observed that hyphal formation is coordinated with the expression of genes required for adhesion, host tissue invasion, antifungal drug resistance, and oxidative stress response [39], suggesting that a subset of fungal pathogens have evolved regulatory circuits that link virulence gene expression with morphologic changes. Notably, despite the importance of the yeast-to-hyphal transition in the infectivity of C. albicans, a systematic analysis of a large-scale C. albicans deletion library showed that morphology can be uncoupled from virulence. Although a large number of mutants were identified to have defects in both virulence and morphology, a sizeable number of mutants were defective for virulence despite maintaining normal morphologic responses in vitro, and some mutants displayed normal infectivity despite having substantial defects in morphogenesis [40]. In contrast, the ability to differentiate into yeast-form cells is likely to be essential for the intracellular parasitism that is characteristic of H. capsulatum infection. Therefore, it may be the case that yeast-phase morphology is categorically linked to virulence in the thermally dimorphic fungal pathogens.
Finally, this work expands the rich history of elucidating fungal transcription circuits to understand regulatory networks in eukaryotic cells. Studies of fungal gene regulation continue to provide examples of novel DNA-binding domains, as was previously the case with the Ryp1 family of proteins [14]. In this study, we show association between DNA and two Velvet family proteins that lack known DNA-binding domains, now implicating this highly conserved protein family in direct gene regulation. The most interesting molecular implication of these data is that the Velvet domain could be a novel DNA binding domain. Additionally, this analysis reveals a fundamental example of transcriptional rewiring: although Ryp4 regulates acetate utilization genes in related fungi outside of the thermal dimorphs, in H. capsulatum its major role is to regulate the transcriptional response to temperature. As a result, the temperature-dependent circuit elucidated here demonstrates cooperation and regulation between three distinct families of transcription factors: WOPR, Velvet, and Zn(II)2Cys6. Expression of each of the Ryp proteins has been wired to be absolutely dependent on all of the others, so strains that lack any of the regulators fail to undergo a developmental program in response to temperature. This network structure is interesting, in part because of the regulatory implications described above, but also because the relationship between orthologs of these transcription factors is not static throughout the fungal kingdom. For example, Hemiascomycetes have retained WOPR family members but lack Velvet family proteins. These phylogenetic differences provide an opportunity to study how regulatory circuits evolve. Finally, the role of the Ryp proteins in the life cycle of other thermally dimorphic fungi has not been examined. Elucidating the nature of this circuit in these related fungal pathogens will provide insight on the evolution of this temperature-responsive circuit.
H. capsulatum strains G217B (ATCC26032) and G217Bura5 (WU15) were kind gifts from Dr. William Goldman (University of North Carolina, Chapel Hill). ryp1, ryp2, ryp3 T-DNA, and knockdown mutants were previously generated [5],[6]. ryp4 knockdown mutants and wild-type strains with control RNAi plasmids were generated in this study.
All plasmids and primers used in this study are listed in Table S6 and Table S7, respectively. All plasmids were maintained in Escherichia coli DH5α strain and sequenced to ensure no mutation was introduced during the cloning process.
Introduction of integrating and episomal RNAi plasmids into H. capsulatum G217Bura5 strain was done as previously described [6]. ryp4 knockdown strains with integrated RNAi constructs were used in all experiments, except for the plasmid loss experiments where episomal RNAi plasmids were used.
Four independent ryp4 mutants generated with two different episomal RNAi plasmids (pSB30 and pSB31) were grown in HMM broth supplemented with uracil (0.2 mg/ml) at 37°C, 120 rpm with 5% CO2. After 4 wk, an inoculating loop was used to transfer cells from each flask onto HMM+uracil agarose plates followed by incubation at 37°C with 5% CO2. Individual yeast-phase colonies were streaked again and tested for uracil auxotrophy on HMM agarose plates. Representative images of the resulting strains are shown in Figure S3A.
All images were obtained using a Zeiss Axiovert 200 microscope with 40× Phase objective.
Total RNA from H. capsulatum strains was harvested using Trizol (Life Technologies-Invitrogen) following manufacturer's instructions. RNA was treated with RNase-free DNase set (Qiagen) and cleaned using the RNeasy mini kit (Qiagen). 2 µg of total RNA was used in cDNA synthesis using Superscript RT II (Life Technologies-Invitrogen) following manufacturer's instructions. cDNAs were used to amplify RYP1, 2, 3, 4, and GADPH transcripts using OAS1057-58, OAS1942-43, OAS1944-45, OAS3320-31, and OAS1452-53, respectively (Table S7). qRT-PCR reactions were performed in the Stratagene Mx3000P QPCR System (Agilent Technologies) and contained 300 nm of each primer and 0.8× FastStart Universal SYBR Green Master mix (Roche). The transcript levels of RYP genes were normalized to the GAPDH transcript levels.
Total RNA from H. capsulatum strains was harvested as previously described [5]. Reference RNA was generated by mixing total RNA from ryp mutants and wild-type control strains in three different ratios: For microarray experiments with ryp T-DNA mutants, equal amounts of total RNA from each ryp mutant and wild-type were combined. For microarray experiments with ryp1, ryp2, and ryp3 knockdown strains, total RNA from wild-type yeast, wild-type filaments, and ryp knockdown strains were mixed in a 1∶1∶2 ratio. For microarray experiments with ryp4 knockdown strains, total RNA from wild-type yeast, wild-type filaments, and ryp4 knockdown strains was mixed in 1∶1∶1 ratio. For each sample, 15 µg of total RNA was used to generate amino-allyl–labeled cDNA as described previously [42] with the following modifications: Reverse transcription reaction was performed with Superscript RT II (Life Technologies-Invitrogen), and reactions were cleaned using QIAquick PCR clean-up kit (Qiagen) with phosphate wash buffer (5 mM KPO4, pH 8.0, 80% ethanol). cDNA from each ryp mutants and wild-type controls was labeled with Cy5 (GE Healthcare Life Sciences-Amersham) as described [42] and competitively hybridized against the Cy3-labeled pooled reference sample using a H. capsulatum whole-genome 70-mer oligonucleotide microarray. For the experiments with ryp T-DNA mutants, there were four to six replicates for each strain and condition, and for the experiments with ryp knockdown strains, there were three to 12 replicates for each strain and condition. Raw data for all hybridizations (total of 88) performed are available through Gene Expression Omnibus (GEO, www.ncbi.nlm.nih.gov/geo/) with accession numbers GSE46936, GSE46937, GSE46938, and GSE46939 (GEO superseries accession number GSE 47832).
Arrays were scanned on a GenePix 4000B scanner (Molecular Devices) to determine the intensity units in the 635- and 523-nm channels (detecting Cy5 and Cy3, respectively) for each spot on the microarray. Data were analyzed by using GENEPIX PRO version 6.0, NOMAD (http://nomad2.ucsf.edu/NOMAD/nomad-cgi/login.pl), and MultiExperiment Viewer 4.0 (www.tm4.org/mev.html) [43],[44]. After removal of values for flagged spots, background subtraction, and median normalization, the ratio of the median Cy5 intensity/median Cy3 intensity for these spots was used for further analysis. For the genes represented by multiple 70 mer probes, only values for the first probe were used for subsequent analysis. Normalized data was analyzed by Bayesian Analysis for Gene Expression Levels (BAGEL) to obtain relative expression levels for each spot in each condition or mutant [45]. Using the values obtained by BAGEL, we compared any two samples by dividing the value of the first sample by the value of the second sample. To determine the number and identity of genes that changed significantly in expression in any given comparison, we used >2.0-fold change in transcript levels and p value <0.01 as a cutoff criteria. Results of these analyses are given in Tables S1 and S3.
A directed graph was constructed from all “is-a” relationships in the 12/7/2011 version of the Gene Ontology (go_daily-termdb.obo-xml.gz downloaded from the GO Consortium website [http://www.geneontology.org/]). GO-to-gene edges were added from the 12/7/2011 versions of the AspGD [http://www.aspgd.org/] [46], CGD [http://www.candidagenome.org/] [47], and SGD [http://www.yeastgenome.org/] [48] GO association files for A. nidulans (An), A. fumigatus (Af), C. albicans (Ca), Candida glabrata (Cg), and S. cerevisiae (Sc). Gene-to-gene edges were added for An/Af/Ca/Cg/Sc genes to their H. capsulatum G217B (HcG217B) orthologs based on InParanoid mapping of each genome pair (using default parameters with no outgroup). InterPro domains (IPR) from the Pfam, TIGRFAMS, SMART, PANTHER, and Gene3D databases from InterPro version 34 were mapped to HcG217B genes with InterProScan version 4.8 [49]. GO-to-IPR and IPR-to-gene edges were added to the graph by parsing the InterProScan results. An/Af/Ca/Cg/Sc genes with no ortholog in HcG217B were pruned from the graph, as were obsolete GO terms. The set of G217B genes deriving from each GO term was tabulated by traversing the graph in reverse topological sort order, assigning each parent node to the union of the G217B genes spanned by its children. For each query gene set, the GO association graph was pruned to the subset of GO terms spanning the query. GO terms in the resulting subgraph with a single GO term parent and child were considered uninformative relative to the query and were replaced by a direct edge from parent to child. We further removed any GO terms that were associated only with a single gene. The graph was further reduced to only GO terms present in the union of the AspGD/CGD/SGD GO slim sets (12/7/2011 version). For each GO term spanning k genes in the query, the probability of that term spanning at least k terms in a random gene set of the same size was calculated from the hypergeometric distribution as in [50]. The probability calculation was carried out using the phyper function in R [51] with graph operations implemented using NetworkX [52]. The top five terms are reported for each query with no adjustment for multiple hypothesis testing.
Initial sets of Ryp4 homologs were identified using HMMer 3.0 and the Pfam hidden Markov models (HMMs) for the fungal Zn(II)2Cys6 binuclear cluster domain (PF00172.13) and the fungal-specific transcription factor domain (PF04082.13). Each HMM was searched against the BROAD predicted gene sets for Coccidioides immitis RS, Trichophyton verrucosum HKI 0517, Paracoccidioides brasiliensis Pb03, Blastomyces dermatitidis er-3, Fusarium graminearum ph-1, Neurospora crassa OR74A, Magnaporthe oryzae 70-15, Botrytis cinerea B05.10, and Stagnospora nodorum SN15; the SGD curated gene set for Saccharomyces cerevisiae S288C; the AspGD curated gene set for Aspergillus nidulans FGSC; and the GSC predicted gene set for Histoplasma capsulatum G217B using hmmscan. Hits were aligned to each HMM using hmmalign, and neighbor-joining trees were generated from the aligned domain sequences using CLUSTALW 2.1. For PF00172.13, Ryp4 was found in a monophyletic clade, sister to Sip4, with orthologs from each target species except for B. cinerea. For PF04082.13, Ryp4 was found in a monophyletic clade identical to the previous one except for the gain of B. cinerea BC1G_13551 and the loss of Sc Cat8. TBLASTN of the Ryp4 protein sequence against the B. cinerea genome yielded a hit to the Zn(II)2Cys6 domain (E = 2e-29) about 170 bp upstream of BC1G_13551, consistent with an incorrectly predicted 5′ for this gene. Based on these results, the union of pezizomycotina hits was annotated as orthologs of Ryp4. The full-length protein sequences of the Ryp4 orthologs, as well as Cat8 and Sip4, were aligned, and a bootstrapped (n = 1,000) neighbor-joining tree was generated using CLUSTALW.
ryp1, ryp2, ryp3, and ryp4 mutants and wild-type (G217B) cells were grown at 37°C. Cells were fixed and collected as previously described [5]. Frozen cell pellets were ground using a mortar and a pestle in liquid nitrogen or in Retsch Mixer Mill MM 400. Chromatin immunoprecipitation was performed as described [53] with the following modifications: 300 mg of ground samples were vortexed vigorously in lysis buffer [50 mM Hepes/KOH (pH 7.5), 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate supplemented with Halt protease and phosphatase inhibitors (Pierce)] for 3 h with 0.5 mm glass beads to lyse the cells. Following cell lysis, lysates were sonicated in a Diagenode Bioruptor for 30 min (30 s on, 1 min off) to shear DNA. Input DNA was collected and the rest of the sample was subjected to immunoprecipitation. For filamentous-phase samples (ryp mutants), lysates from three identical tubes were combined prior to antibody incubation. Each sample was incubated overnight at 4°C with 20 µg of polyclonal antibodies against portion of Ryp1 (ID:2877, ELDKPFPPGEKKRA), Ryp2 (ID:237, QTNRDYPFYNGPDAKRPR), Ryp3 (ID:338, GIKIPIRKDGVKGPRGGQ), or Ryp4 (ID:8, PPPQQSLQGWSPEEW). On the next day, 40 µl of Protein-A Sepharose 4B Fast Flow (Sigma-Aldrich) beads were added to the lysate-antibody mixture and further incubated for 2 h. Subsequently, beads were collected by centrifugation and washed nine times. Protein–DNA complexes that are bound to the beads were collected by incubating beads at 65°C with elution buffer (50 mM Tris HCl pH 8.0, 10 mM EDTA, 1% SDS). Protein–DNA crosslinks in input and ChIPed samples were reversed overnight at 65°C. Proteins were digested with proteinase K and DNA was purified with phenol∶chloroform extraction as described [53]. Both input and output DNA were amplified and labeled with fluorescent dyes (Cy3 and Cy5) using strand displacement amplification following published protocols [53]. Labeled DNA samples were hybridized following Agilent's ChIP-on-chip protocol onto Agilent 2× 400K arrays comprised of 60 mer oligos tiling the entire H. capsulatum genome at a frequency of one probe per 50 bp. Oligos were selected based on a previously published method [54].
Slides were scanned with an Agilent scanner and raw ratios were obtained with the Agilent ChIP_107_Sep09 protocol. Background subtracted probe intensities were transformed to log2 (red/green) ratios (M) and log2 (sqrt(red*green)) geometric mean intensities (A), excluding probes with intensities below background. The M,A values were fit using the LOWESS algorithm [55] as implemented in R [51], and the fit curve was subtracted from the transformed data to yield lowess normalized log2 ratio values. Normalized spot intensities are available through GEO with accession number GSE47341 (GEO superseries accession number GSE 47832). After normalization, data were analyzed using Mochiview [56]. Data from three biological replicates for each Ryp-IP were analyzed together to identify peaks. ChIP-chip experiments done with ryp mutants were used as a negative control in this analysis. Default parameters of Mochiview peak extraction were used with the exception of increasing total random samples to 100,000 and maximum random samples to 100 (p value <0.001). Median+Interquartile range (IQR) was used as a threshold to filter extracted peaks. Peaks that were eliminated based on the noisy enrichment values in ryp mutants were included back if there was >2-fold difference between wild-type and ryp mutant enrichment ratios. Additionally, peaks that were greater than a median value for a given Ryp event and had another Ryp event greater than median+IQR were also included in the finalized list of events (Table S2). ChIP enrichment ratios plotted in Figures 3 and 4 were generated using ChIP tracks that were a combination of three biological replicates for each Ryp-IP.
Each ChIP event was mapped to specific genes using an H. capsulatum G217B strain validated gene set that was defined previously using gene expression and sequence conservation criteria [28]. Genes that have a ChIP event in their 5′ region within a 10 kb distance from the center of the peak were listed as target genes (Table S2). In addition to target genes found in the validated gene set, some additional target genes were included if they had detectable expression levels in the whole-genome transcriptional profiling experiments performed in this study.
Signal peptide and transmembrane helix prediction algorithm Phobius 1.01 [57] was run with default parameters on the full G217B predicted protein set, excluding two sequences with predicted internal stops. Three nondisjoint prediction sets were derived from the Phobius output: genes predicted to have a signal sequence, genes predicted to have at least one transmembrane helix, and the intersection of the previous two sets. For each phobius prediction set spanning k genes in a given Ryp associated set, the probability of that term spanning at least k terms in a random gene set of the same size was calculated from the hypergeometric distribution as in [50]: P(k≤X) = 1−sum∧[k−1]_[i = 0] c(M,i)*c(N−M,n−i)/c(N,i), where c is the binomial coefficient function, N is the size of the full gene set, n is the size of the Ryp associated set, and M is the size of the phobius prediction set. The probability calculation was carried out using the phyper function in R [51].
Each set of Ryp ChIP events was randomly split into two sets using Mochiview. Each set was subjected to motif finding algorithms using Mochiview and Bioprospector [56],[58]. For each set, the top five motifs identified using Mochiview and the top three motifs identified using Bioprospector were further analyzed for specificity in the corresponding randomly split set. Motif searches were carried out using MAST [59] as implemented in version 4.8.1 of the MEME suite [60]. The MAST -hit_list parameter was used to yield all nonoverlapping motif instances in the query sequences with no adjustment for query length or number of motif hits (such that a reported p value reflects only the alignment of the motif instance to the query matrix). For ROC plots, the motif threshold (-mt) was set to 0.01 in order to explore a wide range of possible parameter values. For genome-wide searches, the motif threshold was set to the values determined from the ROC plots using false positive rate of 10% as a cutoff (corresponding to p values of 2.08e-04 for Motif A and 9.18e-05 for Motif B). Motif locations identified in the genome are given in Table S5.
For expression of the Ryp1-N-terminus, Ryp1, Ryp2, and Ryp3 proteins for EMSA experiments, we subjected pSB122, pSB124, pSB128, and pSB130, respectively, to TNT Coupled Wheat Germ Extract systems (Promega) following the manufacturer's instructions. For purification of Ryp proteins, eight 50-µl reactions were pooled and diluted 10-fold in binding buffer (10 mM Tris-HCl, pH 8.0, 50 mM KCl, 5% glycerol, 1% TritonX-100, and 20 mM imidazole) supplemented with HALT phosphatase and protease inhibitors (Pierce). For each sample, 100 µl of Ni-NTA agarose beads (Qiagen) was washed in binding buffer and incubated with extracts for 1 h at 4°C. Following the incubation, beads were washed once with binding buffer, and once with binding buffer with no detergents. His-tagged proteins were recovered with elution buffer (10 mM Tris-HCl, pH 8.0, 50 mM KCl, 5% glycerol, 250 mM imidazole). The presence of His-tagged Ryp proteins was confirmed by SDS-PAGE analysis and Western blotting. Wheat germ extract with no DNA template was subjected to a similar purification process and use as a control in mobility shift assays.
5′-IRDye800-labeled Motif A and Motif B probes were prepared by annealing 5′-IRDye800-CBP1-MotifA-Fwd and CBP1-MotifA-Rev, and 5′-IRDye800-CBP1-MotifB-Fwd and CBP1-MotifB-Rev, respectively, in 10 mM Tris-HCl, pH 7.9, 50 mM NaCl, 10 mM MgCl2, and 1 mM DTT. Nonlabeled competitor probes were prepared similarly with nonlabeled oligonucleotides (Table S7). Two µg of each purified protein (Ryp1-N-terminus, Ryp1, Ryp2, Ryp3, or control extract) and 1 nM of labeled probes were mixed in binding buffer (10 mM Tris-HCl, pH 8.0, 50 mM KCl, 5% glycerol, 1 mM EDTA, 0.5 mM DTT, 100 ug/ml BSA, and 25 ug/ml poly(dI:dC)) and incubated for 30 min at room temperature. Reactions were separated on 6% DNA retardation gels (Invitrogen) in 0.5× TBE buffer. Mobility shifts were visualized and analyzed using the ODYSSEY imaging system (LI-COR Biosciences).
Wild-type (G217B) cells grown to late log phase at 37°C were harvested by filtration, and the pellet was frozen in liquid nitrogen. Whole cell extracts were made by cryogrinding the pellet in Retsch Mixer Mill MM 400. Co-immunoprecipitation experiments were performed using the Dynabeads Co-immunoprecipitation kit from Invitrogen following the manufacturer's instructions. Briefly, 100 ug of polyclonal α-Ryp2 (ID:387, SQSAGHMQSPSQVPPAWG) or α-Ryp3 (ID:356, SHGSKGQDGEGEDWENEG) antibodies were covalently linked to 5 mg of magnetic beads using Dynabeads Antibody Coupling kit. To prepare cell lysate, 2 g of ground samples were mixed with lysis buffer, vortexed, and spun down. Then, supernatant was incubated with 5 mg of antibody-coupled magnetic beads for 8 h at 4°C. After multiple washes, protein complexes that are bound to antibody-coupled beads were eluted in low pH buffer provided by the manufacturer. IPs with ryp2 and ryp3 mutants grown at 37°C and no antibody control were performed similarly. All fractions were separated by SDS-PAGE, visualized by silver staining, and analyzed by Western blotting using standard procedures. Polyclonal α-Ryp1 (ID:3873, ASSYQPGPPASMSWNTAATG), α-Ryp2 (ID:387, SQSAGHMQSPSQVPPAWG), or α-Ryp3 (ID:356, SHGSKGQDGEGEDWENEG) antibodies were used to detect Ryp proteins.
β-galactosidase assays were performed as previously described [61]. At least three independent isolates of each S. cerevisiae strain were grown to late log (for the yeast-two-hybrid assay) or stationary (for the in vivo transcriptional activation assay) phase. Each isolate was assayed in quadruplicate, and the results of representative isolates are shown in Figures 5, 7, and S7.
|
10.1371/journal.pgen.1003616 | The Conserved ADAMTS-like Protein Lonely heart Mediates Matrix Formation and Cardiac Tissue Integrity | Here we report on the identification and functional characterization of the ADAMTS-like homolog lonely heart (loh) in Drosophila melanogaster. Loh displays all hallmarks of ADAMTSL proteins including several thrombospondin type 1 repeats (TSR1), and acts in concert with the collagen Pericardin (Prc). Loss of either loh or prc causes progressive cardiac damage peaking in the abolishment of heart function. We show that both proteins are integral components of the cardiac ECM mediating cellular adhesion between the cardiac tube and the pericardial cells. Loss of ECM integrity leads to an altered myo-fibrillar organization in cardiac cells massively influencing heart beat pattern. We show evidence that Loh acts as a secreted receptor for Prc and works as a crucial determinant to allow the formation of a cell and tissue specific ECM, while it does not influence the accumulation of other matrix proteins like Nidogen or Perlecan. Our findings demonstrate that the function of ADAMTS-like proteins is conserved throughout evolution and reveal a previously unknown interaction of these proteins with collagens.
| Cellular adhesion and tissue integrity in multicellular organisms strongly depend on the molecular network of the extracellular matrix (ECM). The number, topology and function of ECM molecules are highly diverse in different species, or even in single matrices in one organism. In our study we focus on the protein class of ADAMTS-like proteins. We identified Lonely heart (Loh) a member of this protein family and describe its function using the cardiac system of Drosophila melanogaster as model. Loh constitutes a secreted protein that resides in the ECM of heart cells and mediates the adhesion between different cell types - the pericadial cells and the cardiomyocytes. Lack of Loh function induces the dissociation of these cells and consequently leads to a breakdown of heart function. We found evidence that the major function of Loh is to recruit the collagen Pericardin (Prc) to the ECM of the cells and allow the proper organization of Prc into a reticular matrix. Since the function of Loh homologous proteins in other systems is rather elusive, this work provides new important insights into the biology of cell adhesion, matrix formation and indicates that ADAMTS-like proteins might facilitate an evolutionary conserved function.
| The establishment and maintenance of extracellular matrices (ECM) are important tasks to allow proper organ function in metazoans. Among other factors, changes in ECM composition, turnover and homeostasis are crucial mediators of human cardiovascular disease leading to life threatening conditions and premature death. The ECM allows cells to resist mechanical forces, protects complex tissues from being damaged and promotes specific physical properties like elasticity or stiffness in order to maintain organ functionality. While the composition of the ECM is very complex and extremely variable the basic structural constituents can be grouped as collagens, glycoproteins and proteoglycans, which are highly conserved throughout metazoan species [1]. Consequently, defects in ECM proteins or matrix composition cause major developmental defects and strongly contribute to prevalent human disease like fibroses or cancer [2]. During the last years fibrotic disease and mutations in various ECM proteins were correlated to cardiovascular disease. For example mutations in human Col4a1 cause the weakening of the major vasculature leading to life threatening aneurysms or stroke [3] while mutations in murine Col4a1 and Col4a2 induce vascular defects causing internal bleedings and prenatal lethality [4]. Even more recently ADAMTS-like (ADAMTSL, A Disintegrin and Metalloprotease with Thrombospondin repeats) proteins have gained significant importance in the understanding of certain types of fibrillinopathies [5], [6]. Mutations in human ADAMTSL4 were identified in patients suffering from isolated ectopia lentis (EL), a recessive disorder of the occular lense [7], [8] and, more severely, aberrations in ADAMTSL2 cause geleophysic dysplasia a syndrome which, amongst others, manifests in the thickening of the vascular valves and progressive cardiac failure causing premature death [9]. Unfortunately, despite the pathological mutations no ADAMTSL alleles in genetically treatable model systems were described so far.
In the present study we use Drosophila melanogaster as a model of ECM function in the cardiac system. In Drosophila the maintenance of cardiac integrity is of great importance, since no mechanisms of cardiac cell replacement or tissue repair exist. A variety of mutations in ECM genes have been analyzed with respect to their function in different tissues and processes like neurogenesis, muscle attachment, wing development and others [10]–[12]. Cardiogenesis in the fly embryo depends on several ECM components including the evolutionarily conserved toolkit of proteins forming the basement membrane. The basement membrane constitutes a specialized type of ECM consisting of Laminins, Collagen IV, Perlecan and Nidogen found at the basal side of epithelial cells [13]. The interaction of laminins with cellular receptors like integrins or dystroglycan and its self-assembly into a higher meshwork forms the initial step of basement membrane formation in animals [14], [15]. Consequently, mutations in any of the four laminin encoding genes in Drosophila lead to severe embryonic cardiac defects. For example loss of lanB1, encoding the only β-subunit of the laminin trimer, prevents the accumulation of collagen IV and perlecan towards cardiac cells, while mutations in lanA and lanB2 (encoding the α3,5-subunit and the γ-subunit, respectively) cause the detachment of pericardial cells, a specific type of nephrocytes in arthropods, from the heart tube [14], [16]–[18]. The highly abundant proteins forming the basement membrane have in common that they are distributed ubiquitously and cover all internal organs of the fly [14], [19].
Compared to that the cardiac ECM is unique, since it contains the collagen Pericardin (Prc), which is rather specifically decorating the heart tube [20], [21]. Prc displays certain homologies to mammalian collagen IV and was shown to be crucial for heart morphogenesis and cardiac cell to pericardial cell adhesion [20], [22]. However, the question of how Prc accumulates in a cell specific manner in the fly embryo or how specific matrices are specified in the rather open body cavity of insects in general was not addressed in detail so far. Here we introduce the gene lonely heart (loh), which is crucial to maintain cardiac integrity during postembryonic developmental stages. We show that Loh is a member of the ADAMTSL protein family and constitutes the essential mediator of Prc accumulation and matrix formation already in embryonic cardiac tissue. ADAMTSL proteins belong to the evolutionary conserved family of ADAM proteases with the exception that these proteins lack a proteolytically active domain in their primary sequence and therefore its function is unclear [5], [6]. We found evidence that Loh is sufficient to specifically recruit Prc to the ECM of different tissues indicating that Loh regulates the assembly of tissue and organ specific matrices. This is of great interest since the composition of the ECM determines its mechanical properties crucial for correct organ function and cellular behavior [23]. We also address the physiological relevance of cardiac integrity and show that lack of either loh or prc prevents proper blood circulation in the animals and cause a reduction of the fly's life span. The findings presented in here demonstrate that mutations in ADAMTSL proteins lead, like in human disease, to progressive heart failure and premature death in flies, strongly arguing for an evolutionary conserved function.
In order to identify novel mediators of cardiac function we screened a set of pupal lethal EMS induced mutants, known as the Zuker collection, for the presence of postembryonic cardiac malformations [24]. To mark all cells contributing to the mature heart we introduced the previously described handC-GFP reporter into each individual mutant strain [25]. We identified a single allele, lonely heart (loh1), showing a strong detachment of pericardial cells from the heart tube during larval stages (Figure 1 and Figure S1A, B). To map the mutation to the genome we introduced the loh1 allele to a collection of genomic deficiencies and assayed the progeny for the presence of the pericardial cell detachment phenotype. The allele failed to complement the deficiencies Df(2L)Exel7048, Df(2L)BSC453 and Df(2L)BSC144 but complements Df(2L)BSC209 (Figure S1C–F). This allowed us to narrow down the location of the mutation to a 14 kb genomic region at band 31E3-4 containing three open reading frames (Figure S1I). Since EMS is known to promote secondary hits on the same chromosome we decided to assay existing alleles of these three genes for the presence of the pericardial cell detachment phenotype. We were able to identify two alleles, MB05750 and MI02765, that are allelic to loh1 and Df(2L)Exel7048 and produce the heart phenotype in transheterozygous condition (Figure 1A–D and Figure S1G, H). Both mutations were induced by the insertion of minos elements within the locus of the previously uncharacterized gene CG6232 [26], [27]. Based on sequence predictions CG6232 encodes an ADAMTS-like (A Disintegrin and Metalloproteinase with Thrombospondin repeats) protein, containing several Thrombospondin type 1 repeats, a central ADAM-spacer domain and a C-terminal Protease and Lacunin (PLAC) domain (Figure S1J). The primary sequence of Loh/CG6232 shows high homologies to mammalian ADAMTSL6, known to promote the formation of fibrillar matrices in mice [28].
During a parallel reverse genetic approach we also tested transposon induced alleles affecting known ECM genes for the appearance of late cardiac defects. We identified the allele MB03017 carrying a minos element in the pericardin (prc) locus. Homozygous prcMB03017 and transheterozygous prcMB03017/Df(3L)vin6 animals display a strong pericardial cell detachment phenotype similar the loh phenotype (Figure 1E, F and Figure 2E). The Prc protein constitutes a rather heart specific collagen, which shows homologies to vertebrate collagen IV [22]. Previous studies implicated Prc to be involved in dorsal closure as well as cardiogenesis [20]. However, no gene specific mutant was available so far.
To investigate the adhesion defects arising in both loh and prc mutants in more detail we analyzed the morphology of the heart at different developmental stages. During embryogenesis the heart tube arises from two bilateral primordia and forms a simple tube at the dorsal midline. Determination and migration of heart precursor cells is not affected in either lohMB05750/Df(2L)Exel7048 or prcMB03017/prcMB03017 mutant animals (Figure 2A, D and G). During larval development the pericardial cells irreversibly detach from the heart tube with the phenotype becoming fully visible in third instar larvae (Figure 2B, E and H). The loss of cardiac integrity in both mutants does not constrain the development into adult animals and we could detect the pericardial cell detachment phenotype in pharate adult animals, which further develop into viable and fertile flies (Figure 2C, F and I). These findings show that the phenotype arises progressively during development and indicate that proper heart function is not essential for development into the imago. Of note the alleles loh1 and lohMI02765 cause larval lethality in homozygous condition, while the alleles are viable in transheterozygous combination indicating second site mutations or yet unknown dominant effects of the mutated proteins. Since lohMB05750 and prcMB03017 animals are homozygous viable and show the pericardial cell detachment phenotype all experiments predominantly focus on these two alleles.
Postembryonic pericardial cells are enclosed by a dense network of Prc fibers and connected to the alary muscles (Figure 2J–K). Since the heart tube and the alary muscles are not connected via direct cell-to-cell contacts this Prc network is likely to be a fundamental structural component to suspend the heart to the body cavity [29]. To evaluate the adhesion of the heart tube to the alary muscles in more detail we stained transheterozygous lohMB05750/Df(2L)Exel7048 and prcMB03017/Df(3L)vin6 larvae for F-actin and βPS integrin (Figure 2L–N). The detachment of pericardial cells also ruptures the connection between the alary muscles and cardiomyocytes demonstrating that the lack of pericardial cell adhesion consequently lead to a breakdown of the heart's suspension towards the epidermis. Furthermore, the morphology of the cardiomyocytes itself is dramatically altered in lohMB05750/Df(2L)Exel7048 and prcMB03017/Df(3L)vin6 mutants (Figure 2O–Q). While in the wild type cardiomyocytes show a defined arrangement of F-actin fibers in a circular fashion mutant cells exhibit an uncoordinated distribution of actin fibers and an altered cell shape. Since the arrangement of actin fibers might be a secondary effect of a changed cardiac cell polarity we stained mutant embryos for the polarity markers FasIII and αSpectrin (Figure S2A–L). Neither loh nor prc mutant hearts displayed changes in cell polarity proving that the changed actin arrangement is an effect of the defective cellular adhesion.
We next elucidated how heart beat is influenced in the mutants. For this purpose the beating pattern of the heart was recorded in semi-dissected third instar larvae (Movies S1, S2, S3) [30]. Wild type heart beat follows a very regular pattern and the heart walls display systolic and diastolic movements (Movie S1). Compared to that the beating pattern in lohMB05750/Df(2L)Exel7048 and prcMB03017/Df(3L)vin6 mutant larvae is dramatically altered. The disorganized actin fibers cause a changed contraction movement of the whole organ along the posterior-anterior axis (Movie S2 and Movie S3). In addition no systole and diastole are detectable already indicating that the pumping performance of the organ is altered.
To evaluate whether the disruption of heart architecture and the changed beating pattern impairs heart functionality we analyzed the capability of mutant hearts to provide circulatory activity. To visualize the hemolymph flow by dye angiography we injected a fluorescent tracer into the abdomen of adult animals shortly before eclosion (pharate adults) and semi-quantified the pumping capacity of the dorsal vessel by measuring the tracer accumulation within the head (Figure 3A–C) [31]. To verify the reliability of the technique a control strain that does not display any cardiac defects was tested and showed a strong accumulation of the tracer in the head (Figure 3B–C and Movie S4). In contrast homozygous prcMB03017 and lohMB05750 mutant animals displayed a dramatic reduction or total absence of dye accumulation within the examination time, which proves that the observed disruption of heart integrity directly influences the ability to promote circulatory activity (Figure 3C).
Since it is known that heart failure can cause a significant reduction of Drosophila's life span [32], [33] we tested whether the isolated alleles show a direct effect on adult survival. As a wild type control we used the white1118 strain, because this genotype resembles the genetic background of both minos insertion strains. Wild type flies (white1118) revealed an average life time of 46 days, while the mean life span of homozygous lohMB05750 and prcMB03017 animals was decreased by 26% (34 days) or 46% (25 days), respectively (Figure S3). This strongly argues that impaired cardiac function in the mutants reduces the survival of the animals.
We investigated the temporal expression pattern of loh and prc by developmental Northern blots. The loh locus encodes two transcripts - a longer isoform A (3081 bp predicted) and a shorter isoform C (2131 bp predicted) (Figure 4A). While isoform A constitutes the major transcript during embryogenesis, isoform C becomes additionally expressed during the first and second larval stage (L1 and L2). Later on expression declines and becomes weakly re-activated during pupal and adult stages. Compared to loh the temporal expression profile of prc was found to be remarkably similar (Figure 4A). A single transcript (5535 bp predicted) becomes expressed from the embryo to L2 and declines in L3. During metamorphosis expression re-initiates and lasts until adulthood. In order to reveal if both loh isoforms are essentially needed to ensure proper heart integrity we expressed two independent gene specific hairpins either effecting only isoform A (loh-IRNIG6232-2) or both isoforms (loh-IRVDRC31020) under the control of handC-Gal4 to knock down the gene's expression (Figure 4B). Expression of both hairpins causes a pericardial cell detachment phenotype. However, since expression of the loh-IRNIG6232-2 hairpin, which only targets isoform A, resulted in a detachment phenotype (Figure S4) we concluded that isoform A constitutes the relevant one for the observed adhesion defect.
To investigate the effect of the isolated mutations on the expression level we analyzed the total protein amounts by immunoblotting (Figure 4D, E). Therefore we raised a specific peptide antibody recognizing both Loh isoforms. In embryonic extracts the antibody detects a single protein band corresponding to isoform A. The band runs slightly higher compared to the predicted molecular mass of 100 kDa, most likely due to posttranslational modifications (Figure 4D). The protein is absent from extracts of homozygous Df(2L)Exel7048 embryos proving the specificity of the antibody. Significantly, the protein is also undetectable in extracts of homozygous lohMB05750 embryos. RT-PCR analysis proved that lohA transcripts are severely reduced but not absent in these animals (Figure S4A), obviously leading to massively decreased protein levels. Similarly, Prc protein could be detected in extracts of different developmental stages in the control, but is absent from homozygous mutants (Figure 4E).
Given the similar phenotypes of the mutants we sought to analyze the spatial expression pattern of both genes. Transcripts of loh and prc can be detected from embryonic stage 13 onwards until the end of embryogenesis in cardioblast and pericardial cell precursors (Figure 5A–F), where loh seems to be more prominently expressed in the ventricle of late stage embryos (Figure 5C). Additionally, loh transcripts were detected in the chordotonal organs, while prc is expressed by the oenocytes. Since it is known that prc is only expressed by a subset of cardiac cells we analyzed the expression of loh mRNA in combination with the cardiac cell markers Tinman and odd skipped-lacZ [20], [34], [35]. loh transcripts are expressed by both cell types demonstrating that most cardioblasts and pericardial cells contribute to the gene's expression (Figure 5G, H). As previously reported, Prc protein distributes predominantly along the basal side of the cardiomyocytes where it co-localizes with the collagen IV fusion protein Vkg::GFP (Figure 5I) [20], [36]. Strikingly, Loh co-localizes with Vkg::GFP as well as Prc, demonstrating that it constitutes an integral part of the basal cardiac ECM (Figure 5J–K). The detected signal was considered to be specific since it follows the observed mRNA pattern and is undetectable in homozygous Df(2L)Exel7048 embryos (Figure S5A–C). The expression of Loh and Prc supports a function in mediating the adhesion between pericardial cells and cardiomyocytes in the mature heart, while the observed co-localization throughout the whole embryo indicates a cooperative function (Figure 5L).
The data presented so far pointed us to the question if Loh and Prc act cooperatively in the cardiac ECM. To test if the proteins affect each other we analyzed the localization of Prc in loh mutant background and vice versa (Figure 6A–O). In homozygous lohMB05750, lohMI02765 and loh1 embryos Prc becomes normally secreted but strikingly fails to assemble properly in between the pericardial cells and the heart (Figure 6A–F and Figure S6A, B). While in the wild type Prc organizes into a proteogenic sheet at the basal side of the cardiomyocytes this regular distribution is completely disrupted in loh mutant embryos (Figure 6A–F). We also tested whether impaired loh expression affects other ECM proteins like Laminin, Nidogen or Perlecan (Figure 6B, E and Figure S6F, G and I,J). The expression and distribution of all tested proteins was unchanged in loh mutant animals indicating that Loh specifically regulates the correct accumulation of Prc but is not needed for ECM formation in general. The other way around the lack of Prc in homozygous prcMB03017 embryos does not affect the localization of Loh (Figure 6J–O) or any other tested ECM protein demonstrating that the function of both proteins is not mutual (Figure 6G–N and Figure S6H, K). To prove that the phenotypes in lohMB05750 and prcMB03017 definitely arise from the inserted transposons we generated revertants by precise excision of the minos elements [26], which was verified by PCR and subsequent sequencing (Figure S6C–E). The precise remobilization of both transposons lead to a restored Prc expression and distribution in both revertants demonstrating that the mutations are gene specific.
To study the effect of loh and prc mutants on heart cell morphology in more detail we investigated TEM cross sections of wild type and homozygous lohMB05750 and prcMB03017 embryos (Figure 6P–V and Figure S6L–N). Like in wild type the cardiomyocytes are localized along the dorsal midline at the end of embryogenesis in both mutants showing that dorsal closure is not affected (Figure S6L–N). However, frequently the cardiomyocytes in homozygous prcMB03017 mutants fail to seal the lumen properly at the ventral side of the heart tube (Figure S6N). Staining against the ligand Slit, which is involved in heart lumen formation did not reveal any changes in its distribution indicating that the Slit/Robo signaling cascade is not affected (Figure S6O–Q) [37]. Most importantly, the luminal and basal membranes of the cardiomyocytes are covered by a distinct basement membrane in both homozygous mutants supporting the immunocytochemical data (Figure 6P–R). Measuring its thickness does not reveal any significant changes (Figure 6S). However, even if the pericardial cells are not fully detached from the embryonic heart, small gaps between the cells and rupture of the connecting ECM are detectable (Figure 6T–V). Taken together these data demonstrate that Prc and Loh are essential to maintain pericardial cell to cardiomyocyte adhesion and heart integrity but are not involved in ECM formation in general.
Hypothetically the open circulatory system of insects would allow ECM proteins to be expressed by a certain cell type, then be distributed over the blood flow and finally become recruited by specific receptors expressed on the target cells. The embryonic expression pattern of loh and prc argue that both proteins are primarily produced locally by heart cells and become secreted into the cardiac ECM. To analyze the expression of prc during later stages we used the previously described prc-Gal4 driver to express GFP and found that it exactly mimics the expression pattern of prc in the embryo (Figure 7A) [20]. Upon larval hatching the driver becomes strongly activated in the fat body (Figure 7B) raising the question, whether the reporter mimics the endogenous prc expression. To test if Prc becomes produced by adipocytes we trapped the protein by inhibiting the protein secretion machinery of the cell by knocking down the expression of the small GTPase Sar1, which is essential for the establishment of COPII coated vesicles and protein secretion (Figure 7C, D) [38]. Compared to wild type, adipocytes of prc>sar1-IR first instar larvae displayed a strong accumulation of intracellular Prc protein unambiguously demonstrating that it becomes expressed by the larval fat body. To estimate the contribution of fat body derived Prc to the total amount of the protein made, we knocked down prc expression either in heart cells alone (handC-Gal4) or in both heart and fat body (prc-Gal4) and detected the protein by immunoblotting (Figure 7E). The specificity of the knock down was ensured by the use of two independent hairpins (Figure 3C). Prc levels are not markedly changed in handC>prc-IR third instar larvae, while the protein is nearly undetectable in extracts of prc>prc-IR animals illustrating that most of the larval Prc protein becomes secreted by adipocytes. Finally, the pericardial cell detachment phenotype could be induced by knocking down prc expression using both drivers (Figure S7). However, the penetrance of the induced pericardial cell detachment phenotype is strikingly higher if the knock down was mediated via prc-Gal4 (Figure 7F), showing that the protein secreted from adipocytes indeed contributes to pericardial cell adhesion. From these experiments we conclude that the major source of Prc in larvae is non-cardiac tissue. Nevertheless, locally produced Prc contributes to proper heart integrity, since heart specific knock down of Prc expression does induce the detachment phenotype as well. Taken together these experiments prove a developmental switch in Prc expression with embryonic Prc being locally produced by cardiac cells and during later stages becoming mainly secreted by the fat body (Figure 7G). Furthermore, the integration of fat body derived Prc into the cardiac ECM is essential to promote organ integrity.
Although Prc is produced by adipocytes, the protein is not incorporated into the ECM of the fat body indicating that these cells lack specific adhesion properties for Prc (Figure 8A). We found that in third instar larvae the protein almost exclusively accumulates around tissues that initially expressed loh during embryogenesis, but is nearly absent from other mesodermal tissues. From these observations we concluded that Loh might act as a mediator or receptor of Prc matrix formation in Drosophila. To test if Loh is indeed sufficient to induce the formation of Prc matrices we expressed the protein ectopically either in adipocytes or myocytes by using prc-Gal4 or mef2-Gal4, respectively. Even if some sole Prc fibers can be found along both cell types these organs are not naturally covered by a Prc matrix (Figure 8A). Ectopically expressed LohA protein becomes secreted from both cell types and localizes around the cells (Figure 8C). The protein is retained at the cell surface of adipocytes or myocytes indicating proper localization in the ECM. Upon expression in the fat body, LohA distributes along the whole organ showing a higher accumulation at cellular contacts. Similarly, LohA ectopically expressed by myocytes distributes along the whole myotube with higher accumulation at the muscle tendons (Figure 8C, inset). Most importantly, we found that LohA expression strongly induces the formation of an ectopic proteogenic Prc network around both cell types (Figure 8C). Adipocytes and myocytes ectopically expressing LohA are tightly covered by Prc fibers, which are interconnected to each other and form a dense meshwork. Immunoblot analysis on whole extracts revealed that the overall amount of Prc was not changed in these animals (Figure S8A), demonstrating that ectopic LohA expression leads to a re-direction of Prc protein. To evaluate if Loh acts within the ECM we ectopically expressed a secretion defective version of the protein (Figure 8B), lacking the N-terminal signal peptide. The mutated protein localizes to the nuclei of the cells and fails to recruit Prc to the target matrix demonstrating that LohA has to be secreted in order to act as an initiating factor of Prc matrix formation.
To evaluate if both proteins co-localize in such artificial matrices we counterstained dissected prc>LohA third instar larvae for Loh and Prc (Figure 8D,E). High resolution images of dissected fat bodies showed that ectopic LohA distributes as a very faint network at the surface of adipocytes and clusters in a pointy fashion along the cell contacts (Figure 8D) but does not completely co-localize with the recruited Prc fibers. Single slices and optical cross sections further demonstrate that Loh co-localizes with Prc at the anchoring points of the Prc network (Figure 8E), indicating that Loh might connect the root of each Prc fiber to the cell surface. Eventually, co-immunoprecipitation experiments using protein extracts isolated from prc>LohA adults proved a either direct or indirect biochemical interaction of both proteins (Figure 8F). In the respective experiments Prc co-precipitated if Loh was pulled down and vice versa. Based on these findings we hypothesize that Loh acts as a linker protein allowing Prc to interact with the cell surface, and wondered if Loh co-localizes with specific cell surface receptors. We found that LohA co-localizes to βPS integrin in adipocytes of prc>LohA third instar larvae (Figure S8B) tending us to speculate that LohA binds to integrin receptors, which has to be proven by further experiments. In summary we found that LohA is a crucial and sufficient mediator of Prc matrix formation, very likely acting by interconnecting Prc with the cell's ECM.
In this study we demonstrate that the Drosophila ADAMTSL protein Loh constitutes an unique protein of the cardiac ECM, essentially mediating cell adhesion and matrix formation. Loh is the first protein of its family identified and characterized in depth in flies. We isolated three independent alleles of the gene, all displaying the very same phenotype - the detachment of pericardial cells from the contracting heart tube during larval stages. Thus, the gene loh constitutes a novel and essential mediator of heart cell adhesion and cardiac function. Surprisingly, impaired heart function does not hamper proper development into adult animals but significantly reduces life span. This might be explained by the fact that oxygen transport and blood flow is uncoupled in insects and therefore a reduced hemolymph circulation might not immediately result in cytotoxicity. Furthermore, the open body cavity of the larvae might also allow a distribution of hemolymph independently of a pumping organ supporting the finding that larvae seem not to achieve any drawbacks by the loss of heart function.
Based on the primary sequence the domain architecture of Loh is extremely similar to that of vertebrate ADAMTSL6 and is likely to be its ortholog. Furthermore, ADAMTSL6 is the only protein of this family known to produce two transcriptional isoforms from one gene locus. In contrast to Loh the shorter ADAMTSL6 isoform was found to be functional in organizing the ECM in mice [28]. Our data demonstrate that LohA, the larger protein, is functional and sufficient to mediate matrix formation in Drosophila while the role of the shorter isoform C remains elusive by now. However, since the lohC transcript is not expressed during embryogenesis, the critical time window of loh function, we exclude any role of LohC in mediating cardiac ECM formation.
By testing different ECM proteins we demonstrated that Prc, a collagen with a very restricted distribution in the animal, is particularly affected in all isolated loh mutant alleles, emphasizing the specific function of Loh to promote Prc matrix formation. Consequently, we isolated the first prc mutant allele, which phenocopies the cardiac defects found in loh mutant strains. In loh mutant animals Prc mislocalizes along the heart already during embryogenesis, leading to a progressive loss of tissue integrity, which eventually causes the observed collapse of the heart tube and an abolishment of cardiac activity. The main function of both proteins is therefore the mediation of cellular adhesion between the heart, the pericardial cells and the alary muscles which further connect the whole organ system to the body cavity.
In addition to the cell adhesion defects we also found that the process of heart lumen formation was impaired in prc but not loh mutants. Since we have not followed up the details of this phenotype the role of Prc in lumen formation remains elusive for now. However, the data implicates that the presence of Prc is critical to allow cardioblasts to seal the lumen correctly, while the correct localization of Prc into the matrix seems not to be essential for this process.
Analyzing the embryonic and larval expression patterns of loh and prc revealed that both genes are predominantly active during the growing stages of the animal and become deactivated after the heart has grown to its final size. In the embryo, both genes are transcribed in either the same or very proximate cells indicating that the proteins are not distributed over longer distances once they are secreted. Importantly, the final localization of Prc therefore mainly follows the expression of loh. This can be seen best in the oenocytes of the embryo, where Prc becomes secreted but later on mainly localizes to the overlying chordotonal organs that in turn express loh. Thus, loh expression is a prerequisite for the successful establishment of a Prc matrix. This local protein distribution changes during larval stages. As demonstrated by an inhibited secretion in adipocytes of prc>sar1-IR animals, Prc becomes strongly expressed by the fat body during early larval stages. Hence, the protein becomes distributed over longer distances in the larva but still decorates organs and tissues that initially expressed loh. Based on these data, we provide a conceptual model (Figure 7G) in which Loh predetermines the ECM to allow Prc to become coupled to the cell surface and to be organized into a reticular matrix. Previously it was shown that Collagen IV, the major collagen in the basement membrane, becomes also secreted by adipocytes and distributes through the hemolymph [39]. We can now prove that Prc as a second collagen is also synthesized by the larval fat body, which enhances the importance of this organ for ECM biogenesis. The developmental change in prc expression might therefore be explained by the ongoing differentiation of pericardial cells into mature nephrocytes during larval stages. While embryonic pericardial cells are able to secrete large amounts of protein into the extracellular space, the major function of pericardial nephrocytes is endocytosis [40], thus requiring adipocytes to take over Prc production. Finally our results show that the cardiac matrix is maintained during larval growing phases presumably by the consecutive incorporation of fat body derived Prc.
The ectopic expression of Loh showed that the secreted protein is readily incorporated into different matrices raising the question how Loh itself interacts with the ECM in general. At the moment it is not fully understood if ADAMTSL proteins interact with miscellaneous ECM components or require specific cell surface receptors. Based on the spatial proximity of Loh to βPS integrin we speculate that Loh may interact with integrin receptors and link these to Prc bundles, thereby promoting the connection of the Prc network to the cell surface. This idea is supported by the observed changes in fiber orientation of mutant cardiomyocytes. Since it is known that integrins are connected to the underlying Z-disks of muscle cells by a structure called the costamere [41] we propose that lack of integrin-ECM binding induces the redistribution of myofibrils. However, there is no evidence of an interaction between ADAMTSL proteins and integrins or any other cellular receptor so far. Nevertheless, in such a model Loh would allow the specific binding of specialized ECM molecules to only some unique matrices. Since Drosophila possesses only two β integrin subunits the number of α/β-dimers is limited and the use of Loh as an adapter molecule increases the diversity of matrix composition and opens up the possibility to create sub-functional matrices. Furthermore, integrin mediated binding seems to influence the correct assembly of Prc since previous findings already showed that lack of αPS3- or βPS integrin can interfere with the distribution of Prc and induce pericardial cell detachment phenotypes [42].
In addition to a receptor mediated ECM incorporation of Loh, binding might also be achieved by some or all of the five TSR1 domains found in the primary sequence of the protein. Previously it was demonstrated that ADAMTS(L) proteins can bind to the ECM via the various TSR1 motifs that interact with glycosaminoglycans [43]. This would not need special receptors and allow Loh to incorporate into any matrix. The cell specific expression of loh would then mainly decide which matrix will incorporate Prc and this would in turn strongly depend on the cis-regulation of the gene's expression.
On the molecular level we propose that Loh basically acts as a linker protein. Based on the ectopic expression of Loh and the co-immunoprecipitation experiments we can demonstrate that Loh and Prc interact in vivo. In our hands Loh behaves like a secreted receptor molecule that specifically recruits Prc to the cell surface. Our findings indicate that the main molecular function might therefore be binding, but does not exclude additional functions of the protein. It was suggested previously that ADAMTSL proteins act as regulators of extracellular proteases and thereby regulate ECM content and composition [6]. For example it was demonstrated that Drosophila Papilin, another member of ADAMTSL related proteins, is sufficient to inhibit a vertebrate procollagen proteinase in vitro [44]. Thus, it is possible that also Loh regulates a so far unknown proteinase that renders the matrix unsuitable for the accumulation of Prc in some way. In such a model the activity of Loh would then influence the pre-existing microenvironment around a cell to allow Prc to assemble into a network. However, there is no evidence for such a function or the involvement of proteinases so far.
The observed roles of Loh in Drosophila partially reflect the function of ADAMTSL proteins in vertebrates, which were shown to organize Fibrillin-1 (FBN1) microfibrils in specialized matrices. Genetic and biochemical analyses showed that ADAMTSL4 and ADAMSTL6 are sufficient to mediate the formation of FBN1 fibrils in cultured fibroblasts as well as in vivo [28], [45]. ADAMTSL4 acts as a FBN1 binding protein that mediates microfibril assembly in the zonule fibers of the human eye leading to isolated ectopia lentis (IEL) if mutated. Thus, IEL is caused predominantly by altered mechanical properties of the zonular fibers leading to a progressive dislocation of the lens [45]. In Drosophila, where no FBN1 homolog exists, Loh interacts with Prc and mediates its distribution within the ECM in a very similar manner. Therefore, the correct assembly of Prc between the pericardial cells and the heart tube could promote the mechanical properties needed to sustain the permanent mechanical forces during heartbeat. The clinical phenotypes of geleophysic dysplasia (GD) observed in ADAMTSL2 mutant patients exceed a function of simply promoting mechanical stability of the ECM. It was shown that ADAMTSL2 binds to FBN1 but also interacts with LTBP1, a regulator of TGFβ signaling, and therefore the phenotypes of GD also include growing defects, muscular hypertrophy and thickening of the skin [9]. None of these additional phenotypes were observed in Drosophila loh mutants. Therefore, it is obvious that ADAMTSL proteins developed novel functions during evolution making them essential mediators of ECM development and homeostasis. So far there are no reports of interactions between any ADAMTSL proteins with collagens but the obviously similar functions in flies and vertebrates strongly argue for a conserved function in organizing fibrillar matrix proteins.
Flies were kept under standard conditions at 25°C on cornmeal agar. The following fly stocks were obtained from the Bloomington stock center: w1118; Mi(ET1)prcMB03017/TM6c,Sb1, w1118; Mi(ET1)lohMB05750, y1,w1118; Mi(MIC)lohMI02765/SM6a, Df(2L)Exel7048/CyO, Df(3L)vin6/TM3, Sb1,Ser1, w1118; Sco/SM6a,P{hsILMiT}2.4, w1118; UAS-eGFP and balancer stocks KrIf-1/CyO,Kr>GFP and Dr1/TM3,Kr>GFP.
Further fly stocks used are: handC-GFP and handC-Gal4 [25], oddrk111 (odd-lacZ) (C. Rauskolb), vkg::GFP-454 [36], UAS prc-IR41320, UAS prc-IR100357, UAS loh-IR31020 and UAS Sar1-IR34191 [46], UAS loh-IR6232-2 (Drosophila Genetic Resource Center, Kyoto), mef2-Gal4 (H. Nguyen) and prc-Gal4 [20].
Precise excision of minos elements was carried out essentially as described before [26]. Briefly, homozygous w1118; Mi(ET1)lohMB05750 or w1118; Mi(ET1)prcMB03017 males were mated to w1118; Sco/SM6a, P{hsILMiT}2.4 “jump starter” females. After two days adults were removed and the F1 progeny was heat shocked each day at 37°C for 1 h until hatching. F1 males, carrying the minos element (expressing GFP) and the transposase source (recognized by the SM6a balancer) were mated to adequate balancer stocks. In the F2 generation revertant chromosomes were identified by the absence of GFP expression and isolated via backcrossing to the F1 balancer stocks. Revertant lines were established and removal of the minos elements was evaluated by amplifying closely flanking sequences of the transposon by PCR and sequencing. Oligonucleotides (minos-flank) used for PCR and sequencing are: loh-fwd GCGGTCAGCTAAATAGCATC, loh-rev GAATTGGTTTGTCCCACAACG, prc-fwd CACACAGTGGAGCGAGATCC and prc-rev CCTTTCGAAGTGTAAAGTGC.
Embryos were prepared for staining by chemical or heat fixation as described previously [47], [48]. Staining of larvae was done on dissected tissue samples, fixed 1 h in 3.7% formaldehyde in 1× PBS. Primary antibodies used are: guinea pig anti-Loh (1∶500, heat fixation, this study), mouse anti-Prc/EC11 (1∶5, Developmental Studies Hybridoma Bank, DSHB), mouse anti-βPS integrin/CF.6G11 (1∶3, DSHB), mouse anti-FasIII/7G10 (1∶3, DSHB), mouse anti-αSpectrin/3A9 (1∶3, DSHB), mouse anti-Slit/C555.6D (1∶3, heat fixation, DSHB), rabbit anti-Perlecan/Trol (1∶1.000) [49], rabbit anti-Nidogen/Entactin (1∶1.000, a gift from S. Baumgartner), rabbit anti-Laminin (detects only secreted Laminin trimers; a gift from J. Fessler), rabbit anit-Tinman (1∶800) [34] and rabbit anti-GFP (1∶1.000, Abcam). Secondary antibodies used are anti-mouse-Cy2/Cy3 (1∶100/1∶200, Dianova), anti-rabbit-Cy2/Cy3 (1∶100/1∶200, Dianova) and anti-guinea pig-Cy2/Cy3/Alexa633 (1∶100/1∶200/1∶200, Dianova and Abcam). F-Actin was visualized by staining fixed tissues using TRITC coupled phalloidin (Sigma), at a concentration of 0.4 µg/ml in 1× PBS, for 1 h at room temperature. All images were acquired using a Zeiss LSM 5 PASCAL confocal microscope and standard objectives.
The ability of insect nephrocytes to sequester colloids from solutions can be used to specifically label living cells. Therefore colloidal toluidine blue was used as vital stain. Third instar larvae were dissected in 1× PBS and incubated in 0,1 mg/ml colloidal toluidine blue solution for 1 min. Living nephrocytes specifically take up the dye resulting in a deep blue staining. Unspecific signals were removed by three consecutive washes in 1× PBS and animals were photographed immediately.
Embryonic protein extracts were isolated from 20 selected embryos, which were homogenized in 25 µl ECM extraction buffer (1 mM EDTA, 1,5% Triton-X 100 and 2 M urea). Samples were supplemented with 25 µl 2× SDS sample buffer, cooked at 99°C for 2 min and 20 µl were used for SDS-PAGE. Larval and adult extracts were obtained from 10 whole animals homogenized in extraction buffer. Primary antibodies were diluted in 10% dry milk powder (w/v) in TBS-T and incubated overnight at 4°C. Antibodies used were guinea pig anti-Loh (1∶5.000, this study), mouse anti-Prc/EC11 (1∶200, DSHB) and mouse anti-βTub/E7 (1∶5.000, DSHB). Alkaline phosphatase coupled secondary antibodies (Dianova, Germany) were diluted 1∶10.000 and phosphatase activity was visualized by colorimetric NBT/BCIP reaction. Total protein was stained using 0.1 µg/ml amido black 10B (Sigma) in 7% acetic acid.
Animals were equilibrated for 20 min and heart beat was recorded on a Zeiss Axioplan upright microscope equipped with a 10× air objective (n.a. = 0.30). Single pictures were recorded at 80 frames per second (fps) using a Hamamatsu EM-CCD C9100 camera. Images were processed using Fiji and transformed into movie files.
For dye injections staged pharate adults (<90 h APF) were glued on a glass object slide using double sided scotch tape. After 10 min the operculum was removed with fine forceps to allow imaging of dye accumulation. One single injection per animal was carried out, using a glass capillary applied to a micro manipulator and an Eppendorf FemtoJet microinjector. The capillary was filled with 10 µl uranin solution (1 µg/µl in PBS) that was injected laterally into the abdomen of the animal. Dye accumulation was recorded over three minutes using a stereo microscope equipped with an UV lamp, a corresponding filter set and a consumer digital camera (Canon PowerShot A650 IS). Pixel intensities were measured using the “Plot Z-axis profile” tool of Fiji within a region of interest (R.O.I) of the head (excluding the eyes due to different pigmentation).
Freshly hatched animals were collected and separated according their sex and genotype. The flies were kept in plastic vials filled with standard cornmeal agar in groups of less than 20 animals at 22°C. The number of living animals was evaluated every three to five days and the flies were transferred onto new vials.
Late stage embryos were selected according their genotype, judged by balancer expression. Fixation of embryos, sectioning and image acquisition was described previously [47]. The thickness of the basement membrane (BM) was investigated in sections of three independent animals (two sections per animal) of each genotype using Fiji. Therefore, BM thickness was measured at ten randomly picked positions in each image leading to a total number of 60 values per genotype.
Northern blot was done as described previously with 15 µg total RNA loaded per lane [50]. Hybridization was carried out at 66°C for 24 h.
The cDNA of lohA was amplified from cDNA clone GM15606 (BDGP). Oligonucleotides used were lohA-EcoRI-F TACTCAGAATTCATGGCGAAGCTGTTGTTAATATTCAG and lohA-KpnI-R TACTCAGGTACCTTAAATGCCACCCGTGCAGGAAAAAC. The lohAΔSP coding DNA was amplified using the modified oligonucleotide lohAΔSP-EcoRI-F TACTCAGAATTCATG GATTTAACAACTAAAGAGCG. The resulting DNA fragments were cloned into the pUAST vector and transgenic flies were established after standard protocols (TheBestGene Inc., USA).
An antiserum against Loh was generated by injecting two guinea pigs with the sequence specific peptide VFDYHRIDGAEDSNGVTEW-C bound to KLH. Harvested antiserum was affinity purified against the peptide. Peptide synthesis, serum production and affinity purification were carried out by a commercial service (Pineda Antikörperservice, Berlin).
All steps were carried out at 4°C or on ice. Total protein from 100 mg adult prc-Gal4/+; UAS-LohA/+ flies (∼100 flies) was extracted in 500 µl ECM extraction buffer (1 mM EDTA, 1,5% Triton-X 100 and 2 M urea). Flies were homogenized, pulled 6-times through a syringe (Ø = 0,8 mm) and debris was spun down at 8.000 g for 30 min. The supernatant was centrifuged again at 13.100 g for 30 min. The soluble protein fraction was split into four 100 µl aliquots. One aliquot served as input. The other aliquots were supplemented with 10 µl Protein A-Sepahrose 4B (Sigma), 0,1% BSA and either 10 µl PBS (negative control), 10 µl anti-Loh or 67 µl anti-Prc antibody and incubated under constant shaking overnight. Protein A slurry was spun down at 13.100 g for 10 min and the pellet was washed in 500 µl ice cold 1M NaCl. The washing step was repeated three times, afterwards the pellets were resolved in 60 µl 2× SDS sample buffer and used for Western blotting
|
10.1371/journal.pgen.1005178 | The Protein Quality Control Machinery Regulates Its Misassembled Proteasome Subunits | Cellular toxicity introduced by protein misfolding threatens cell fitness and viability. Failure to eliminate these polypeptides is associated with various aggregation diseases. In eukaryotes, the ubiquitin proteasome system (UPS) plays a vital role in protein quality control (PQC), by selectively targeting misfolded proteins for degradation. While the assembly of the proteasome can be naturally impaired by many factors, the regulatory pathways that mediate the sorting and elimination of misassembled proteasomal subunits are poorly understood. Here, we reveal how the dysfunctional proteasome is controlled by the PQC machinery. We found that among the multilayered quality control mechanisms, UPS mediated degradation of its own misassembled subunits is the favored pathway. We also demonstrated that the Hsp42 chaperone mediates an alternative pathway, the accumulation of these subunits in cytoprotective compartments. Thus, we show that proteasome homeostasis is controlled through probing the level of proteasome assembly, and the interplay between UPS mediated degradation or their sorting into distinct cellular compartments.
| The accumulation of misfolded proteins threatens cell fitness and viability and their aggregation is commonly associated with numerous neurodegenerative disorders. Cells therefore rely on a number of protein quality control (PQC) pathways to prevent protein aggregation. In eukaryotes, the ubiquitin proteasome system (UPS), a supramolecular machinery that mediates the proteolysis of damaged or misfolded proteins, plays a vital role in PQC by selectively targeting proteins for degradation. Although the critical role-played by the UPS in PQC, and the severe consequences of impairing this pathway are well established, little was known about the mechanisms that control dysfunctional proteasome subunits. Here, we reveal that the interplay between UPS mediated degradation of its own misassembled subunits, and sorting them into cytoprotective compartments, a process that is mediated by the Hsp42 chaperone, determines how proteasome homeostasis is controlled in yeast cells. We believe that the mechanism of proteasome regulation by the PCQ in yeast may serve as a paradigm for understanding how homeostasis of this essential complex is controlled in major chronic neurodegenerative disorders in higher eukaryotes.
| Protein homeostasis encompasses the systems required by the cell for generating and maintaining the correct levels, conformational states, distribution, and degradation of its proteome. Maintaining protein homeostasis is crucial to cells given the toxic potential of misfolded proteins and aggregates. Cells therefore rely on a number of protein quality control (PQC) pathways that survey proteins both during and after synthesis to prevent protein aggregation, promote correct protein folding, and target terminally misfolded proteins to degradation. In eukaryotes, the ubiquitin proteasome system (UPS) plays a vital role in PQC by selectively targeting proteins for degradation [1–4].
The eukaryotic 26S proteasome is a highly conserved 2.5-MD multisubunit protease responsible for degrading a large fraction of intracellular proteins. The 26S proteasome comprises a 20S core particle (CP) and two 19S regulatory particles (RP) that are further divided into lid and base complexes [5]. The degradation of most proteins is mediated by polyubiquitin chains labeling, which leads to their recognition by the 26S proteasome [6]. A diverse array of fundamental biological processes are controlled by the UPS, including cell cycle progression, DNA repair, signal transduction, and PQC in which the cell removes abnormal and toxic proteins generated as a result of environmental damage [7,8]. Under such conditions, chaperones are tasked with accompanying terminally misfolded and aggregated proteins to disposal, or to limit inclusion of these proteins, thereby preventing protein aggregates from causing cell toxicity and from being transferred to the next generation [2]. This chaperone mechanism, alongside the UPS, is termed spatial quality control, and consists of the juxtanuclear quality control compartment (JUNQ) and the insoluble protein deposit (IPOD), which were identified in yeast [9,10]. The JUNQ provides a specialized environment for chaperone-mediated refolding or proteasomal protein degradation. Proteins that are not refolded or degraded in the JUNQ are mobilized to the IPOD.
Alongside its PQC role, the UPS plays an essential role in regulating the degradation and function of nuclear proteins [11–13]. Accordingly, immune-electron microscopy experiments [14] and fluorescent microscopy of GFP-tagged proteasome subunits [15] have established that the 26S proteasome is highly enriched in the nucleus. In exponentially growing yeast cells, 80% of the 26S proteasomes are localized inside the nucleus throughout the cell cycle [16]. Given the importance of the UPS, proteasomal nuclear mislocalization may have severe consequences, for example, a deleterious effect on DNA double strand break repair [13].
Although ubiquitin-mediated proteasomal degradation of many proteins plays a key role in the PQC system, the proteasome itself can become dysfunctional as a result of transcriptional and translational failures, genomic mutations, or diverse stress conditions, leading to misfolded proteins existing in every compartment of the cell. The regulatory pathways, and the identity of the cellular machinery that mediates the sorting sequestration and elimination of dysfunctional proteasomal subunits remain poorly understood.
By using a mutated proteasome lid subunit (termed rpn5ΔC), we show that the nuclear mislocalization, and the cytosolic aggregates formed by this mutant represent a misassembled proteasome lid. With this experimental tool in hand, we were able to demonstrate how the dysfunctional proteasome is controlled by the PQC machinery. We found that among the multilayered quality control mechanisms, the UPS-mediated degradation of its own dysfunctional subunits is the favored pathway. However, in the absence of a functional proteasome, peripheral aggregates that represent misassembled proteasome accumulate in the IPOD, a process that is mediated by the Hsp42 chaperone. We further demonstrate that while the proteasome structure can tolerate the structural defects of the rpn5ΔC mutant and assemble into a functional proteasome, the PQC machinery takes-over, and the mutated protein is spatially removed by the PQC machinery, leading to proteasome dysfunction. Overall, our results demonstrate that proteasome homeostasis is controlled through cellular probing of the quality of proteasome aggregates, and the interplay between UPS-mediated degradation of dysfunctional subunits and alternatively, their accumulation in cytoprotective compartments.
Recently, we screened a collection of temperature sensitive (Ts) mutants in the yeast Saccharomyces cerevisiae for those that show a chromosomal instability (CIN) phenotype [17]. This screen identified proteasome subunit genes, such as the Ts allele of the regulatory particle (RP) subunit RPN5, and the core particle (CP) subunit PUP2. The Ts allele of RPN5 was generated by random mutagenesis [17]. Sequence analysis of this mutant revealed that a single base pair insertion resulted in a premature stop codon, leading to a 34 amino acid (aa) truncation at the C-terminal domain (CTD) of Rpn5 (termed rpn5ΔC). To examine the subcellular localization of Rpn5ΔC, GFP was fused at its N-terminus. An identical N-terminal GFP fusion was constructed for the wt RPN5 gene, which was used as a control (both fusion proteins were expressed from a galactose-inducible promoter (GAL1)). RPN5 is an essential gene, and therefore, growth on a glucose-containing medium (which shuts-off the expression of both GAL1-GFP-RPN5 (GFP-RPN5), and GAL1-GFP-rpn5ΔC (GFP-rpn5ΔC)) resulted in cell death (Fig 1Ai-top). In contrast, when the expression of the wt and the truncated gene were induced by growing the cells on 2% galactose, the growth was fully restored at the semi-permissive temperature (30°C) (Fig 1Ai-bottom). Furthermore, as shown in Fig 1Aii-top, similarly to the wt proteasome, the control GFP-Rpn5 protein localized predominantly to the nucleus in logarithmically growing cells. These results indicate that GAL1-GFP-RPN5 fully supports the proteasome function. In contrast, while GFP-Rpn5ΔC cells could still grow at the semi-permissive temperature (30°C), the GFP-Rpn5ΔC protein were detected in cytosolic inclusions in 81% of the cells (n>200) (Fig 1Aii, bottom, other localization patterns are shown in S2A Fig).
Next, we wanted to determine whether the mislocalization caused by rpn5ΔC could be attributed to a failure in proteasome assembly. Previous studies have suggested a model whereby the two parts of the 26S proteasome, namely the CP and RP, are formed and imported to the nucleus independently of each other [18,19]. Our results are in agreement of this model and show that rpn5ΔC leads to the specific mislocalization of another proteasomal lid subunit (Rpn11) (S1Ai and S1Aii Fig), while the core subunits are retained in the nucleus (S1Aiii and S1Aiv Fig). Similar results were obtained in a reciprocal experiment in which the Ts mutant of the CP pup2 affects the nuclear enrichment of Pre6-GFP (another CP), while the RP Rpn11-GFP is unaffected (S1B Fig). To further demonstrate the importance of an intact CTD for proteasome integrity, we generated diploid cells in which the original truncation of 34 aa from the CTD was extended to 45 aa (rpn5Δ45). In this case, no viable haploid rpn5Δ45 spores could be obtained following tetrad dissection (Fig 1B). Moreover, a cross-species complementation experiment revealed that the expression in yeast of a full-length human homolog of RPN5, PSMD12, but not psmd12ΔC, was able to rescue the temperature sensitivity of the rpn5ΔC strain (Fig 1C).
Next, we wished to test the interaction of truncated RPN5 with other proteasomal lid subunits. To this end, we used the protein fragment complementation assay (PCA) [20] to examine the interaction between Rpn5ΔC and several other proteasomal lid subunits (Rpn3, Rpn6, Rpn7, Rpn8, Rpn11, and Rpn12) that were previously shown to interact with Rpn5 [7,21,22]. In this assay (for details see S1C Fig) the interaction between two proteins of interest can be detected by cell growth on media in the presence of the dihydrofolate reductase (DHFR) enzyme inhibitor, methotrexate (MTX). The results show that Rpn5ΔC fails to exhibit the expected interactions when compared to the wt Rpn5 control at the semi-pemissive temperature (Fig 1D).
This result was supported biochemically by an immunoprecipitation experiment showing that when Rpn8 fused to a FLAG-Tag (Rpn8-Flag) is pulled-down, it shows a decreased interaction with Rpn5 containing a 34 aa deletion in its CTD (Rpn5ΔC-F[3]) (S1D Fig).
By generating a series of strains with defined deletions at the C-terminus of Rpn5 (Fig 1D and 1E), we next showed that the interaction with other proteasomal lid subunits is impaired only when the deletion at the CTD domain is greater than 20aa. The physiological importance of these interactions is highlighted by the clear correlation between the extent of truncation and cell viability at the restrictive temperature (34°C) (Fig 1F).
Moreover, we analyzed total protein by immunoblotting with anti Ub Abs. The results (Fig 1G) show a clear increase in protein ubiquitination in GFP-Rpn5ΔC cells, when compared to the GFP-Rpn5 control, at the restrictive temperature of 34°C. We therefore concluded that the presence of the truncated form of RPN5 is associated with proteasome dysfunction.
Our results are in agreement with a previous study showing that in rpn5-1 cells, containing a different CTD-truncated Ts mutant of RPN5, lid subcomplexes are not assembled, even to a partial extent, at the restrictive temperature [18]. This study also specifically examined the effect of the rpn5-1 mutation on the UPS, by evaluating the stability of three model substrates of the ubiquitin–proteasome pathway. The results demonstrated that compared with the wild-type cells, rpn5-1 cells maintained the normally short-lived substrates at a higher level at the semi-permissive temperature, indicating that the rpn5-1 mutation caused a defect in the UPS. [18]. Additional studies mapped the interaction between the subunits of the RP by electron microscopy, tandem mass-spectrometry, affinity purification analysis, and other methods [7,21,23]. These approaches demonstrate that Rpn5, Rpn8, Rpn9, and Rpn11 form a stable soluble subcomplex, and the authors have proposed a subunit interaction map, supporting the notion that Rpn5 is a core component in the lid formation. Furthermore, it was also shown that in yeast, Rpn5 is independently incorporated through its CTD into the proteasomal lid [24].
These results, together with our observation that the RP subunit Rpn11-RFP co-localizes with 98% of the cells containing a large GFP-Rpn5ΔC cytosolic aggregate (n>100) (Fig 2A), suggest that these aggregates represent misassembled proteasome lid intermediates.
Next, we tested the degree of lid assembly into 26S proteasome holocomplexes in GFP-Rpn5, or GFP-Rpn5ΔC at the semi-permissive and restrictive temperatures using the in-gel peptidase activity assay. In this assay proteasomes are resolved by non-denaturing PAGE according to their charge/mass ratio directly from whole cell extract, and visualized based on inherent peptidase activity as described in [25]. The results (Fig 2B) clearly show that similarly to cells expressing the endogenous levels of Rpn5 in the wt, the over-expression of GFP-Rpn5 by a galactose inducible promoter, had no effect on proteasome integrity, as in both cases similar amounts of proteasomes were found as a mixture of RP2CP, and RP1CP. In contrast, the over production of GFP-Rpn5ΔC resulted in structural defects, as evident by higher levels of RP1CP at the semi permissive temperature (30°C), and free BaseCP and CP mostly at the restrictive temperature (34°C). Similar defects were previously reported by Yu et al, when using rpn5ΔC mutant, expressed through its endogenous promoter [24]. Taken together, these results clearly show that proteasome assembly is inhibited in rpn5ΔC cells, and that the misassembled lid in rpn5ΔC is not associated with active proteasome.
Aggregation prone proteins are partitioned between the JUNQ and the IPOD [9,10,26]. Proteins that are ubiquitinated by the PQC machinery are delivered to the JUNQ where they are processed for degradation by the proteasome [9]. We therefore hypothesized that the impairment of the PQC degradation pathway by the rpn5ΔC mutant, should lead to the accumulation of misassembled proteasomal lid subunits in the IPOD. To test this idea, we followed the localization of Hsp104, a commonly used IPOD marker [9,10,27], fused to TFP (Hsp104-TFP), in a GFP-Rpn5ΔC strain grown in rich galactose containing medium. Our results show that in all cases where GFP-Rpn5ΔC cytosolic inclusions could be detected (81% of the cells, n>200, S2A Fig), the largest inclusion always co-localized with Hsp104-TFP (n>200) (Fig 2C). Similar results were obtained with the glutamine-rich prion protein Rnq1 fused to mCherry (Rnq1-mCherry), another well-established IPOD marker (Fig 2D) [9,10,27]. Taken together, these findings suggest that peripheral foci containing misassembled Rpn5ΔC are directed to the IPOD when a functional proteasome is scarce, such as in rpn5ΔC cells.
It should be noted that since the RPN5 proteasome subunit is essential for cell viability [17,28], our experiments were performed mainly at the semi-permissive temperature (30°C), to enable its partial (hypomorphic) function. At this temperature, some proteasomes apparently function sufficiently to support cell growth. Indeed, in 19% of the cells (n>200), the GFP-Rpn5ΔC signal was limited to the nucleus, the expected localization in proliferating cells (S2A-top Fig). Since the sequestration of aggregates to the IPOD was shown to preclude their delivery by the parental cells to subsequent generations [10], the nuclear GFP-Rpn5ΔC signal, may also represent the functional proteasomes in recently separated daughter cells, as demonstrated in S1 Movie. This idea is also supported by calcofluor staining showing that the GFP-Rpn5ΔC signal was not detected as cytosolic inclusions in 98% of the daughter cells issued from IPOD containing mother cells (Fig 2E).
In addition to the nuclear localization, 81% of cells showed cytosolic inclusions of the GFP-Rpn5ΔC signal with the following patterns (representative images are shown in S2A Fig, n>200): 25% exhibited a single large cytosolic aggregate representing the IPOD, and 39% contained a second smaller juxtanuclear protesomal signal, likely representing the JUNQ [9]. Finally, it was recently demonstrated that under acute stress (such as in the case of partially assembled proteasome), misfolded proteins are initially collected and processed in the form of multiple puncta named Q-bodies, which are rapidly reversible structures, that can be dynamically directed to folding and refolding by the chaperone machinery, or to degradation by the proteasome or autophagy system [10,26]. Indeed in 17% of the cells, we detected many cytosolic aggregated bodies, which probably represent these structures.
One of the challenges faced by the cell in maintaining protein homeostasis is the presence of misfolded proteins. The UPS, in particular, plays a critical role in PQC by selectively targeting proteins for degradation. To test whether the UPS can also regulate the degradation of its own misassembled subunits, we introduced a functional allele of RPN5 by mating the haploid GFP-rpn5ΔC strain to another haploid containing a wt copy of RPN5. The complementation of GFP-rpn5ΔC by RPN5 was indicated by the restoration of growth at the restrictive temperature, and the nuclear localization of the Rpn5 wt protein (Rpn5-RFP) (Fig 3A and 3B). Similarly Rpn11-GFP also localized to the nucleus in rpn5ΔC/RPN5 cells (S2B Fig when compared to rpn5ΔC haploids (Fig 2A). Interestingly, in the presence of a functional copy of RPN5, the GFP-Rpn5ΔC puncta was no longer detected in 97% of the cells (n>100) (Fig 3C). Furthermore, the addition of the proteasome inhibitor, MG132, stabilized the GFP-Rpn5ΔC signal, and led to its reproducible accumulation in the IPOD in 68% of the cells (n>100), compared to the DMSO control (Fig 3D). The accumulation of GFP-Rpn5ΔC in cells treated with MG132 was further confirmed by GAL promoter shutoff experiments followed by Western Blot analysis (S2C Fig).
Previous studies suggested that the ubiquitination level of a protein determines whether it is sequestered into the IPOD or JUNQ compartments [9,29]. Impairing misfolded protein ubiquitination blocked their accumulation in the JUNQ, and instead resulted in excessive accumulation in the IPOD [9]. The UPS-mediated degradation of Rpn5ΔC in heterozygote diploids (containing a wt copy of RPN5), suggests that Rpn5ΔC is in a ubiquitinated form, when localized to the IPOD. In order to explain this discrepancy, we propose that in rpn5ΔC haploids, the cells are under constant acute stress, due to proteasome impairment. Under such conditions, the degradation mechanism is blocked, and therefore there is no alternative, aside from targeting the misassembled lid to the IPOD. This idea is supported by previous reports that proteasome inhibition leads to the accumulation of substrates that are normally degraded by the UPS (such as Ubc9ts) into the JUNQ and IPOD [9,26]. In conjunction with this, other studies have demonstrated that when the degradation capacity of the JUNQ declines, with JUNQ-localized proteasomes becoming inactive, a protective alternative is furnished by the IPOD, and toxic aggregating species are rerouted from the JUNQ to the IPOD [30]. We therefore suggest that in our case, when cells fail to degrade misassembled proteasome lid subunits, they can be targeted to the IPOD when still conjugated to ubiquitin.
Taken together, our results suggest that the preferred PQC pathway of misassembled proteasome subunits is degradation by the UPS. As functional proteasomes are scarce in rpn5ΔC cells grown at the restrictive temperature, the UPS pathway is hindered. Thus, misassembled lid subunits are ultimately diverted to the IPOD.
Molecular chaperones prevent aggregation and misfolding of proteins, and are thus central to maintaining protein homeostasis. Two chaperones that were previously shown to mediate spatial sequestration of misfolded proteins are the small heat shock proteins Hsp26 and Hsp42. These chaperones efficiently co-aggregate with misfolded proteins, thereby altering the properties of protein aggregates and facilitating disaggregation by other chaperones [31]. This process is a key molecular event that determines whether such a protein is sorted to the JUNQ or to a peripheral site [32,33]. As shown in Fig 4A and 4B, both Hsp26 and Hsp42, fused to TFP, colocalized with GFP-Rpn5ΔC in 100% of the cells containing a large cytosolic aggregate, representing the IPOD (n>200). Similar results were obtained when we investigated the co-localization of Hsp42-TFP with Rpn11-GFP on a rpn5ΔC background expressed from its endogenous promoter (Fig 4C). Since we have shown that Rpn5ΔC co-localizes with Rpn11 (Fig 2A), and that Hsp42 co-localizes with the IPOD marker Hsp104 on a rpn5Δc background in 95% of the cells (n>200) (Fig 4D), we conclude that Hsp42 colocalizes with the misassembled proteasome lid in the IPOD. This co-localization is probably associated with physical interactions between Hsp42 and misassembled proteasome subunits, as indicated by the co-immunoprecipitation between the lid subunit Rpn8, and Hsp42 in rpn5ΔC cells (S2D Fig).
Given the role of HSP42, HSP26 and HSP104 in controlling the sequestration of protein aggregates into deposition sites [32–34], we hypothesized that the sequestration of GFP-Rpn5ΔC to the IPOD may also depend on these chaperones. To investigate this possibility, we examined the localization of GFP-Rpn5ΔC in Δhsp42, Δhsp26 and Δhsp104 cells. In contrast to Δhsp26, and Δhsp104 which only had a minor effect on GFP-Rpn5ΔC cytosolic peripheral focus (S2E and S2F Fig), in cells lacking HSP42, the GFP-Rpn5ΔC signal was no longer observed in the cytosolic periphery. Instead, it showed the nuclear enrichment expected of the wt proteasome (Fig 4E-bottom). Similar results were obtained with Rpn11-GFP in a strain mutated in both HSP42 and RPN5 (Δhsp42, rpn5Δc) (Fig 4F-bottom). We therefore concluded that association of misassembled proteasome lid subunits with Hsp42 is required for their accumulation in the IPOD. Strikingly, while the deletion of HSP26, and HSP104 had no effect, the nuclear relocalization of GFP-Rpn5ΔC in Δhsp42 cells was clearly associated with increased survival at 34°C (Fig 4G), and as revealed by the in gel peptidase activity assay, with the reassembly of functional proteasomes (Fig 4H). Taken together, these results show that HSP42 plays an essential role in mediating the sequestration of misassembled proteasome lid subunits to the IPOD.
Our data is in agreement with previous studies that mapped the interactions between subunits of the proteasome regulatory particles, and led to the notion that Rpn5 is a core component in lid formation [7,21,23], and that in yeast, Rpn5 is independently incorporated through its CTD into the proteasomal lid [24]. Our model (Fig 5) suggests that there is competition between assembly, degradation and aggregation of proteasome subunits. The employment of rpn5ΔC shifts the balance, since this mutation partially impairs proteasome lid assembly, which in contrast to wt RPN5, triggers the activation of the PQC machinery. At the restrictive temperature, the lid is mostly misassembled, and the degradation pathway is blocked, which results in the independent recruitment of Rpn5ΔC, and other lid subunits to the IPOD in an HSP42 dependent manner. This slower assembly has a dual effect: It increases the amount of the unassembled subunits, and at the same time decreases its degradation, because there is less proteasomes available. Hence, the misassembled subunits aggregate in the IPOD.
The fact that the deletion of HSP42 restores cell growth at 34°C, suggests that at this temperature, cells can still tolerate the CTD truncation in Rpn5, and assemble partially functional proteasomes (Fig 4G and 4H). A previous study supports this idea by proteasome fractionation using a glycerol gradient, showing that although the truncation influenced integration of additional subunits, Rpn5ΔC could still integrate into the proteasome at the semi-permissive temperature [8]. However, the cell stress imposed by partial misassembly of the lid in the rpn5 mutant activates the PQC machinery. This activation causes further damage, since lid subunits are independently removed to the IPOD by Hsp42, which in turn leads to complete lid misassembly, proteasome dysfunction, and cell death at the restrictive temperature (34°C). In agreement with this model, the omission of HSP42 probably prevents the rapid sequestration of Rpn5ΔC into aggregates, allowing more time for 26S proteasomes to assemble, and to degrade the unassembled lid subunits.
Thus, while it was believed that the Rpn5ΔC mutation causes purely structural defects [18], our study provides a plausible alternative mechanism. We suggest that spatial separation of misassembled proteasome lid subunits mediated by the PQC machinery is the key pathway leading to proteasome dysfunction, rather than the structural defects within the RPN5 mutant.
Taken together, our results reveal that proteins harboring mutations that activate the PQC can be eliminated from the cells, even when the protein is still functional, and the damage ensuing from diverting essential protein products. In light of this, cells have adopted numerous PQC pathways to aid folding, mediate degradation, or to accumulate such proteins in stress foci [3]. This idea is nicely demonstrated by the ΔF508 mutation within the fibrosis transmembrane conductance regulator (CFTR), highly associated with cystic fibrosis (CF) [35]. Although the mutated protein retains significant chloride-channel function, the protein is rapidly recognized by the PQC machinery and is degraded shortly after synthesis, before it can reach its site of activity at the cell surface [36].
Although the critical role-played by the UPS in PQC, and the severe consequences of impairing this pathway are well established, little was known about the mechanisms that control dysfunctional proteasome subunits. Our results demonstrate for the first time that proteasome homeostasis is controlled through the interplay between UPS mediated degradation of its misassembled subunits, and sorting into the IPOD, a process that is mediated by the Hsp42 chaperone, which determines how proteasome homeostasis is controlled in yeast cells.
The assembly of the proteasome is an intricate process due to the number of subunits that must associate to form an active complex. We used a synthetic mutant that induces proteasome dysfunction. However, the UPS function can be naturally impaired by many factors, including mutations, errors during transcription, RNA processing and translation, trapping of a folding intermediate, incorrect incorporation into multimeric complexes or oxidative damage, all of which are processes that are accelerated during aging [37]. Dysregulation of this pathway results in intracellular deposits of ubiquitin protein conjugates which can be seen in age-related pathologies and in all the major chronic neurodegenerative disorders such as Alzheimer’s, Parkinson’s and Huntington’s diseases as well as amyotrophic lateral sclerosis (ALS) and others [37] [38].
The mechanism of proteasome regulation by the PQC in yeast may serve as a paradigm to understand how homeostasis of this essential complex is controlled in higher eukaryotes. Identifying additional chaperones that work in conjunction with Hsp42, and elucidating the identity of the structurally abnormal features that the PQC machinery recognizes in the misassembled proteasome, will provide further insight into the recognition and targeting mechanisms of dysfunctional proteasomes in cells.
Unless otherwise stated, all the strains used in this study are isogenic to BY4741, BY4742, or BY4743 [39]. The relevant genotypes are presented in S1 Table. Deletions, GFP, TFP, and mCherry fusions were generated using one step PCR mediated homologous recombination as previously described [40,41]. For all deletions, the selection markers replaced the coding region of the targeted genes. GFP, TFP and mCherry were fused at the 3’ end of the coding region of the targeted genes, by replacement of their stop codons [40,41]. A GAL1 promoter was placed at the N-terminal of RPN5 and RPN5ΔC by replacement of their start codon. We have shown that the expression of the fusion protein GAL1-GFP-RPN5 has no effect on proteasome normal phenotype (Fig 1A). Strains containing different mDHFR-F[1,2], and mDHFR-F[3] C-terminal fusion proteins were obtained from the PCA collection (commercially available from Open Biosystems), or in cases in which the strains were absent from the collection, by one step PCR mediated homologous recombination, as described by Tarassov, K, et al. [20]. For C-terminal truncations by mDHFR-F3 (Fig 2D and 2E), this fragment was targeted to replace the truncated amino acids at the 3’ end.
Microscopy was performed as previously described [42]. Briefly, cells were observed in a fully automated inverted microscope (Zeiss observer. Z1 Carl Zeiss, Inc.) equipped with an MS-2000 stage (Applied Scientific Instrumentation), a Lambda DG-4 LS 300 W xenon light source (Sutter Instrument), a 63x Oil 1.4 NA Plan-Apochromat objective lens, and a six-position filter cube turret with a GFP filter (excitation, BP470/40; emission, BP525/50; Beamsplitter, FT495), a HcRed filter (excitation, BP592/24; emission, BP675/100; Beamsplitter, FT615), and a DAPI filter (excitation, G365; emission, BP445/50; Beamsplitter, FT395) from Chroma Technology Corp. Images were acquired using a CoolSnap HQ2 camera (Roper Scientific). The microscope, camera and shutters (Uniblitz) were controlled by AxioVision Rel. 4.8.2. Images are a single plane of z-stacks performed using a 0.5 μm step.
The PCA was performed as described previously [20]. Strains were mated on YPD, and diploids were selected on YPD supplemented with clonNAT and hygB. SD supplemented with noble agar (Difco), and methotrexate (MTX; Bioshop Canada) was used for the final selection steps. Drugs were added to the following final concentrations: clonNAT (100 μg/ml, Werner Bioagents); MTX (200 μg/ml (prepared from a 10 mg/ml methotrexate in DMSO stock solution, Bioshop Canada); and HygromycinB (100 μg/ml, Calbiochem).
Co-immunoprecipitations and Western blot analysis were carried out as described previously [43]. The antibodies used for the Western blot analysis were anti-DHFR-[F3] (Sigma), anti-FLAG (Sigma), anti-Ubiquitin (Dako Dk-z045801-2), anti-GFP (Roche 11–814460001).
Cultures were grown overnight and washed twice with DDW and once with chilled buffer A (25 mM Tris [pH 7.4], 10 mM MgCl2, 10% glycerol, 1 mM ATP, and 1 mM dithiothreitol [DTT]). Pellet was resuspended in two volumes of buffer A and lysed using glass beads at 4°C. Native lysates were clarified by centrifugation at 16,000×g for 15 min.
Proteasome peptidase activity was studied in native PAGE using the substrate succinyl-LLVY-7-amido-4-methylcoumarinfluorescent peptide (Bachem, Bubendorf, Switzerland) as previously published [24,25].
|
10.1371/journal.ppat.0030048 | Type III Effector Activation via Nucleotide Binding, Phosphorylation, and Host Target Interaction | The Pseudomonas syringae type III effector protein avirulence protein B (AvrB) is delivered into plant cells, where it targets the Arabidopsis RIN4 protein (resistance to Pseudomonas maculicula protein 1 [RPM1]–interacting protein). RIN4 is a regulator of basal host defense responses. Targeting of RIN4 by AvrB is recognized by the host RPM1 nucleotide-binding leucine-rich repeat disease resistance protein, leading to accelerated defense responses, cessation of pathogen growth, and hypersensitive host cell death at the infection site. We determined the structure of AvrB complexed with an AvrB-binding fragment of RIN4 at 2.3 Å resolution. We also determined the structure of AvrB in complex with adenosine diphosphate bound in a binding pocket adjacent to the RIN4 binding domain. AvrB residues important for RIN4 interaction are required for full RPM1 activation. AvrB residues that contact adenosine diphosphate are also required for initiation of RPM1 function. Nucleotide-binding residues of AvrB are also required for its phosphorylation by an unknown Arabidopsis protein(s). We conclude that AvrB is activated inside the host cell by nucleotide binding and subsequent phosphorylation and, independently, interacts with RIN4. Our data suggest that activated AvrB, bound to RIN4, is indirectly recognized by RPM1 to initiate plant immune system function.
| Many bacterial pathogens use a specialized protein “injection needle” called a type III secretion system to help colonize cells of higher organisms. The type III secretion needle attaches to a host cell and is the delivery conduit for a variety of bacterial proteins that act inside of the host cell. These proteins are called type III effectors. They manipulate host cell biology in order to help the bacterial pathogen colonize the host. We studied one type III effector from plant pathogenic bacteria called Pseudomonas syringae. This effector, termed avirulence protein B (AvrB), is targeted to the inner face of the plant cell plasma membrane, where it interacts with a membrane-bound host protein called RIN4 (resistance to Pseudomonas maculicula protein–interacting protein). RIN4 is phosphorylated in the presence of AvrB and an as-yet-unknown additional host factor. We provide a structural basis for the binding of AvrB to RIN4 and a possible mechanism of action for AvrB inside the host. AvrB activation and its ability to bind RIN4 have evolved to help the pathogen, yet in Arabidopsis, the AvrB-dependent phosphorylation of RIN4 is sensed by the plant immune system, leading to a rapid halt in pathogen growth.
| Many Gram-negative bacterial pathogens of plants or animals employ type III secretion systems (TTSSs) to translocate type III effector proteins into host cells [1]. Type III effector proteins manipulate host cellular targets and signaling pathways to promote the infection process in genetically susceptible hosts [2,3]. In the plant immune system, specific nucleotide-binding leucine-rich repeat (NB-LRR) disease resistance proteins can monitor the homeostasis of type III effector targets [4–6]. In several cases, when a type III effector perturbs its target, the corresponding NB-LRR protein is activated. NB-LRR activation leads to a complex output including hypersensitive cell death (HR) and a suite of cellular responses that render the plant resistant to infection by pathogen strains expressing that type III effector.
The Arabidopsis NB-LRR protein RPM1 (resistance to Pseudomonas maculicula protein 1) recognizes the action of two distinct type III effector proteins, avirulence protein Rpm1 (AvrRpm1) and avirulence protein B (AvrB), which are found in various strains of the plant pathogen Pseudomonas syringae [7,8]. The RPM1-interacting protein (RIN4) is required for RPM1 function triggered by either AvrRpm1 or AvrB [9]. RIN4 physically associates in vivo with both AvrB and AvrRpm1, and with RPM1 [9]. The presence of either AvrRpm1 or AvrB in the plant cell leads to phosphorylation of RIN4, although neither of the type III effectors has sequence similarity to kinases [9]. RIN4 also interacts with, and is required for the function of, a second NB-LRR protein, RPS2 (resistance to P. syringae protein 2) [10,11]. The corresponding type III effector, avirulence protein Rpt 2 (AvrRpt2), is an autoprocessed cysteine protease that is activated by a host cyclophillin after delivery [12,13]. AvrRpt2 cleaves RIN4 at two sites [14,15], and the disappearance of RIN4 drives RPS2 activation [10,11].
One of the AvrRpt2 cleavage sites overlaps the AvrB binding site on RIN4 [15]. Thus, at least two independent type III effector proteins have evolved to target a small approximately 30–amino acid domain on RIN4 using at least two different biochemical mechanisms. It remains to be determined whether AvrRpm1 also targets this region of RIN4. This region is shared among several otherwise unrelated Arabidopsis proteins of unknown function and may represent a common motif targeted by plant pathogens [14,15]. RPM1, RPS2, the type III effectors, and RIN4 have been demonstrated, or are predicted to be, localized to the inside of the plant plasma membrane; AvrB, AvrRpm1, and RIN4 are acylated as a requirement for this localization [11,15–17].
Despite the wealth of genetic information for AvrB, its biochemical function remains elusive. A crystal structure of free AvrB revealed that it adopts a novel bilobal fold with no structural homologies to previously characterized proteins or functional domains [18]. A small upper lobe (amino acid residues 123 to 217) contains three alpha-helices (α5, α6, and α7) surrounding a five-stranded antiparallel beta-sheet (β1-β5-β4-β3-β2). A large lower lobe (residues 28 to 122 and 218 to 317) is composed strictly of alpha-helices [18]. The junction of the two lobes forms a large solvent-accessible cleft with a volume of over 900 Å3 that extends into the lower lobe to form a pocket rich in conserved residues. Chimeras of AvrB with the closely related paralog avirulence protein C (AvrC) (which does not activate RPM1) were used to demonstrate that the upper lobe of AvrB (residues 126 to 216) is required for RPM1 activation [18]. The lack of sequence conservation in the upper lobe among AvrB paralogs from various plant pathogens that do not trigger RPM1 mediated responses (see below) further supports this contention. In contrast, the lower lobe and the interlobal cleft are highly conserved in sequence among the AvrB family members. The conserved cleft was hypothesized to be an enzymatic active site that binds to substrate and/or cofactor [18].
In this study, we present the structure of AvrB bound to RIN4 and we identify interacting residues in the upper lobe of AvrB that are required for both RIN4 binding and activation of RPM1. We also identified a highly conserved nucleotide-binding pocket contained largely in the lower lobe of AvrB that is also required for activation of RPM1. In addition, phosphorylation of AvrB occurs in the presence of a host cofactor or kinase and may represent a third prerequisite for recognition by RPM1.
Previous work using gel filtration and native gel electrophoresis demonstrated that AvrB bound a small RIN4 fragment consisting of amino acids 142 to 179 [15]. We cocrystallized AvrB with RIN4142–176 (Figure 1A, 1C, and 1E; Table 1). This fragment binds AvrB with affinity similar to that of the full-length protein (kd = 3 μM versus 4 μM as determined by isothermal titration calorimetry [ITC]; see below). Thus, essentially all of the binding energy for interaction with AvrB is contained in RIN4142–176. The structure of AvrB complexed to RIN4142–176 revealed that the peptide forms a “Z” shape with the N-terminal half contacting the upper lobe and the C-terminal half straddling the interface between the two lobes (Figure 1A, 1C, and 1E; Table 1). The interaction of AvrB with RIN4142–176 does not alter the overall structure of AvrB (unpublished data). Using this cocrystal structure, we identified residues of AvrB that interact with RIN4142–176 and could potentially be important for binding affinity (Figure 2A and 2C). The N-terminal half of RIN4142–176 interacts mainly through hydrophobic burial of RIN4 W154 plus a hydrogen bond between the indole nitrogen of RIN4 W154 and AvrB D213 (Figure 2A). RIN4 W154 is part of an AvrRpt2 cleavage site on RIN4 [14]. Additionally, AvrB T182 contacts RIN4 Y151, and AvrB V128 contacts RIN4 D155, S161, and G162.
The most extensive interactions of RIN4142–176 with AvrB involve the C-terminal half of the RIN4 peptide. First, there is a set of antiparallel beta-strand hydrogen-bonding interactions between main-chain AvrB residues 120 to 124 (within strand β1) and RIN4 residues 167–171. Second, there is a hydrogen bond between the main-chain of RIN4 N170 and the AvrB R209 guanidinium group. These interactions allow this region of RIN4 to form an additional strand of the beta-sheet in the upper lobe of AvrB (Figure S1). Importantly, in forming this strand, the RIN4 T166 side-chain is directed toward the interior of AvrB such that the hydroxyl group hydrogen bonds with the hydroxyl group of AvrB T125, which in turn hydrogen bonds to the imidazole ring of AvrB H217 (Figure 2C).
Above this hydrogen-bonding arrangement, the aromatic rings of RIN4 Y165 and F169 as well as interleaving H167 are thrust against the surface formed from the last two turns of AvrB helix α7 (Figure 2A). This “ring-stack” of RIN4 spans from the AvrB upper lobe (R209) to the interlobe boundary (R266) (Figure 2A). These aromatic and indole side-chains stack in a nearly parallel arrangement with each other that is energetically very favorable. The guanidinium groups of AvrB R209 and AvrB R266 buttress the termini of the ring-stack, contacting RIN4 F169 and Y165, respectively, via π–π interactions such that the guanidium groups are also almost parallel to the three stacked aromatic rings of RIN4. In the middle of this ring stack, AvrB Q208 forms additional contacts with RIN4 H167 (Figure 2A). AvrB Y210 may play a weak role in RIN4142–176 binding since additional electron density from RIN4142–176 appears to be contacting this residue (Figure 1A). No additional electron density is observed for the last four residues of the RIN4 peptide (sequence REER).
RIN4 and AvrB interact in yeast two-hybrid assays and can be coimmunoprecipitated from plant tissue [9]. This interaction is direct, because both purified proteins comigrate as an in vitro complex on native gel filtration columns [15]. To test whether the aforementioned AvrB residues are involved in RIN4 binding (listed in Table 2), we mutated the relevant amino acids individually to alanine. These mutants were assayed for their ability to bind full-length RIN4 by in vitro gel filtration and/or in vivo by yeast two-hybrid assay (Table 2). In addition, the binding affinity of selected AvrB mutants and RIN4142–176 was measured by ITC (Figure 3).
The cocrystal structure revealed that the side-chains of several AvrB upper lobe residues make contacts with the N-terminal region of RIN4142–176. For example, D213 of AvrB interacts with W154 within the N-terminal portion of RIN4142–176. However, substitution of D213 to alanine did not disrupt either binding to RIN4 or the ability to trigger RPM1-mediated responses. The substitution of nearby residues of AvrB (i.e., V128A and T182A) also did not disrupt binding to RIN4 or activation of RPM1 (Table 2; unpublished data). These AvrB residues contact the N-terminal region of RIN4142–176 that contains an AvrRpt2 cleavage [14,15]. Hence, our data suggest AvrB and AvrRpt2 target two distinct amino acid sequences of this short, 35–amino acid residue region of RIN4.
In contrast, disruption of AvrB residues contacting the C terminus of RIN4142–176 strongly affected RIN4 binding. AvrBT125A and AvrBH217A were unable to bind RIN4142–176 under our ITC conditions (Figure 3). Loss of RIN4 binding activity was not due to significant structural disruptions, since AvrBT125A or AvrBH217A were properly folded, as measured by circular dichroism (unpublished data). Similar mutation of AvrB residues directly supporting the side-chains of the ring-stack also affected RIN4 binding. The single mutation AvrBR209A had very little effect on RIN4 binding (Table 2) but a triple mutation of Q208, R209, and Y210 to alanine (hereafter AvrBQRY/AAA) significantly lowered the affinity for RIN4 (kd = 9 μM for AvrBQRY/AAA) and destabilized the AvrB/RIN4 complex (Figure 3 and Table 2). Overall, two regions of the upper lobe make distinct and functional contacts with the C-terminal region of RIN4142–176: (1) AvrB residues Q208, R209, and Y210, which directly support the side-chains of the ring-stack, and (2) AvrB residues T125 and H217, which interact with RIN4 T166, between Y165 and H167 of the ring-stack.
AvrB induces phosphorylation of RIN4 [9], but it remained unclear whether AvrB must interact with RIN4 to trigger RPM1-mediated disease resistance and HR. We reasoned that if the RIN4–AvrB interaction is a prerequisite for RPM1-mediated HR or disease resistance, then AvrB mutants that do not bind to RIN4 should not trigger either phenotype. The preceding experiments identified AvrB mutants that varied in binding affinity to RIN4 (Figure 3 and Table 2). We expressed these avrB alleles in Pseudomonas syringae pv. tomato (Pto) DC3000. All of the AvrB alleles that exhibited altered function, as described below, were expressed at essentially normal levels in P. syringae (Figure S2D). Arabidopsis leaves were infected with Pto DC3000 strains expressing wild-type AvrB, each AvrB mutant, or an empty vector and assessed for RPM1-mediated HR using both trypan blue as a qualitative measure and the leakage of cellular ions into media and consequent changes in media conductivity over time as a quantitative measure (Figures 4A, 4B, S2A, and S2C). We also quantified the RPM1-mediated restriction of pathogen growth for selected AvrB mutants (Figure 4C). The results of these functional tests are summarized on the structures in Figure 2A and 2C.
V128, T182, and D213 are in a region of the AvrB upper lobe that is not required for interaction with RIN4 (Table 2). Unsurprisingly, these mutants have no effect on the ability of AvrB to activate RPM1 (Figure S2). The residues of AvrB that directly support the ring-stack of RIN4, especially R209 and Y210 (Figure 3 and Table 2), are more important for complex formation. Thus, while AvrBR209A bound RIN4 (Table 2), and only marginally diminished RPM1-mediated HR in both assays (Figure 4A and 4B), disruption of several of the ring-stack interactions in AvrBQRY/AAA drastically diminished RIN4 binding activity in vitro (Figure 3; Table 2) and resulted in significantly reduced RPM1-mediated HR (Figure 4A and 4B). These levels of altered HR correlated with slightly reduced RPM1-mediated restriction of bacterial growth. Pto DC3000(avrBR209A) and Pto DC3000(avrBQRY/AAA) grew in planta more than bacteria expressing wild-type avrB (Figure 4C).
Furthermore, mutations that disrupt the hydrogen bonding between AvrB T125 and H217 and RIN4 T166 (Figure 2A and 2C) eliminated or greatly reduced binding to RIN4 and the ability of these AvrB mutants to trigger RPM1 function (Table 2 and Figures 3 and 4). Translocation assays confirmed that AvrBT125A was delivered into Arabidopsis cells (as are other loss-of-function AvrB alleles described here; Figure S3 and Table 2). Therefore, interactions that either directly or indirectly support the ring-stack of RIN4 are a major determinant for complex formation. Taken together, these results strongly suggest that binding of RIN4 by AvrB is a prerequisite for its ability to efficiently activate RPM1.
The AvrBR266A mutation was unable to trigger RPM1 function (Figure 4). The AvrB structure contains a cavity in the large lobe [18]. AvrB R266 lies at the interlobe boundary between the major RIN4-binding groove and this cavity (Figure 2). We hypothesized that this large cavity, rich in residues that are highly conserved across the AvrB protein family (see below), could serve as a binding site for a cofactor for AvrB activity. Because the presence of AvrB renders RIN4 marginally hyperphosphorylated [9], we wondered whether AvrB is an atypical kinase. We therefore examined binding to nucleotides, including adenosine triphosphate (ATP), adenosine diphosphate (ADP), guanosine diphosphate, and nonhydrolyzable ATP and guanosine diphosphate analogs, by soaking these into AvrB crystals. We found that crystallized AvrB could bind ADP within the lower lobe pocket of the large cavity (referred to as the ADP pocket; Figures 1B, 1D, 1F, 2B, and 2D; Table 1). We observed only weak density corresponding to other nucleotides in the lower lobe pocket (unpublished data). Significant interactions with ADP involve AvrB residues Y65, R99, and R266 (Figure 2B and 2D). Y65 stacks below the adenine ring and hydrogen bonds to the 2-OH group of the ribose, while R99 and R266 form salt bridges with the alpha- and beta-phosphates, respectively. N62 and F113 also contact the adenine base and phosphates, respectively (Figure 2B and 2D). F113 also appears to contribute to proper positioning of R99 and R266. The conserved AvrB residues Y131 and D297 located at the interlobe boundary are approximately 4 Å from ADP and appear to contact it only via bridging water molecules.
Mutation of ADP-binding residues in AvrBY65A, AvrBR99A, and AvrBR266A resulted in complete loss of AvrB-induced RPM1 function (Figure 4). Mutation of ADP-interacting residues AvrBN62A and AvrBF113A resulted in partial losses of RPM1-mediated HR (Figure S2B and S2C). Surprisingly, mutations that appear to contact ADP only via water bridges, AvrBY131A and AvrBD297A, also abrogated RPM1-mediated HR and bacterial growth restriction (Figure 4). Thus, all of the AvrB loss-of-function mutants in the ADP binding cavity eliminated the initiation of RPM1 function. We note, however, that they are each produced and properly folded (Figure S2D and circular dichroism, unpublished data) and translocated into host cells (Figure S3). Although AvrBY65A and AvrBD297A failed to bind RIN4 in ITC (Figure 3), they, along with AvrBR99A, did bind RIN4 in yeast two-hybrid assay, gel filtration, or both (Table 2). We discuss these data further below.
Since RIN4 phosphorylation is induced by AvrB, it was hypothesized that AvrB may possess kinase activity. However, we detected no AvrB-dependent phosphorylation of RIN4 or RIN4142–176 using in vitro radiolabeling experiments in the presence or absence of Arabidopsis extracts. We also found no evidence for in vitro autophosphorylation of AvrB (Figure 5A). We did, however, observe phosphorylation of AvrB in the presence of wild-type Arabidopsis extracts (Figure 5A). AvrB phosphorylation was sensitive to ethylenediaminetetraacetic acid (EDTA) and heat denaturation of the plant extract by boiling prior to assays, suggesting that AvrB phosphorylation requires cations and a heat-labile plant factor (unpublished data). Phosphorylation activity was specific for AvrB, the only member of this type III effector family demonstrated to trigger RPM1 function, since related paralogs (Figure 5C) did not or only weakly incorporated radiolabeled phosphate (Figure 5A). In this assay, AvrB phosphorylation appears to be RIN4 independent, as it occurs in extracts from rin4 rpm1 rps2 mutant plants (unpublished data). Titrating increasing amounts of purified full-length RIN4 protein into extracts from rin4 rpm1 rps2 did not alter the level of AvrB phosphorylation in vitro (unpublished data). Furthermore, phosphorylation of AvrB does not require RIN4 binding, since AvrBQRY/AAA, AvrBT125A, and AvrBH217A were readily phosphorylated (Figure 4A). These data show that RIN4 is not the plant cofactor that either directly or indirectly regulates AvrB phosphorylation in planta.
In contrast, AvrB residues that directly contact ADP are important for phosphorylation (Figure 5A). AvrBR266A and AvrBY65A were partially and completely compromised for phosphorylation, respectively (Figure 5A). The requirement of ADP interacting residues for AvrB phosphorylation suggests that nucleotide binding by AvrB is critical for this event. On the other hand, AvrBY131A and AvrBD297A were phosphorylated to the same levels as wild-type AvrB (Figure 5A). This correlates with the minor contribution of these residues to nucleotide binding observed from the crystal structure. Nevertheless, AvrB Y131 and D297 are required for the triggering of RPM1 (Figure 4), suggesting their involvement in another aspect of AvrB activation.
Our data significantly extend previous observations defining the key functional regions of the type III effector protein AvrB: its upper lobe and interlobal cleft mediate contact with the Arabidopsis target protein RIN4, and the lower lobe contains a pocket suitable for binding ADP or another similarly shaped molecule (Figure 5B). We defined three correlates for triggering AvrB-dependent, RPM1-mediated disease resistance function in Arabidopsis. These are AvrB's interaction with RIN4, its binding of nucleotide, or another small molecule of similar shape, and its likely phosphorylation in the presence of Arabidopsis extract.
Our structure-based functional analysis of the AvrB-RIN4142–176 complex identified two main regions of interaction: AvrB T125 and H217 and AvrB Q208, R209, and Y210. These AvrB residues interact with bulky and aromatic RIN4 residues Y165, T166, H167, and F169 (previously termed the AvrB binding site [BBS] [15]) in a ring-stacking arrangement just C-terminal to the previously identified AvrRpt2 cleavage site (RCS2; [14,15]). Mutation of these AvrB residues interferes with the ability to trigger RPM1 function, demonstrating that physical interaction of AvrB with RIN4 is required for recognition by RPM1. AvrB residues Q208, R209, and Y210 are poorly conserved in other AvrB family members (Figure 5C), suggesting that it might be the specificity determinant for RIN4 binding.
The BBS and RCS are a functional, bipartite domain in RIN4 and approximately 11 additional proteins in Arabidopsis proteins (pfam05627). This RCS-BBS domain is widely distributed in multicellular plants evolutionarily distant from Arabidopsis, such as the moss Physcomitrella patens and the fern Cerapteris richardii (http://plantta.tigr.org) and thus may have conserved roles in the plant immune system. Within this domain, the RCS is highly conserved, while the BBS and the spacing between the RCS and the BBS is more variable. A number of other RCS-BBS proteins are cleaved by AvrRpt2 (which targets the RCS [14,15]), yet it remains to be seen whether the additional 11 BBS-containing proteins are also bound by AvrB. It is noteworthy that the N-terminal RCS-BBS domain of RIN4 does not bind AvrB [15], suggesting that the amino acid divergence between the N-terminal RCS-BBS and C-terminal RCS-BBS sequences of RIN4 is functionally relevant.
AvrB is also recognized by a second resistance gene, Rpg1-b, in soybean [19]. A random mutational analysis of AvrB identified nine individual amino acids required for induction of both RPM1-mediated HR on Arabidopsis and Rpg1-b function on soybean [20]. Among these, AvrB T125, Q164, and I215 were required for the HR phenotypes on both RPM1-expressing Arabidopsis and Rpg1-b–expressing soybean. All of these residues are in the AvrB upper lobe. The hydrophobic I215 lies underneath T125, and its mutation would likely disrupt the structural integrity of this region, and hence interaction with RIN4. Q164 is not solvent exposed and is also likely to disrupt AvrB structure when mutated. S268 is also in the lower lobe, and its substitution to isoleucine abrogated RIN4 binding in yeast two-hybrid assays [20]. The polar side-chain of S268 is partially buried and its mutation to a residue with a bulkier, nonpolar side-chain could have detrimental effects on the structure of AvrB, thus resulting in loss of RIN4 binding. A RIN4 ortholog is present in soybean and possesses an intact BBS domain. Whether this soybean RIN4 ortholog interacts with AvrB and is required for Rpg1-b–mediated resistance remain to be determined.
We also observed binding of ADP to the lower lobe pocket of AvrB (Figures 2 and 5B). Nucleotide contact residues (AvrB Y65, R99, and R266) are conserved between AvrB family members (Figure 5C) and are required for RPM1-mediated disease resistance responses. Thus, the ability of AvrB to trigger RPM1 function in Arabidopsis requires both interaction with RIN4 and an intact nucleotide binding pocket. Strikingly, Rpg-1b–mediated HR in soybean also required an intact AvrB nucleotide-binding pocket [20]. In these studies, AvrBR266A did not elicit an HR in soybean. In addition, AvrB G46 and A269, which make contacts with ADP in our crystal structure, were also required for this activity in soybean [20].
We hypothesize that AvrB binds to ADP, or another similarly shaped nucleotide, and interacts with RIN4 to induce RIN4 phosphorylation in Arabidopsis. If so, then AvrB might have a kinase or protokinase activity required to phosphorylate RIN4. Tertiary structure-matching programs such as DALI or MSD-Fold did not reveal significant structural similarity between AvrB and known kinases. However, the AvrB structure is similar to typical Ser/Thr protein kinases such as cAMP-dependent kinase, in that both contain a bilobal structure with a large lobe composed predominantly of alpha-helices and a small lobe with a mixed alpha-helix beta-sheet content [18]. We present a model of a ternary AvrB/RIN4/ADP structure created by superimposing the AvrB/RIN4 and AvrB/ADP coordinates and subsequent insertion of the ADP coordinates into the AvrB/RIN4 structure (Figure 6A). In this model, the distance between the oxygen atoms of the beta-phosphate of ADP and the T166 Oγ atom is quite short (4.2 Å), indicating that RIN4 T166 is a strong candidate for phosphorylation by AvrB.
We compared our modeled ternary complex to the structure of a ternary complex between cAMP-dependent kinase, AMPPNP, and an inhibitor peptide in which the acceptor serine 17 is replaced by an alanine [21] (Figure 6B). This comparison revealed that AvrB residues necessary to elicit an RPM1-dependent HR correspond in position to residues critical for the kinase activity of this cAMP-dependent kinase. Also, the positions of the phospho-acceptor in the inhibitor peptide and the putative acceptor T166 of RIN4 are similarly positioned. In cAMP-dependent kinase, the nucleotide binds in a cavity between the small lobe and large lobe, with the small lobe making up the majority of the interactions, whereas in AvrB the nucleotide binds in the major cavity of the large lobe. As a result, the orientation of the nucleotide is different between the two structures.
In cAMP-dependent kinase D168 deprotonates the acceptor serine/threonine residue and T201 positions D168 by hydrogen bonding. A similar role can be envisioned for AvrB H217 and T125, respectively, with regard to the putative acceptor T166 of RIN4. In addition, cAMP-dependent kinase K72 plays a role in phosphate binding and in stabilization of the transition state. It is analogous in position to R99 in AvrB. Although R266 in AvrB could also play this role, this residue is spatially most similar to D184 in cAMP-dependent kinase, which is involved in ligating a Mn2+ ion that chelates the terminal phosphate group. In the AvrB/ADP structure, we could not definitively observe a Mg2+ ion. The mechanism of metal binding in AvrB complexes remains to be elucidated.
Finally, protein kinases contain an activation loop joining the two lobes whose phosphorylation is necessary for protein kinase activity. In cAMP-dependent kinase, phosphorylation of T197 in the activation loop (residues 191 to 199) is necessary for kinase activity. In AvrB, residues 115 to 121 could form the activation loop, and three residues in this loop (T118, S119, and T121) could serve as phosphorylation sites. Additionally, this loop also borders the region (residues 120 to 125) that forms an antiparallel beta-sheet with residues 166 to 171 of RIN4. Hence, phosphorylation of this loop region may also affect RIN4 binding.
Despite this plausible similarity to kinases, pure AvrB neither autophosphorylates nor transphosphorylates pure RIN4 in vitro (unpublished data). However, we found that AvrB is phosphorylated in the presence of Arabidopsis extracts, suggesting that AvrB's possible kinase activity would require accessory plant factors. In vitro, this phosphorylation event is RIN4 and RPM1 independent. The P. syringae type III effector avirulence protein Pseudomonas tomato (AvrPto) is also phosphorylated in the presence of plant extracts independently of its corresponding plant disease resistance proteins, Pto and Prf [22]. Additionally, NopL and NopP are TTSS effector proteins from Rhizobium sp. NGR234 that are phosphorylated in the presence of protein extracts from L. japonicus [23,24]. Hence, there is a class of type III effector proteins that are phosphorylated once delivered to the host cell. We anticipate that this modification is linked to effector activation in all of these cases.
AvrB phosphorylation in the presence of Arabidopsis extract is dependent on the AvrB nucleotide-binding residues we defined, such as Y65 and R266. While nucleotide binding residues are required for AvrB to be phosphorylated, they are not sufficient, because AvrC and Xanthomonas campestris campestris avirulence protein (AvrXccC) encode the conserved nucleotide-binding region and are not phosphorylated in the presence of Arabidopsis extract. These data suggest that nucleotide bound to AvrB is required for the recruitment and/or function of a host kinase. Alternatively, AvrB could act as a “protokinase” that lacks intrinsic phosphor-transfer activity that can be enhanced by association with plant accessory protein(s). We speculate that each AvrB family member has evolved to usurp a plant species-specific protein that contributes to their activation by phosphorylation following delivery into host cells.
The ability of AvrB to trigger RPM1 function requires its nucleotide-binding pocket, which in turn is required for AvrB phosphorylation induced by an unknown Arabidopsis cofactor(s) or kinase, and interaction with RIN4. Neither AvrBY65A, AvrBY131A, nor AvrBD297A nor bound RIN4 in ITC (Figure 3), although all three did in gel filtration and/or yeast two-hybrid assays (Table 2). The most parsimonious explanation for these binding data requires very slow association and dissociation rates of complex formation for AvrB and RIN4. For the wild-type proteins, these slow rates result in a modest equilibrium (kd approximately 1 μM) measurable by ITC and sufficient for isolation of AvrB–RIN4 complexes when the two proteins are incubated together prior to gel filtration chromatography. Similarly, the two proteins would likely interact when coexpressed during yeast two-hybrid analysis. We propose, however, that AvrBY65A, AvrBY131A, and AvrBD297A preferentially decrease the on-rate for AvrB–RIN4 complex formation. Hence, complexes could still form and be isolated by gel exclusion chromatography or inferred from yeast two-hybrid data due to the very slow off-rate. However, the resulting decrease in the overall equilibrium constant would prevent accurate affinity measurement by ITC. Implicit in this model is the likely requirement for a conformational change in AvrB that is slow and required prior to the binding of RIN4. Importantly, Y131 and D297 reside at the interlobal boundary of AvrB, and their substitution can easily be envisaged to shift the conformational equilibrium of AvrB and lock it in a state unfavorable for RIN4 association. This interpretation is consistent with the finding of Ong and Innes (2006) that AvrBD297A enhanced binding of AvrB to RIN4 in their yeast two-hybrid system.
Our data are consistent with the activity of the AvrB protein family (Figure 5D). None of the AvrB homologs induce RPM1 function, although the Xanthomonas campestris protein AvrXccC interacts with RIN4 in yeast two-hybrid experiments and binds RIN4 weakly in vitro (unpublished data). Most of the AvrB-RIN4 contact residues are poorly conserved, particularly those in α6 and α7 helices, including AvrB Q208, R209, and Y210, which support the ring-stack of RIN4. For example, these AvrB residues correspond to AvrC residues A239, A240, and S241. AvrB T125, located on β1, is essential for RIN4 interaction and is conserved in all homologs of AvrB. Hence, the regions of AvrB that support the ring-stack of RIN4 appear to contribute to the functional specificity of AvrB for RIN4-dependent, RPM1-mediated HR. By contrast, the important ADP-contacting residues Y65, R99, and R266 in the lower lobe pocket are conserved in all AvrB homologs, suggesting that nucleotide binding is a core function for this entire type III effector family. However, an intact ADP binding cavity is not sufficient for phosphorylation of AvrB by Arabidopsis extracts, since the AvrC and AvrXccC possess them (including N62 and F113) but are not readily phosphorylated (Figure 5A). Furthermore, while AvrXccC interacts with RIN4, at least in yeast, and possesses the nucleotide-binding pocket, it does not trigger RPM1. This suggests that RIN4 interaction and nucleotide binding are not sufficient for the activation of RPM1 by AvrB family members.
AvrB Y131 and D297 are 4 Å away from the nucleotide binding site and are located in the solvent-exposed region of the interlobe cleft. Many of the solvent-exposed residues of the interlobe cleft are highly conserved within the AvrB family (Figure 5C). The electrostatic surface of the interlobe cleft and the sheer size of the cleft highly suggest that this might be an active site required for an as-yet-undefined activity of AvrB, or for the docking of an Arabidopsis protein that is required for AvrB phosphorylation. We speculate that the divergent sequences of the AvrB paralogs, centered on the interlobe cleft, are critical for recruitment of plant species-specific cofactors that enhance nucleotide turnover on each AvrB family member and/or align substrates of AvrB with potential catalytic residues in the cleft. We speculate that this set of atomic events would be required to trigger RPM1 (or other NB-LR) function in diverse plant species.
Besides acting as an avirulence factor to trigger resistance in Arabidopsis and soybean, AvrB can also serve as a virulence factor in susceptible hosts. In the absence of a functional Rpg1-b gene, AvrB enhances the growth of P. syringae on susceptible soybean cultivars [19] Similarly, AvrB also induces a cytotoxic yellowing response on Arabidopsis plants lacking RPM1 that is attributable to either its function in virulence or a weak disease resistance response [17]. Mutation of AvrB T125, R266, or S268 to alanine abrogated the virulence phenotype in soybean and compromised the chlorosis phenotype in Arabidopsis [20]. Combined with our structural data, these results indicate that RIN4- and ADP-binding regions, as well as functions provided by the interlobe cleft, are required for the virulence activity of AvrB. Therefore, interaction with RIN4, nucleotide binding, and host phosphorylation are correlated with both the virulence activity of AvrB and for its recognition by independent plant NB-LRR proteins. This corroborates the “guard” model hypothesis where NB-LRR proteins monitor the activity of type III effectors for recognition rather than direct interaction [4,6].
Our three-dimensional crystal structures of AvrB in complex with its host target RIN4 or the ADP, and the combined functional studies detailed here and in Ong and Innes (2006), suggest a plausible series of events required for both AvrB virulence activity on susceptible hosts and for its ability to trigger disease resistance recognition in two plant species: AvrB is delivered by the TTSS of Pto DC3000. Inside the host cell, AvrB binds to a nucleotide, or another small molecule of similar shape. The AvrB–nucleotide complex recruits a plant cofactor that transforms AvrB into a kinase capable of autophosphorylation. Alternatively, the nucleotide-bound form of AvrB mimics a substrate for an unknown plant kinase and becomes phosphorylated itself. Because soluble AvrB can be labeled in an Arabidopsis extract, we infer that phosphorylation of AvrB is independent of, and hence might precede, myristoylation and plasma membrane localization [17]. Phosphorylated AvrB becomes myristoylated and directed to the plasma membrane. At the plasma membrane, AvrB interacts with RIN4, and its conformation is altered. This stable heterodimer guides RIN4 phosphorylation. The AvrB–RIN4 complex, and potentially the phosphorylated form of RIN4 itself, triggers RPM1-mediated activation of disease resistance. RIN4 functions as a negative regulator of basal defense [25]. Hence, in the absence of RPM1, AvrB is similarly activated and subsequently interacts with, and indirectly induces post-translational modifications of RIN4 and other BBS-containing proteins [15] in order to curb basal defense responses and contribute to disease.
For expression in P. syringae, AvrB-HA downstream of the AvrRpm1 promoter [17,26] was constructed by PCR amplification using primers that incorporated an XhoI site upstream of the AvrRpm1 promoter (XhoI-AvrRpm1p) and a BamHI site downstream of the HA epitope tag (BamHI-HA). The resulting PCR product was digested with the restriction enzymes XhoI and BamHI and cloned into the broad host range vector pBBR1 MCS-2 [27] digested with the same enzymes. Mutations of AvrB were generated by PCR using PFU turbo high-fidelity polymerase (Stratagene, http://www.stratagene.com). Overlapping primers incorporating the mutation of interest were synthesized, and PCR was conducted using the sense primer with BamHI-HA and the antisense primer with XhoI-AvrRpm1p. The resulting PCR products were gel purified, combined, and used as a template for a second PCR using the XhoI-AvrRpm1p and BamHI-HA primers. The resulting PCR product was cloned into the TOPO TA cloning vector (Invitrogen, http://www.invitrogen.com) and sequenced to ensure that no additional mutations had been introduced. The insert was then cleaved using the restriction enzymes XhoI and BamHI and cloned into pBBR1 MCS-2 digested with the same enzymes. pBBR1 MCS-2 containing the mutant AvrB-HA genes were then introduced into Pto DC3000 by triparental mating.
AvrB alleles from X. campestris pv. campestris strain 8004 (Xcc), P. syringae pv. glycinea race 0 (AvrC), and P. syringae pv. syringae strain B728A were PCR amplified from the corresponding bacterial strains and cloned into TOPO-TA (Invitrogen) cloning vectors (Z. Nimchuk and J. L. Dangl, unpublished data). To add an N-terminal glutathione-S-transferase (GST) tag, alleles were PCR amplified from TOPO-TA vectors using oligonucleotides that incorporated an N-terminal TEV cleavage site and subcloned into the pENTR D-TOPO vector (Invitrogen). The genes were then recombined into pDEST-15 vector using the LR Clonase enzyme mix according to manufacturer's instructions (Invitrogen). Similarly, mutated versions of AvrB were amplified from TOPO-TA and subcloned into pENTR D-TOPO followed by recombination into pDEST15 using LR recombination enzymes (Invitrogen).
AvrB in pProEX-HTa was induced with 0.75 mM isopropyl-β-d-thiogalactopyranoside (IPTG) at 18 °C for 6 h in BL21 Rosetta cells (Stratagene). All protein purification steps were performed at 4 °C. Cell pellets were resuspended in buffer A (20 mM Tris [pH 8.0], 300 mM NaCl, and 10 mM imidazole) plus one “Complete EDTA-free” protease inhibitor tablet (Roche, http://www.roche.com), a few crystals of Lysozyme (Sigma, http://www.sigmaaldrich.com), and DNase (Sigma). After resuspension, cells were lysed using an Avestin Emulsiflex-C5 (Avestin, http://www.avestin.com) and centrifuged for 45 min at 15,000 rpm in an SS-34 rotor. The supernatant was loaded on to a 5 ml High Trap chelating column (GE Healthcare, http://www.gehealthcare.com) preloaded with nickel as described in the manufacturer's instructions. The column was then washed with 10 column volumes of buffer A, followed by 10 column volumes of buffer A augmented with 50 mM imidazole. Specific elution of AvrB was performed with 5 column volumes of buffer A containing 400 mM imidazole. Relevant fractions were pooled and dialyzed overnight at 4 °C in low-salt buffer containing 20 mM Tris (pH 8.0) and 100 mM NaCl in the presence of tobacco etch virus (TEV) protease to facilitate removal of N-terminal His tags. Removal of tags was verified by SDS-PAGE. The dialysate was loaded on an 8-ml Source Q (GE Healthcare) anion exchange column and eluted with a 0 to 400 mM NaCl gradient. If samples were not sufficiently pure at this point, relevant fractions were concentrated to approximately 10 ml and applied to a HighPrep 26/20 Sephacryl S200 (GE Healthcare) column equilibrated with 20 mM HEPES (pH 7.5), 150 mM NaCl, and 1 mM DTT. Purified protein was exchanged into 20 mM Tris (pH 8.0), 50 mM NaCl, and 3 mM DTT; concentrated to approximately 20 mg/ml and flash-frozen using liquid N2; and stored at −80 °C.
All AvrB mutants were cloned into the PD15 plasmid (GATEWAY; Invitrogen) and thus isolated as TEV-cleavable GST fusions. Induction was also performed with 0.75 mM ITPG at 18 °C for 6 h in BL21 Rosetta cells (Stratagene). Cell pellets were resuspended in buffer B (20 mM Tris [pH 7.5], 300 mM NaCl, and 1 mM DTT) and lysed as described for wild-type AvrB. Clarified lysates were loaded on a 5-ml High Trap glutathione column (GE Healthcare). The column was then washed with 10 column volumes buffer B, followed by specific elution by 3 to 5 column volumes of buffer B plus 10 mM glutathione. Eluted protein was digested overnight at 4 °C with TEV protease, and completely digested protein was diluted 5-fold, loaded on an 8-ml Source Q column, and eluted with a 0 to 400 mM NaCl gradient. Relevant fractions were then concentrated to less than 1 ml, flash-frozen in liquid N2 in approximately 250-μl aliquots, and stored at −80 °C.
RIN4 was cloned into pGEX-6P-1 as a GST-fusion cleavable with PreScission protease (GE Healthcare) and expressed in RIL codon-plus cells (Stratagene). Cells were grown to an OD of approximately 0.4 at 37 °C, and then the temperature was decreased to 25 °C for 45 min, and cells were induced for 3 h with 0.5 mM IPTG. Cell pellets were resuspended in buffer C (20 mM sodium phosphate [pH 6.5], 2 mM DTT, 1 mM EDTA) plus one “Complete EDTA-Free” protease inhibitor tablet (Roche), a few crystals of Lysozyme (Sigma), and DNase (Sigma). Cells were lysed as described for AvrB. Clarified lysates were then loaded on a hand-poured 20-ml Fast Flow S (GE Healthcare) column, washed with low-salt buffer, and then eluted with a 20 column volume gradient of buffer C plus 0 to 500 mM NaCl. The resulting broad peak was concentrated to 50 ml, 50 units/ml PreScission protease was added, and the mixture was dialysed overnight into buffer D (20 mM HEPES [pH 7.5], 50 mM NaCl, 2 mM DTT). Completely digested protein, as verified by SDS-PAGE, was then loaded on a 8-ml Source S column (GE Healthcare) and eluted with a 0 to 400 mM NaCl gradient. Relevant fractions were then pooled and again dialyzed overnight in buffer C. The dialysate was then run on a 8-ml Source Q column (GE Healthcare) and eluted with a 0 to 400 mM NaCl gradient. The purity of the samples was verified by SDS-PAGE, concentrated to approximately 2.5 mg/ml, flash-frozen in liquid N2 in approximately 250-μl aliquots, and stored at −80 °C.
Gel filtration experiments of the mutant AvrB proteins were performed by loading approximately 0.3 to 0.5 mg of protein either alone or with roughly equimolar RIN4 in a volume of 1 ml onto a hand-poured calibrated 16/70 Superdex S-75 column. Flow rate was 0.9 ml/min, and 3-ml fractions were collected from 30 to 80 ml over a 150-ml run.
Circular dichroism experiments were run on a Pistar-180 Circular Dichroism/Fluorescence spectrophotometer (Applied Photophysics, http://www.photophysics.com). Samples at approximately 0.1 mg/ml were exchanged into a buffer containing 20 mM potassium phosphate (pH 7), and placed in a 0.1-cm cuvette, and scans were taken from 185 to 260 nm with 0.2-nm increments and 30,000 repetitions per increment.
ITC experiments were performed on a VP-ITC microcalorimeter (MicroCal, http://www.microcalinc.com). To verify binding, we used concentrations of wild-type and variant AvrB ranging between 5 and 25 μM AvrB. RIN4142–176 peptide with concentrations ranging from 50 to 450 μM was titrated in 6-μl injections with stirring at 255 rpm. Experiments involving wild-type AvrB and full-length RIN4 used 6 μM and 120 μM, respectively. Once binding was confirmed (for wild-type and QRY/AAA), experiments were repeated in triplicate. Nonbinding was confirmed by increasing the concentrations of both the proteins to as much as 25 μM and 450 μM, respectively. Nonbinding variants were confirmed by repeating the experiment twice. Thermodynamic parameters were fit to the data using Origin v 7.0383 software (OriginLab, http://www.originlab.com).
Conditions for crystallization of AvrB and AvrB/RIN4142–176 were similar to those reported for AvrB previously [18]. Free AvrB was crystallized by vapor diffusion at 4 °C of a 1:1 mix of protein (8 to 10 mg/ml) with well solution (100 mM glycine [pH 9.0], 20% to 30% polyethylene glycol 550 monomethyl ether [PEG 550 MME]). AvrB/RIN4142–176 was also crystallized by vapor diffusion at 18 °C of an approximately 1:2 to 3 ratio of protein to peptide, mixed with an equal volume of well solution (100 mM Tris [pH 7.5], 20% to 30% PEG 550 MME). Initial crystals were of poor quality and were heavily twinned. This crystalline mass was resuspended in 50 μl of well solution, broken up and serially diluted 10-, 100-, and 1,000-fold, and used for microseeding. In general, seeds yielded suitable quality crystals within 2 d. AvrB/RIN4142–176 crystals belonged to space group P2(1) with cell dimensions a = 45.9 Å, b = 58.2 Å, c = 119.8 Å, and β = 89.9°, which corresponds to a very different packing than the P6(5) crystals found for crystals of free AvrB [18]. Soaks of nucleotide were performed by exchanging drop and reservoir solutions with 20 μl of 27% PEG 500 MME and 100 mM Tris 7.5 (with and without 5 mM MgCl2), followed by a final exchange in the drop of this solution plus 5 mM nucleotide. Nucleotides were soaked for approximately 1 d. Following soaks, crystals were found to pack in the P6(5) space group, with cell dimensions a = b = 122.7 Å, c = 64.1 Å, which is within 2.5 Å of the cell dimensions reported for free AvrB [18]. All crystallization solutions described in this section are inherently cryoprotective, and no further cryoprotection was found to be necessary.
AvrB/RIN4142–176 diffraction data were collected on a Rigaku RU-H3R (http://www.rigaku.com) rotating anode generator equipped with Osmic confocal “blue” optics, and diffraction intensities were recorded on an R-Axis IV++ image plate system. For the ADP-soaked crystals, diffraction data were collected at the SER-CAT beamline (ID-22; Advanced Photon Source, http://www.aps.anl.gov). All structures were solved by molecular replacement using AMoRe [28] using the previously solved free AvrB structure (PDB code 1NH1) [18]. Upon molecular replacement followed by rigid-body refinement, simulated annealing, and individual B factor refinement using CNS [29], difference electron density could be found for both the peptide and nucleotide (see Figure 1A and D). Peptide and nucleotide were then modeled into the resulting difference density using the program O [30]. Definitions of ADP torsions in O, as well as topology and parameter files for all nucleotides, Tris, and trifluoroacetic acid, were taken from the Hic-Up server (http://xray.bmc.uu.se/hicup). This was followed by iterative cycles of simulated annealing, B factor refinement, and water picking to reach the results shown in Table S2. In addition, restrained 2-fold noncrystallographic symmetry was used during refinement of the AvrB/RIN4142–176 structure. From two AvrB/RIN4142–176 complexes in the asymmetric unit, 5,389 atoms are modeled, including residues 26 to 53 and 56 to 321 of AvrB and residues 150 to 172 of RIN4, in both complexes. In addition, 325 water molecules, four Tris molecules, and two trifluoroacetic acid molecules were included in the model. For the AvrB/ADP structures, there are 1,503 atoms, including residues 16 to 319 of AvrB, 1 nucleotide, one Tris molecule, and 82 water molecules.
For Western blot analyses, 1.5-ml overnight cultures grown in KB with the appropriate antibiotics were pelleted, washed with hrp gene-inducing minimal media [26], and resuspended to an OD600 of 0.1 in minimal media. Then, 2.5 ml of the 0.1 OD600 culture was induced overnight and spun down the next day. Pellets were resuspended in 250 μl of 20 mM Tris/HCl (pH 8.0) and sonicated twice for 10 s with a 1-min interval between. The sonicated culture was spun down at 4 °C for 20 min at 20,000g. Then, 200 μl of the supernatant was removed carefully so as not to disturb the pelleted and centrifuged again at 4 °C for 20 min. Next, 150 μl of the supernatant was carefully removed and soluble protein quantified. And 20 μg of protein of soluble protein from the wild-type and mutant AvrB-HA expressing Pto DC3000 strains was loaded onto SDS-PAGE gels after equalizing volumes with 20 mM Tris/HCl (pH 8.0), 6× Laemmli buffer was added, and it was boiled for 5 min. Immunodetection was performed by standard methods using anti-HA antibodies (Roche) at a dilution of 1:1,000.
Lactophenol–trypan blue was used to visualize dead cells 5 h postinoculation with Pto DC3000 expressing wild-type and mutant AvrB-HA constructs as previously described [31]. Electrolyte leakage assays were carried out as previously described [31]. Briefly, fully expanded leaves from 3-wk-old plants were hand-inoculated with 0.1 OD600 (approximately 5 × 107 cfu/ml) Pto DC3000 DC3000 expressing wild-type or mutant AvrB-HA constructs. At 2 h after infection, 7.5-mm leaf discs were collected and washed extensively with distilled water for 1 h. Four leaf discs were placed in a tube with 6 ml of distilled water (four replicates per treatment), and conductivity was measured over time with a conductivity meter (model 130; Orion Research, http://www.thermo.com).
Plant extracts were prepared by grinding two or three 2-cm2 leaves to a fine powder in liquid nitrogen using a mortar and pestle. Then, 1 ml of grinding buffer (20 mM Tris/HCl [pH 8.0], 50 mM NaCl, 0.01% Triton X-100, 5 mM DTT) was added to the powder in a 1.5-ml microcentrifuge tube, and the mixture was vortexed for 30 s, followed by centrifugation for 10 min at 2,000g to remove large cell debris. The supernatant was collected and used as total plant extract. The 20-μl reactions contained 100 ng of purified AvrB allele or AvrB mutant, 1 μg of plant extract, 10 μM [γ32P]ATP (1.2 μCi; Amersham Biosciences, http://www.amershambiosciences.com), 100 μM ATP, and 10 mM MgCl2. Reactions were allowed to proceed for 10 min and terminated by adding 5 μl of 5× Laemmli buffer and boiling for 5 min. Reactions were loaded onto 12% SDS-PAGE gels, and incorporated radiolabel was visualized by autoradiography. As an equal loading control for proteins used in the kinase reactions, the SDS-PAGE gels were stained by Coomassie blue after detection.
Pto DC3000 strains containing the wild-type or mutant AvrB-HA constructs were streaked out onto King's medium B (KB) plates containing the appropriate antibiotics and incubated at 28 °C overnight. Bacteria were then scraped off the plate and resuspended to an OD600 of 0.0002 (approximately 1 × 105 cfu/ml) in 10 mM MgCl2. Three-week-old plants were hand-inoculated with the diluted bacterial solution. Each sample was collected in quadruplicate using four leaves for each time point (16 discs per time point). Leaf discs were bored from the infiltrated area, ground in 10 mM MgCl2, and serially diluted to quantify bacterial numbers.
For yeast two-hybrid analysis, avrB and its mutant derivatives were cloned into the Gateway-compatible LexA binding domain (BD) fusion vector pEG202 using LR Clonase II enzyme mix (Invitrogen). The LR reaction was left to proceed overnight at 16 °C. Yeast two-hybrid analysis was performed using the MATCHMAKER LexA system (Clontech, http://www.clontech.com) following the manufacturer's protocols. The yeast strains used in this study are EGY48 (Clontech) and RFY206. RIN4 was expressed from the plasmid pJG4–5 as B42 activation domain (AD) fusions and transformed into yeast strain EGY48 (MATα). avrB and its mutant derivatives were expressed from plasmid pEG202 and transformed into the yeast strain RFY206 (MAT a) carrying the lacZ reporter plasmid pSH18–34 (+pSH18–34). Preparation of highly competent yeast cells and small-scale lithium acetate transformations were performed using the Frozen-EZ Yeast Transformation II Kit (Zymo Research, http://www.zymoresearch.com). The RFY206 (+pSH18–34) transformants carrying pEG202:AvrB and its mutants were selected on minimal SD glucose agar base [0.7% yeast nitrogen base without amino acids and 2% bacto-agar supplemented with –Ura–His Dropout (DO)] (Qbiogene, http://www.qbiogene.com). The EGY48 transformants expressing pJG4–5 containing RIN4 were selected on minimal SD glucose agar base supplemented with –Trp DO (Qbiogene). After plating, the plates were incubated for 3 to 4 d at 30 °C until colonies appeared.
Pairwise matings were set up between RFY206 (+pSH18–34):LexABD-avrB strains and EGY48:B42AD-RIN4 or EGY48:B42AD. The standard yeast mating procedure was followed according to the manufacturer's protocols (Clontech Yeast Protocols Handbook). A 100-μl aliquot of the mating culture was spread on SD Glucose Agar base supplemented with –Ura/–His/–Trp DO to select for diploids. Plates were incubated at 30 °C for 3 to 4 d to allow yeast cotransformants to form visible colonies. In the diploid strain, the two reporters are LEU2 and lacZ. To assay for protein–protein interactions, yeast cotransformants were replica plated onto two different selective media containing galactose (Gal) to induce the expression of B42AD-RIN4 protein. Plates were incubated at 30 °C for 3 to 4 d until growth was detected.
1. SD Gal agar base supplemented with –Ura/–His/–Trp DO to confirm the nutritional phenotypes of the diploid by selecting for the LexABD, B42AD, and pSH18–34 vectors.
2. SD Gal agar base supplemented with –Ura/–His/–Trp/–Leu/X-Gal DO to screen for Leu2 and lacZ reporter gene expressions. Growth and blue color were monitored based on activation of the reporter genes and were scored as a positive interaction between the fusion proteins. No interaction was scored if the replicates grew only on Gal/SD/–Ura/–His/–Trp DO plate and not on agar base supplemented with Gal/SD/–Ura/–His/–Trp/–Leu/X-Gal.
To determine LexABD-AvrB accumulation in yeast, RFY206 yeast cultures were grown in selective medium overnight. The cultures were diluted to an OD600 = 0.15 – 0.2. The cultures were continuously monitored until an OD600 of 0.4 to 0.6 was reached. The cells were pelleted and proteins were extracted and boiled in 50 mM sodium phosphate (pH 7.0), 25 mM 2-morpholinoethanesulfonic acid (MES) (pH 7.0), 3 M urea, 1% SDS, 10% β-mercaptoethanol (BME), and 0.1% Bromophenol Blue supplemented with Complete Protease Inhibitor Pellets (Roche). The boiled samples were spun briefly in the tabletop centrifuge to pellet cell debris. A volume equivalent to 0.5 total OD600 was loaded into each well on a 10% SDS-PAGE gel. After electrophoresis, the proteins were transferred from the SDS-PAGE gel to nitrocellulose membrane support. Western blots were done by standard methods. Anti-LexBD antibody was used at 1:100 (Santa Cruz Biotechnology, http://www.scbt.com). Detection of LexA-AvrB was with the goat monoclonal antibody. Detection of the peroxidase signal of the secondary antibody-HRP conjugate was performed with ECL (Amersham Biosciences).
Selected AvrB mutants were fused to Δ79AvrRpt2 by cloning into Gateway-compatible pBBR1-MCS2 [32,33] using LR clonase and transformed into Escherichia coli DH5α. Each construct was introduced in Pto strain DC3000 by triparental matings. Infiltrations of Arabidopsis rpm1–3 were done as described [9]. The HR was scored 24 h after Pto DC3000 inoculations. Results were compared with leaves infiltrated with Pto carrying full-length avrRpt2 or an empty vector.
Crystallographic coordinates are deposited at the RCSB Protein Data Bank (http://www.rcsb.org/pdb) with the codes 2NUD and 2NUN for the AvrB/RIN4 and AvrB/ADP complexes, respectively; and 1CDK for inhibitor peptide in which the acceptor serine 17 is replaced by an alanine. The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession numbers for the paralogs used in Figure 5 are AvrB (P13835), AvrC (P13836), AvrPphC (AAV68743), AvrB2 (YP_272229), AvrB4–1 (YP_275207), AvrB4–2 (YP_273068), AvrXccC (XCC2109), AvrB Psy1 (AAN85189), AvrB Psy2 (AAF71496), and AvrB3 (YP_275207). |
10.1371/journal.pntd.0005244 | Prevalence of Chagas Disease in a U.S. Population of Latin American Immigrants with Conduction Abnormalities on Electrocardiogram | Chagas disease (CD) affects over six million people and is a leading cause of cardiomyopathy in Latin America. Given recent migration trends, there is a large population at risk in the United States (US). Early stage cardiac involvement from CD usually presents with conduction abnormalities on electrocardiogram (ECG) including right bundle branch block (RBBB), left anterior or posterior fascicular block (LAFB or LPFB, respectively), and rarely, left bundle branch block (LBBB). Identification of disease at this stage may lead to early treatment and potentially delay the progression to impaired systolic function. All ECGs performed in a Los Angeles County hospital and clinic system were screened for the presence of RBBB, LAFB, LPFB, or LBBB. Patients were contacted and enrolled in the study if they had previously resided in Latin America for at least 12 months and had no history of cardiac disease. Enzyme-linked immunosorbent assay (ELISA) and immunofluorescence assay (IFA) tests were utilized to screen for Trypanosoma cruzi seropositivity. A total of 327 consecutive patients were screened for CD from January 2007 to December 2010. The mean age was 46.3 years and the mean length of stay in the US was 21.2 years. Conduction abnormalities were as follows: RBBB 40.4%, LAFB 40.1%, LPFB 2.8%, LBBB 5.5%, RBBB and LAFB 8.6%, and RBBB and LPFB 2.8%. Seventeen patients were positive by both ELISA and IFA (5.2%). The highest prevalence rate was among those with RBBB and LAFB (17.9%). There is a significant prevalence of CD in Latin American immigrants residing in Los Angeles with conduction abnormalities on ECG. Clinicians should consider evaluating all Latin American immigrant patients with unexplained conduction disease for CD.
| Chagas disease (CD) affects an estimated 300,000 people in the United States, but warning signs for the disease have not been closely studied. CD is usually acquired in Latin America, and can remain in the body for years or decades without producing any symptoms. However, in about 30% of patients, it can eventually result in heart failure and death. The electrocardiogram can detect potential heart problems before patients begin to feel symptoms, providing an early warning. If patients with CD receive monitoring and treatment in time, it may prevent the development of more serious heart problems. We checked for the presence of CD in a sample of 327 patients with abnormal electrocardiogram readings, all of whom had resided in Latin America for at least 12 months. Seventeen patients, or 5.2% of the total sample, were positive for CD. Our study discusses the association of different electrocardiogram readings with CD in the United States, and explores variations based on patients’ gender and country of origin. The electrocardiogram can be a valuable tool for detecting and measuring the progression of CD in patients from Latin America so that proper treatment can be offered.
| Chagas disease (CD), caused by the protozoan Trypanosoma cruzi, is a slow-progressing, multi-organ disease endemic to Latin America. There are an estimated 6 million infected individuals worldwide.[1, 2] CD has an acute and chronic phase, with the chronic phase beginning 4–8 weeks after the initial infection.[3] The chronic phase begins in an asymptomatic indeterminate form characterized by seropositivity for antibodies against T. cruzi, a normal electrocardiogram (ECG), and a normal chest radiograph. Without treatment, at least 30–40% of patients with the indeterminate form will develop an advanced or determinate form 10–30 years after the initial infection.[3, 4] The advanced chronic form of CD can lead to irreversible cardiac damage resulting in conduction disease, apical aneurysms, cardiomyopathy, and sudden cardiac death. [4]
CD is traditionally associated with endemic regions in Latin America. However, given migration trends, there has been increasing recognition of populations with CD in Europe and the United States. A recent meta-analysis of European studies, which in aggregate screened 10,000 Latin American immigrants, found a CD prevalence of 4.2%.[5] Another study estimates 300,000 cases of CD in the US, contributing to 30–45,000 cases of cardiomyopathy.[1] Between 2007 and 2013, 1908 cases of CD were identified in the blood donation system.[6] In a study of blood samples in Los Angeles, 1 in 1,993 were positive for T.cruzi antibodies.[7] Nonetheless, an overwhelming majority of CD patients in the US are undiagnosed and untreated.[6, 8]
Conduction disorders are characteristic of chronic determinate Chagas disease, and are often the initial presenting finding. A study in Bolivia found ECG abnormalities in 46% of seropositive children, the most frequent being incomplete right bundle branch block (RBBB).[9] Another study in Mexico found that ECG abnormalities including RBBB were significantly higher among seropositive versus seronegative individuals.[10] In a sample of 1,389 people in a rural community of Brazil with a T. cruzi prevalence of 6.6%, ECG abnormalities were observed in 43.5% of seropositive compared with 18.3% of seronegative individuals.[11] Further, ECG abnormalities can help identify patients who are at higher risk of developing impaired systolic function. The presence of ECG abnormalities at baseline was a significant predictor of decrease in left ventricular ejection fraction (LVEF) after 17 months of follow-up in a cohort of Brazilian patients.[12]
Conduction abnormalities and cardiomyopathy are also strongly associated with CD in Latin American immigrants in the United States and Europe. In our center in Los Angeles, among adult Latin American patients with nonischemic cardiomyopathy, defined as an LVEF <40%, we found a CD prevalence of 19.2%.[13] Another study in New York identified five seropositive cases among 39 immigrants from CD-endemic countries with dilated cardiomyopathy, a prevalence of 13%.[14] Among a sample of 17 T. cruzi-positive blood donors in southeast Texas, 7 (41%) exhibited evidence of cardiomyopathy on electrocardiograph.[15] In Spain, an investigation of 485 T. cruzi-positive immigrants, of whom 459 (94.6%) were Bolivian, determined 31.5% had at least one ECG abnormality.[16] The purpose of this study is to assess the prevalence of CD in a population of Latin American immigrants with conduction abnormalities on electrocardiogram in a Los Angeles County Hospital.
Olive View-UCLA Medical Center is a 377-bed Los Angeles County Hospital which serves a population of 2.1 million people within a catchment area of 999 square miles. Forty percent of this population is Hispanic/Latino. In 2013, nearly 10% of residents earned less than 200% of the federal poverty level, and 27% of adults (ages 18–64) were uninsured all or part of the year.[17]
All ECGs performed as part of regular clinical care at Olive View-UCLA Medical Center and three affiliated clinics between January 2007 and December 2010 were reviewed. This included ECGs for preoperative or routine examinations and patients who presented with non-specific clinical complaints such as chest pain, palpitations or shortness of breath. Enrollment criteria were: age 18–60 years old; an ECG with evidence of RBBB, LBBB, LAFB, and/or LPFB; and history of residence in Latin America for at least 12 months. All ECGs were examined for evidence of conduction abnormalities and classified by two board-certified cardiologists blinded to the study, with discrepancies resolved by a third board-certified cardiologist with consensus opinion. Duration of residency in country of origin and US were determined by interview/questionnaire. Exclusion criteria were: any known history of cardiac disease, including coronary artery disease, valvular heart disease, or cardiomyopathy, defined as LVEF ≤40%.
A total of 399 subjects were identified and met enrollment criteria: 67 subjects could not be successfully contacted and 5 subjects refused participation, resulting in a final study size of 327. A single 5 mL blood sample was obtained from all patients for T. cruzi serology testing, and a questionnaire regarding demographic information was completed at the time of study enrollment. Serological testing was performed through the Centers for Disease Control and Prevention (CDC). All samples underwent Enzyme-Linked Immunosorbent Assay (ELISA, Chagatest ELISA recombinant v. 3.0, Wiener Laboratories, Argentina) and Immunofluorescence Assay (IFA). Subjects were considered seropositive for CD only if both assays resulted positive.
We computed frequencies and proportions for categorical variables, and means and standard deviations for continuous variables. Chi-square tests for independence or Fisher’s exact tests, as appropriate, were used to detect associations between categorical variables, and t-tests were employed for continuous variables. All p values are two-sided, with p < 0.05 considered significant for all analyses. Analyses were conducted with SPSS software, version 23 (SPSS Inc., Chicago IL).
The study was approved by the Institutional Review Board at Olive View-UCLA Medical Center. All participants provided written informed consent prior to participating. There was no compensation for participation.
Study participants had a mean age of 46.3±10.8 years and had resided in the U.S. for a mean of 21.3±10.7 years (Table 1). Countries of origin for the study sample were Mexico (n = 197, 60.2%), El Salvador (n = 70, 21.4%), Guatemala (n = 31, 9.5%) and other (n = 29, 8.9%: Honduras 6, Peru 6, Nicaragua 5, Argentina 5, Costa Rica 2, Colombia 2, Bolivia 2, and Chile 1). Conduction abnormalities among the study group were as follows: RBBB 40.4%, LAFB 40.1%, LPFB 2.8%, LBBB 5.5%, RBBB and LAFB 8.6%, and RBBB and LPFB 2.8% (Table 2).
Seventeen patients were positive for T. cruzi by both IFA and ELISA, resulting in an overall prevalence rate of 5.2% in this cohort of patients with unexplained conduction disease. These patients had not been previously diagnosed and were unaware they had CD. In the seropositive group, the mean age was 50.8±10.7 years with a mean time of residence in country of origin of 28.1±10.0 years. The difference in mean ages (4.8 years) between the seropositive and seronegative group was not statistically significant at the p<0.05 level (p = 0.08). A much smaller proportion of seropositive patients (n = 5, 29.4%) were male, compared with the seronegative group (n = 170, 54.8%), and this difference was significant (p = 0.048). The countries of origin of the seropositive patients were as follows (prevalence within subgroup in parentheses): El Salvador 8 (11.4%), Mexico 5 (2.5%), Guatemala 2 (6.5%), and other 2 (6.9%) (Table 1). There was substantial variation between countries; the prevalence was significantly lower for Mexicans yet higher for Salvadorans (p = 0.001).
We found the following conduction abnormalities within the seropositive group: RBBB (n = 7, 41.2%), LAFB (n = 5, 29.4%), and RBBB in conjunction with LAFB (n = 5, 29.4%) (Fig 1, Table 2). No positive patients had LBBB, LPFB, or RBBB and LPFB. We calculated CD prevalence according to each type of conduction abnormality. For RBBB, 7/132 patients (5.3%) were seropositive, for LAFB, 5/131 (3.8%), and for RBBB/LAFB 5/28 (17.9%). The risk for positive CD diagnosis in patients with both RBBB and LAFB, compared to other conduction abnormalities in the sample, was five times greater (OR = 5.2, CI = 1.7–16.0, p = 0.002).
The majority of previous research investigations on CD-associated ECG findings have focused on populations residing in Latin America.[9, 11, 18] The prevalence of CD among Latin American immigrants in the U.S. has not been well studied. Based on immigration rates and the prevalence of T. cruzi infection in countries of origin, the CDC estimated a prevalence of 1.31% among Latin immigrants in the U.S.[1] Among a subset of Latin American-born patients with conduction abnormalities, we found a much higher T. cruzi prevalence of 5.2%. This group of patients may represent a high-risk group with the presence of early stage cardiac CD, and may benefit from closer monitoring and treatment.
Conduction abnormalities can serve as markers for early stage cardiac involvement in CD.[11, 19] Prior research has demonstrated RBBB and LAFB, isolated or in combination, to be more frequently present in CD.[9, 11] Similarly, our study found the highest prevalence in those with both RBBB and LAFB (17.9%), followed by RBBB (5.3%) and LAFB (3.8%). As expected, no patients with LBBB or LPFB had CD. The prevalence of CD among patients with both RBBB and LAFB approximates that among patients with nonischemic cardiomyopathy in an earlier study at our center.[13]
We found substantial variation by gender and country of origin. Although previous studies have noted more frequent cardiomyopathy in males with CD,[4] in our sample of patients with conduction abnormalities, there was a higher proportion of seropositive females. However, this could be due to other underlying causes for which we did not collect data. The variation in prevalence by country of residence has important clinical ramifications. The proportion of CD among those from El Salvador (11.4%) was more than twice the overall prevalence and over four times that of patients from Mexico. The WHO estimates a prevalence of CD of 1.3% in El Salvador,[2] but other evidence suggests active transmission is still disproportionately affecting some areas of the country.[20] Further, while the prevalence among Mexican patients was comparatively low in our study (2.5%), it still greatly exceeds the national prevalence estimate of 0.78.[2]
Ideally, patients born in Latin America would receive screening for presence of T. cruzi antibodies in primary care; confirmed positive cases should also undergo an ECG and other diagnostic tests to assess cardiac involvement. However, because up to 99% of people with CD are undiagnosed and T. cruzi infection is not routinely screened in the U.S. outside of blood banks,[6] it is likely many at-risk patients who present with ECG abnormalities within the medical system have not had a prior test for CD. When patients from endemic countries show ECG abnormalities characteristic of CD, it is essential to ensure they have been tested for T. cruzi antibodies so that this diagnosis can either be ruled out or utilized to inform subsequent treatment. In this study, one in five patients with bifascicular block (RBBB and LAFB) had T. cruzi infection.
Conduction abnormalities can serve as predictors of impaired systolic function in Chagas patients.[12, 18] Given the potential benefits of antiparasitic therapy at an earlier stage of CD,[21] timely identification of conduction abnormalities are an important criterion in assessing the urgency of providing treatment. Although the BENEFIT trial did not identify an advantage for antitrypanosomal therapy for patients who already had developed moderate to severely impaired systolic function, the study had a short follow-up period (5 years) and exhibited considerable intercountry variation in outcomes, which may reflect differences in T. cruzi strains.[22] The lack of advantage to treating patients with preexisting impaired systolic function in the BENEFIT trial underscores the importance of considering antitrypanosmal therapy before patients progress to the advanced form of chronic Chagas disease, especially since parasite persistence is a potential trigger of cardiac damage. [4, 23] In an Argentinian study with 21 years of follow-up, patients who received treatment with benznidazole were less likely to progress to a more severe Kuschnir classification compared to placebo.[21] Treatment with benznidazole may thus be a viable option for patients who exhibit RBBB or other conduction abnormalities yet no other signs of cardiomyopathy.
To our knowledge, no other study has retrospectively evaluated the prevalence of CD in patients with conduction abnormalities on ECG in the U.S. Our data demonstrate a significant presence of CD in this population, which is substantially higher than the proportion detected through blood sample surveillance. The presence of bifascicular block (RBBB and LAFB) and history of residence in El Salvador appear to be additional risk factors. Awareness of these potential risk factors can help focus screening to identify patients within the U.S. health system who have undiagnosed CD, so that proper treatment can be provided.
We did not account for potentially confounding factors such as age, diabetes mellitus, or hypertension in our analyses. The subgroup of seropositive patients was small, creating wide confidence intervals in the calculation of risk factors. Exclusion of patients with underlying cardiac disease could possibly lead to an underestimation of prevalence of CD. This study is based on a sample of patients from a Los Angeles County public hospital system; the results may not be generalizable to other locations.
|
10.1371/journal.pntd.0005551 | Improvement of a tissue maceration technique for the determination of placental involvement in schistosomiasis | Schistosomiasis in pregnancy may cause low birth weight, prematurity and stillbirth of the offspring. The placenta of pregnant women might be involved when schistosome ova are trapped in placental tissue. Standard histopathological methods only allow the examination of a limited amount of placental tissue and are therefore not sufficiently sensitive. Thus, placental schistosomiasis remains underdiagnosed and its role in contributing to schistosomiasis-associated pregnancy outcomes remains unclear. Here we investigated an advanced maceration method in order to recover a maximum number of schistosome ova from the placenta. We examined the effect of different potassium hydroxide (KOH) concentrations and different tissue fixatives with respect to maceration success and egg morphology. Placental tissue was kept either in 0.9% saline, 5% formalin or 70% ethanol and was macerated together with Schistosoma mansoni infested mouse livers and KOH 4% or 10%, respectively. We found that placenta maceration using 4% KOH at 37°C for 24 h was the most effective method: placental tissue was completely digested, egg morphology was well preserved and alkaline concentration was the lowest. Ethanol proved to be the best fixative for this method. Here we propose an improved maceration technique in terms of sensitivity, safety and required skills, which may enable its wider use also in endemic areas. This technique may contribute to clarifying the role of placental involvement in pregnant women with schistosomiasis.
| Schistosomiasis in pregnant women is associated with prematurity, low birth weight and stillbirth of the fetus. Schistosome eggs may be trapped in the placental tissue and, thus, contribute to fetal harm. As the placenta is a large organ, current microscopic histopathological examinations commonly lack sensitivity. The yield of schistosome eggs can be increased by the use of tissue maceration. However, the applied maceration procedures are labor intensive, time-consuming and cumbersome. To develop an improved maceration technique in terms of sensitivity, safety and required skills which enable its wider use in endemic areas we examined the effect of different potassium hydroxide (KOH) concentrations and different tissue fixatives with respect to maceration success and egg morphology. Placenta maceration using 4% KOH at 37°C for 24 h was the most effective method: placental tissue was completely digested, egg morphology was well preserved and alkaline concentration was the lowest. This improved technique may contribute to clarify the role of placental involvement in pregnant women with schistosomiasis.
| Schistosomiasis may involve the placenta of pregnant women when ova are trapped in placental tissue. This has been described in both Schistosoma haematobium and Schistosoma mansoni (S. mansoni) infections [1, 2, 3, 4, 5]. Lesions are predominantly located in the placental villi, the intervillous space and the decidua [6, 7]. Ova can be found with or without surrounding granulomatous inflammations [7].
Although placental involvement of schistosomiasis has been described in the literature since the beginning of the 20th century, only few cases have been published so far [1, 3, 4, 5, 6]. Diagnostic proof of schistosome ova in placental tissue is normally performed by investigation of histological cross sections. As the placenta is a large organ of which only a limited volume can be routinely analyzed and the tissue density of schistosome ova is usually low, routine histopathological examinations are not sufficiently sensitive.
Maceration techniques aim to remove soft tissue by the destruction of biomolecules. Common methods include the use of inorganic chemicals like sodium hydroxide, ammonium hydroxide and other alkaline solutions [8]. In the early 1970s a tissue maceration technique using 10% potassium hydroxide (KOH) was described to screen larger placenta specimens for schistosome ova [2]. In the studies of Sutherland et al. (1965) and Renaud et al. (1972) a total number of 322 placentas were examined for schistosome ova. By use of the maceration method, ova were found in 79 cases whereas only two cases had been detected by histological examinations. However, the applied maceration procedures are labor intensive, time-consuming and cumbersome. Therefore, this technique was rarely used in recent studies or routine diagnosis.
The aim of this study was to develop an improved maceration technique that can easily be applied, allows the examination of larger naïve and fixed placenta specimens and improves the sensitivity for ova detection. Thus, we examined the effect of different tissue fixatives on the maceration success and the effect of different potassium hydroxide concentrations on the morphology of schistosome ova.
All women signed an informed consent allowing the use of the placental tissue for scientific purposes. The scientific use of these residual samples was approved by the Ethical Committee of the University Hospital Jena, Germany.
Placental tissues were obtained from healthy women who delivered spontaneously or by caesarean section.
Two livers from Schistosoma mansoni (Puerto Rican strain) infected Swiss-Webster female mice were kindly provided by the Biomedical Research Institute (BRI), Rockville, MD 20852. Livers were kept refrigerated in 0.9% saline.
The placenta was cut into several approximately 5 x 5 cm large specimens as described by Renaud et al. (1972) and kept either in 0.9% saline, 5% formalin or 70% ethanol. Each placenta specimen was cut into several smaller specimens (1 x 1 cm) that were placed in a 50 mL tube. Subsequently, 10% potassium hydroxide (KOH) was added until a final volume of 45 mL was achieved. Tubes were gently shaken to allow the 10% KOH to cover all of the placental tissue. The 50 mL tubes were placed in a warming cabinet and incubated for 24 h at 37°C. The caps were not tightened but placed onto the tube in order to allow developing gases to escape. After 24 h the test tubes were checked for maceration success. All experiments were performed in duplicates.
As we observed an incomplete maceration process of the 5% formalin fixed placenta tissues after 24 h, specimens used for this experiment were conserved either in 0.9% saline or 70% ethanol. As described above one placenta piece was cut into several smaller pieces and placed in a 50 mL test tube. Two Schistosoma mansoni infected naïve mouse livers were cut into approximately 0.5 x 0.5 cm pieces and three pieces from different liver areas were added to each tube with placental tissue, respectively. Tubes were filled with either 10% KOH or 4% KOH and placed in a warming cabinet for 24 h at 37°C as described above. Mouse liver pieces incubated alone with 10% KOH or 4% KOH served as controls to verify that schistosomal eggs were not digested by KOH. After 24 h the test tubes were centrifuged for 10 minutes at 2,500 rpm and room temperature. The supernatants were discarded and the whole pellets were examined under the microscope at 100-fold and 400-fold magnification for eggs of Schistosoma sp. with particular attention to their morphology. The experiments were performed in duplicates.
No quantitative assessment of the eggs trapped in the mouse liver pieces was made. Therefore no experimental data about the lower limit of detection of schistosome ova in placenta pieces were collected.
After 24 hours the placental tissue conserved either in 0.9% saline or 70% ethanol was completely digested. Placental tissue fixed in 5% formalin was not fully macerated as the tube still contained a large number of undigested tissue pieces.
After 24 hours S. mansoni ova were found in all preparations. In all experimental groups with 0.9% saline + 10% KOH or 4% KOH, and 70% ethanol + 10% KOH or 4% KOH the shells of almost all ova were intact and the lateral spine was clearly visible (Fig 1A and 1B). Ova contained either a more or less decomposed miracidium or were without content. When mouse livers were directly incubated with 10% KOH all ova were nearly completely digested and detection of the lateral spine or the miracidium was challenging (Fig 1D). Egg morphology was preserved when mouse liver tissue was incubated with 4% KOH as observed in the experimental groups (Fig 1C).
In our experiments we found that placenta maceration using 4% KOH at 37°C for 24 h was the most effective method: placental tissue was completely digested, egg morphology was well preserved and alkaline concentration was lowest. In previous studies placental tissue was commonly digested in 10% KOH at 37°C [1, 4] or 56°C [2,3] for 24–48 hours. However, Renaud et al. (1972) reported, that it was usually impossible to determine the schistosomal species in digested placental tissue as the egg shape was often altered by the potassium hydroxide and the spine was not visible. In animal experiments digestion with 4% KOH for 24–48 hours was previously described for the extraction of trapped eggs in different tissues though this procedure was never used in human samples for diagnostic purpose [9].
Renaud et al. (1972) and Gelfand et al. (1970, 1971) used 250 mL glass receptacles or glass bottles, collected the sediment by a pipette and transferred it into centrifuge tubes. However, collection of the whole sediment is hard to ensure and the transfer of potassium hydroxide is hazardous for the personnel because of the risk of chemical burns. Furthermore, Renaud et al. (1972) described that the maceration process was incomplete when using 2 x 22 cm glass tubes. Therefore, in our experiments, we used disposable 50 mL tubes for the maceration and subsequent centrifugation. The tubes are easy to handle and preclude the loss of macerated tissue during the transfer process from receptacle bottom to centrifuge tube. By mincing the placental tissue the complete maceration was achieved as smaller pieces may offer a larger contact surface for the KOH solution.
Contrary, the complete digestion of 5% formalin fixed placenta samples failed in our experiments as tissue pieces were still found after 24 hours. This confirms a similar observation made in animal experiments (Tucker et al. 2013). Surprisingly, Gelfand et al. (1970, 1971) described the complete digestion of tissue samples that were fixed in 10% formalin. However, several washing steps were applied before application of KOH and a much higher incubation temperature of 56°C was used for the maceration process.
Schistosomiasis has been implicated in fetal harm including prematurity, low birth weight and stillbirth [6, 10, 11, 12, 13, 14]. The physiopathology of negative pregnancy outcomes in schistosomiasis may be multifactorial. Pregnant women with chronic schistosomiasis are known to be more frequently undernourished, anemic due to chronic iron loss, to suffer from chronic protein loss and vitamin deficiencies [15]. The role of placental involvement is not sufficiently elucidated as only a few studies have been undertaken. In two earlier studies from South Africa and the Ivory Coast a total of 343 placentas from both, normal deliveries and miscarriages, premature births or intrauterine fetal deaths were examined by digestion with 10% KOH [1, 4]. Schistosome ova were found in 79 cases but there was no evidence that placental schistomiasis had any impact on the pregnancy outcome or the development of the embryo or of the fetus. However, a symptomatic intestinal or urogenital schistosomiasis was not an inclusion criterion in this study and serology to detect an asymptomatic chronic infection was not performed, so that the percentage of placental involvement during a florid schistosomiasis could not be determined with sufficient accuracy. On the other hand there are several case reports describing that placental schistosomiasis was associated with various complications e.g. tubal pregnancy and stillbirth diagnosed by retrospective histological examinations of placental tissue [6].
Due to the described diagnostic difficulties, it has not been sufficiently investigated whether adverse birth events might be induced by a local inflammation process of the placenta caused by trapped eggs or with complex generalized inflammation due to the helminth infection resulting in increased inflammatory cytokines like tumor necrosis factor α (TNF- α) and interleukin 6 (IL-6) [13,16]. As studies have shown in vitro, first trimester trophoblasts (HTR8/SVneo cell line) showed significantly higher levels of the pro-inflammatory cytokines IL-6 and IL-8 after incubation with serum obtained from S. japonicum infected women at 32 week gestation in comparison to an uninfected control group [17]. Furthermore, incubation of the HRT8/SVneo cell line and placental cytotrophoblasts with soluble egg antigen (SEA), that was found circulating during schistosome infection, inhibited the migratory and invasive properties of the extravillous trophoblasts and also increased the pro-inflammatory cytokine levels [17, 18]. These data indicate that schistosomal antigen can activate pro-inflammatory response and might influence fetal health.
Contrary, in two recent studies from the Philippines and Uganda the effects of praziquantel treatment on the perinatal outcome of S. japonicum or S. mansoni infected mothers were examined [19, 20]. Both studies revealed that the treatment with praziquantel in the second or third trimester of gestation did not have a significant effect on the birthweight of the newborns nor on the haemoglobin levels of the mothers. However, in none of the studies the placentas were examined regarding schistosome ova and therefore no prediction can be made about the effect of placental schistosomiasis on the pregnancy outcome.
In summary, placental schistosomiasis remains underdiagnosed if examined by standard histopathological methods which are not sufficiently sensitive. Future investigations on schistosomiasis in pregnancy should include placenta maceration in order to establish the role of placental involvement especially in schistosomiasis-endemic settings. We have developed an improved maceration technique in terms of sensitivity, safety and required skills which enable its wider use in endemic areas.
|
10.1371/journal.pbio.2002257 | Lef1-dependent hypothalamic neurogenesis inhibits anxiety | While innate behaviors are conserved throughout the animal kingdom, it is unknown whether common signaling pathways regulate the development of neuronal populations mediating these behaviors in diverse organisms. Here, we demonstrate that the Wnt/ß-catenin effector Lef1 is required for the differentiation of anxiolytic hypothalamic neurons in zebrafish and mice, although the identity of Lef1-dependent genes and neurons differ between these 2 species. We further show that zebrafish and Drosophila have common Lef1-dependent gene expression in their respective neuroendocrine organs, consistent with a conserved pathway that has diverged in the mouse. Finally, orthologs of Lef1-dependent genes from both zebrafish and mouse show highly correlated hypothalamic expression in marmosets and humans, suggesting co-regulation of 2 parallel anxiolytic pathways in primates. These findings demonstrate that during evolution, a transcription factor can act through multiple mechanisms to generate a common behavioral output, and that Lef1 regulates circuit development that is fundamentally important for mediating anxiety in a wide variety of animal species.
| Humans, mice, fish, and even flies exhibit anxiety-like behavior despite the fact that their brain anatomy varies widely. This study reveals another common thread that runs through these diverse animals: the molecular origins of their shared behavior. Gene knockout experiments in mouse and zebrafish show that the molecular signal Wnt acts through the transcription factor Lef1 to inhibit anxiety in both species. The pathway is required for formation of anxiolytic neurons in a highly conserved brain region, the hypothalamus. From there, however, the process diverges. In the fish, the pathway triggers genes including corticotropin-releasing hormone binding protein (crhbp), but in mice the same pathway calls into action a different gene, Pro-melanin concentrating hormone (Pmch). By comparison, the fruit fly Drosophila activates crhbp, similar to zebrafish. Furthermore, CRHBP and PMCH show extraordinarily coordinated expression in the primate hypothalamus, indicating that they may act together downstream of Wnt and Lef1 to regulate human behavior. This work reveals the surprising finding that conserved signaling pathways can regulate common behavioral outputs through diverse brain circuits during evolution.
| Recent work has demonstrated that innate behaviors can be highly conserved across diverse animal models [1]. Individual neuronal populations that mediate these behaviors are specified during embryogenesis by transcription factors that can also be conserved across species [2]. However, molecular signaling pathways that regulate the development of common behavioral circuits have not been identified. As brain anatomy and connectivity change through evolution, it is possible that a single pathway could act through diverse molecular and cellular targets to establish a single behavioral output, which is the ultimate constraint on gene function.
Wnt/ß-catenin signaling plays important evolutionarily conserved roles in brain development, and thus represents an ideal candidate pathway to link gene regulation with the evolution of behavioral circuits. The Wnt pathway acts through Tcf/Lef transcription factors [3], and both Wnt signaling and Lef1 are required for neurogenesis in the zebrafish hypothalamus [4], an evolutionarily ancient brain structure that regulates innate behaviors [5]. However, the identity and behavioral function of Lef1-dependent hypothalamic neurons, and their degree of evolutionary conservation, are unknown. Here, we show that Lef1 is required for the differentiation of hypothalamic neurons that inhibit anxiety in both zebrafish and mice, but through divergent molecular and cellular mechanisms in the 2 species. Generation of neurons expressing corticotropin-releasing hormone binding protein (crhbp) requires Lef1 in zebrafish but not in mice, whereas neurons expressing Pro-melanin concentrating hormone (Pmch) are Lef1-dependent in mice but not in zebrafish. Furthermore, zebrafish and Drosophila have common Lef1-dependent crhbp expression in their respective neuroendocrine organs, consistent with an ancient conserved pathway that has diverged in mammals. Finally, the Genotype-Tissue Expression (GTEx) project [6] reveals a top-ranked positive correlation between CRHBP and PMCH in the human hypothalamus, suggesting co-expression and/or co-regulation. Both genes are also correlated with LEF1 expression in humans, and are expressed in the same region of the marmoset hypothalamus, consistent with a conserved regulatory pathway in primates. These findings suggest that the gene expression network regulated by a transcription factor can change during evolution while still generating a common behavioral output. Our data also suggest an anxiolytic role for Wnt signaling in the human hypothalamus, with potential implications for the etiology and treatment of anxiety disorders.
We sought to first characterize the earliest cellular defect in lef1 null zebrafish mutants [4], so that we could perform a transcriptome analysis at that stage to identify Lef1-dependent genes. Despite grossly normal morphology, mass, and brain size, lef1 mutants have a smaller caudal hypothalamus (Hc) at 15 days post-fertilization (dpf) [4], and we found that the size reduction occurred at as early as 3–4 dpf (Fig 1A and S1A and S1B Fig). At 3 dpf the tissue already contained fewer Wnt-responsive cells [7] (Fig 1B), as well as fewer serotonergic cells and ventricular GABAergic HuC/D+ neurons (Fig 1C and S1C Fig). However, th2:GFP+ dopaminergic neurons [8] were unaffected (S1D Fig), indicating that not all neuronal subtypes are Lef1-dependent. In addition, the number of BLBP+ cells was increased (S1E Fig), confirming an inhibitory role of Wnt signaling in the formation of hypothalamic radial glia [4,9].
To determine the cellular mechanism underlying the decreased populations in lef1 mutants, we measured apoptosis and proliferation. We observed an increase in p53-dependent apoptosis within the Hc at 3 dpf (Fig 1D), but no change in proliferation at 3 dpf and beyond (Fig 1E and S1F–S1H Fig). Rescue of apoptosis by loss of p53 (Fig 1D) did not restore HuC/D expression in lef1 mutants (Fig 1F), consistent with a primary defect in progenitor differentiation. To confirm a failure in neurogenesis, we performed BrdU pulse-chase experiments, and observed fewer newly born serotonergic and ventricular HuC/D+ cells in lef1 mutants (S1I Fig). To test whether Lef1 functions cell-autonomously, we transplanted cells from lef1+/- donors into the hypothalamic anlage of lef1 mutant hosts during gastrulation, and observed rescue of ventricular HuC/D expression only in donor cells (Fig 1G). Together these data suggest that Lef1 functions cell-autonomously to promote hypothalamic neurogenesis; in lef1 mutants, neural progenitors fail to differentiate and subsequently undergo cell death, leading to a smaller Hc. Our data also justified 3 dpf as the optimal time point to perform a transcriptome analysis.
To identify Lef1-dependent genes, we next performed RNA sequencing (RNA-seq) analysis of whole hypothalami dissected from 3 dpf control and lef1 mutant zebrafish embryos, and found 144 genes with an adjusted P value (AdjP) <0.1, among which 53 genes had a fold change >2 (Fig 2A, S2 Table). Most of these genes had reduced expression in lef1 mutants (Fig 2A), consistent with Lef1 functioning as a Wnt transcriptional activator [10]. Surprisingly, Ingenuity Pathway Analysis (IPA) identified Lef1-dependent genes as being most highly associated with anxiety and depressive disorder (Fig 2B and S3 and S4 Tables). In contrast, genes associated with other hypothalamus-mediated behaviors, such as feeding (neuropeptide Y [npy], agouti-related protein [agrp], and proopiomelanocortin [pomc]) or sleep (hypocretin [hcrt]), were unaffected (S2 Table). We performed in situ hybridization on 3 dpf offspring of lef1+/- incrosses and confirmed that all Lef1-dependent genes with specific detectable hypothalamic expression showed predicted changes in approximately 25% of embryos, consistent with Mendelian segregation (Fig 2C and 2D and S2A–S2C Fig). These included several known Wnt targets such as sp5a and sp5l [11] (Fig 2C), and anxiety-related genes identified from IPA (Fig 2B and 2D). Expression of neuronal markers such as crhbp and 5-hydroxytryptamine receptor 1A b (htr1ab), was lost specifically in the Hc of lef1 mutants while remaining intact in the rostral hypothalamus (Fig 2D), resulting in their relatively small fold change in whole hypothalamus RNA-seq analysis (S2 Table). In contrast, expression of other genes, such as 2 phosphodiesterase 9a (pde9a) paralogs, was lost in the rostral hypothalamus and Hc of lef1 mutants (Fig 2D and S2A Fig), consistent with lef1 expression in both regions (Fig 2C). We also observed expression of Lef1-dependent genes in the Hc of wild-type (wt) adult zebrafish (S2D Fig), suggesting the presence of Wnt activity and Lef1-dependent neuronal populations throughout life. Together these results suggested that lef1 mutants might have an anxiety-related behavioral phenotype.
lef1 mutants raised with siblings had decreased survival and size (S3A and S3C Fig). When separated at 15 dpf, mutants survived normally (S3B and S3C Fig), but were still smaller than control siblings at culture densities that maximized their growth (Fig 3A and S3D Fig), a phenotype potentially due to enhanced anxiety [12]. We then performed a novel tank diving test to measure anxiety-related behavior [13]. We found that lef1 mutant larvae had a longer latency to enter the upper half of a novel tank and spent less overall time in this zone during the initial exploration phase (Fig 3B and 3C and S1 Video), consistent with elevated anxiety. Notably, lef1 mutants travelled less distance during this phase, partially due to more frequent freezing behavior as indicated by increased time in immobility (Fig 3D and 3E and S1 Video), and again consistent with elevated anxiety. Importantly, lef1 mutants no longer displayed anxiety-related behavior after the exploration phase (Fig 3F). The body growth and anxiety phenotypes in lef1 mutants could be explained by reduced expression of multiple hypothalamic genes including crhbp (Fig 2D), which encodes a corticotropin-releasing hormone (CRH) inhibitor [14]. However, pleiotropic phenotypes in zebrafish lef1 mutants [4,15] could also contribute to defects in growth or motor behavior. Therefore, we sought to create a tissue-specific mouse knockout model to examine the hypothalamic function of Lef1, and to determine whether it is evolutionarily conserved.
Lef1 is expressed in the mouse Hc from embryonic day (E) 10.5 to adulthood [16,17], and while previously characterized Lef1 null mutants exhibit postnatal lethality and a smaller body size, no hypothalamic phenotypes were reported [18,19]. We created a mouse hypothalamus knockout model using Nkx2-1Cre and Lef1flox alleles [20,21]. We also introduced the Cre reporter RosatdTomato [22] to create the conditional knockout allele Nkx2-1Cre/+;Lef1flox/flox;RosatdTomato/+ (herein referred to as Lef1CKO) and control littermates Nkx2-1Cre/+;Lef1flox/+;RosatdTomato/+ (herein referred to as Lef1CON), which were used for all experiments. We confirmed successful recombination by tdTomato expression (S5A Fig), and loss of hypothalamic Lef1 and Wnt reporter [23] expression in Lef1CKO mice (S5B and S5C Fig), which were viable, fertile, and morphologically indistinguishable from Lef1CON littermates. However, both male and female Lef1CKO mice gained weight more slowly after weaning (Fig 4A), similar to the phenotype we observed in zebrafish lef1 mutants (Fig 3A), and again consistent with elevated anxiety [12].
To directly measure anxiety-related behavior, we used an elevated plus maze (EPM) test and found that male Lef1CKO mice spent significantly less time in the open arms and more time in the closed arms (Fig 4B) despite normal mobility (S4A Fig). In an open field test (OFT), male Lef1CKO mice spent significantly less time in the center zone (Fig 4C) despite normal mobility (S4B Fig). These results are consistent with elevated anxiety in male Lef1CKO mice. We also observed enhanced anxiety specifically in OFT with estrous female Lef1CKO mice, but not with diestrous or all females, or with EPM testing of any females (Fig 4B and 4C and S4A and S4B Fig), likely due to reported variations in anxiety-related behavior between different sexes [24] and different behavioral assays [25]. Together, these results suggest a conserved role of hypothalamic Lef1 in inhibiting anxiety.
Consistent with the neurogenesis defect we observed in zebrafish, we found fewer HuC/D+ cells in the mouse hypothalamic ventricular zone in Lef1CKO embryos at E14.5 (Fig 5A). Importantly, this effect was restricted to coronal sections in which endogenous Lef1 is expressed (S5B Fig). To identify Lef1-dependent genes in the mouse hypothalamus, we performed RNA-seq analysis of hypothalami dissected from E14.5 Lef1CON and Lef1CKO embryos, and surprisingly identified only 1 protein-coding gene that mapped to a unique locus with an AdjP <0.1 and a fold change >2, Pmch (Fig 5B and S5 Table). Pmch expression normally overlaps with Lef1 in the premammillary hypothalamus, and extends into the lateral hypothalamus (Fig 5C) [17,26]. We confirmed loss of Pmch expression in E14.5 Lef1CKO embryos by quantitative real-time PCR (qPCR) and immunostaining (Fig 5D and S5D and S5E Fig). The only other significantly affected protein-coding gene identified by RNA-seq, Ribosomal Protein L34 (Rpl34) (Fig 5B, S5 and S6 Tables), is a repetitive processed pseudogene that could not be conclusively mapped to a single genomic locus, although one copy is located adjacent to Lef1.
Reduced Pmch expression in Lef1CKO embryos was unexpected because its orthologs were not significantly affected in RNA-seq analysis of zebrafish lef1 mutants (S2 Table). To determine if any Lef1-dependent genes were conserved with zebrafish later in development, we performed another RNA-seq analysis at postnatal day (P) 22, when Lef1CKO mice begin to exhibit a growth defect (Fig 4A). In this experiment, we identified only 2 affected protein-coding genes mapped to unique loci with an AdjP <0.1: Pmch and Tachykinin receptor 3 (Tacr3) (Fig 5B, S6 Table). Tacr3 is known to be co-expressed in Pmch+ neurons, along with CART prepropeptide (Cartpt) [27]. We confirmed their reduced expression in the lateral hypothalamus of P22 Lef1CKO mice by qPCR and in situ hybridization (Fig 5D and 5E and S5E Fig), consistent with loss of Pmch+ neurons. Decreased body weight observed after ablating Pmch+ neurons [28,29] may therefore be related to an anxiolytic role for these cells [12], which is further supported by characterization of their inputs and activity [30].
Orthologs of multiple Lef1-dependent anxiety-related genes in zebrafish are expressed near Lef1 in the mouse hypothalamus, such as Pde9a and Nitric oxide synthase 1 (Nos1) at E14.5 [26], and Crhbp and Histidine decarboxylase (Hdc) in adults [16]. However, RNA-seq analysis indicated that expression of these genes was Lef1-independent in mice (S5 and S6 Tables), and we confirmed this result for Crhbp by qPCR and in situ hybridization (Fig 5D and S5E and S5F Fig). In addition, we confirmed that expression of zebrafish pmch orthologs [31] does not depend on Lef1 at either 3 dpf or 15 dpf (S6A–S6C Fig). While we cannot rule out the possibility that our RNA-seq analysis of the mouse hypothalamus lacked the sensitivity to identify other conserved Lef1-dependent genes, it is clear that the identity of Lef1-dependent neurons relevant for anxiety differs between zebrafish and mice.
Interestingly, many Lef1-dependent genes in zebrafish encoding components of anxiety-mediating transmitter pathways, such as GABA, 5-HT, and CRH (Fig 2B), have a conserved function in Drosophila anxiety-like behavior [1]. Therefore, we hypothesized that hypothalamic Lef1-dependent neurons in zebrafish may represent an evolutionarily ancient pathway. The Drosophila pars intercerebralis (PI) and pars lateralis (PL) represent neuroendocrine organs equivalent to the vertebrate rostral hypothalamus and Hc, respectively [32]. In Drosophila, a single Lef/Tcf family member, pangolin (pan), functions as a Wnt activator [33,34]. Consistent with our hypothesis, we detected specific pan expression at stage 14 and the crhbp ortholog CG15537 expression at stage 16 in the Drosophila PL primordium [32] (Fig 6A–6C). Furthermore, we observed a loss of crhbp expression in the PL of pan mutants [34] at stage 16, despite intact expression in the PI and normal PL morphology (Fig 6C–6E). Drosophila crhbp in the PL may also be anxiolytic by inhibiting CRH/CRH-like diuretic hormone in the PI [1,32,35], thus these results support a relationship between neuroendocrine Lef1 function and the development of anxiolytic Crhbp+ neurons dating back to a common bilaterian ancestor. By contrast, Pmch is a vertebrate specific gene, and Lef1-dependent Pmch+ neuronal circuitry in mice may reflect a more recent mammalian divergence that co-evolved with new brain structures [36].
Our animal models suggest that in humans Lef1 may also regulate the formation of Pmch+ and/or Crhbp+ hypothalamic neurons. To test this hypothesis, we compared the hypothalamic RNA-seq transcriptomes of 96 human individuals from the GTEx project [37] (S7 Table). Despite the fact that these data did not include prenatal samples, we found that expression of PMCH and CRHBP are both moderately correlated with LEF1, which is expressed at a relatively low level in the adult human hypothalamus (Fig 7A and 7B). Notably, PMCH and CRHBP were both within the top 100 LEF1-correlated genes, along with known Wnt targets such as Sal-like protein 4 (SALL4) [38] and SP5 [11] (Fig 7C and S8 Table).
In the course of this analysis, we noticed similar correlation profiles for CRHBP and PMCH (Fig 7A and 7B), suggesting a possible expression correlation between these 2 genes. Surprisingly, we found CRHBP and PMCH to be the most highly correlated genes with each other (Fig 7D–7F and S8 Table), a relationship that has never been reported previously. Among the top 200 PMCH- or CRHBP-correlated genes, we also found 2 Wnt ligands and 1 Wnt co-activator: R-Spondin 1 (RSPO1) [40] (Fig 7D and 7E). As a comparison, AGRP is the most highly correlated gene with Neuropeptide Y (NPY) (Fig 7G and S8 Table), consistent with their co-expression in the same hypothalamic neurons [41]. Interestingly, while Pmch and Crhbp are expressed in different regions of the mouse hypothalamus [16], they are expressed in the same hypothalamic nuclei in another primate, the marmoset according to the Marmoset Gene Atlas (https://gene-atlas.bminds.brain.riken.jp). Importantly, the results of all our correlation analyses are recapitulated on GeneNetwork (www.genenetwork.org) [42], which imported an older version of GTEx’s datasets and calculated Pearson correlation across a population (See Materials and methods). Together these data suggest co-expression of PMCH and CRHBP in the primate hypothalamus and potential regulation by LEF1-mediated Wnt signaling in humans.
In this study, we demonstrate that Lef1-mediated hypothalamic Wnt signaling plays an evolutionarily conserved role in regulating the formation of anxiolytic neurons (See Fig 8 for summary). In zebrafish lef1 mutants, neural progenitors fail to differentiate and undergo apoptosis, resulting in a smaller Hc (alternatively named the hypothalamic posterior recess, the posterior part of the paraventricular organ, or the caudal zone of the periventricular hypothalamus [4,43,44]). Any or all of the 20 anxiety-related genes that are misregulated in the zebrafish mutant (Fig 2B) may contribute to the behavioral phenotypes that we observe. Likewise, our data do not conclusively prove that crhbp+ neurons, or indeed any individual Lef1-dependent neuronal populations, mediate the effect of Lef1 on anxiety. Such a conclusion would require either rescue of the lef1 mutant phenotype by restoration of missing neurons, or phenocopy by specific ablation of the cells. However, the specific loss of Pmch+ neurons in our mouse conditional knockout (Fig 5B), combined with the unexpected expression correlation between PMCH and CRHBP in the human hypothalamus (Fig 7F), is consistent with a common role for these 2 genes in behavior. While we also cannot rule out the possibility that Lef1 mutants may have other behavioral defects, genes that are known to regulate other hypothalamus-driven behaviors, such as Npy, Agrp, Pomc, and Hcrt, are unaffected in our mutants (S2, S5 and S6 Tables). In addition, pure assessment of other behaviors cannot distinguish a direct phenotype from an anxiety-related secondary phenotype.
While the major product of Pmch, melanin-concentrating hormone (MCH), is an anxiolytic factor in teleosts [45], studies in mammals have reported it to be either anxiolytic, anxiogenic or having no effect [46,47]. In addition, the Pmch propeptide makes at least 2 more neuropeptides, neuropeptide-glutamic acid-isoleucine (NEI), and neuropeptide-glycine-glutamic acid (NGE), which are also involved in stress response and anxiety [48]. Germline Pmch mouse knockouts gain weight more slowly than controls, a phenotype originally attributed to decreased food intake [49]. However, on a different background strain, the same group reported that the knockout mice were not hypophagic, while retaining a growth phenotype [50]. Interestingly, all rodent models ablating Pmch [49–53] or Pmch+ neurons [28,29] exhibit a reduced growth rate. One possible underlying mechanism could be enhanced anxiety [12], which was not directly tested in any of these studies. Therefore, we hypothesize that in Lef1CKO mice, loss of hypothalamic Pmch+ neurons is responsible for elevated anxiety, leading to a secondary growth phenotype.
Our data suggest that the gene expression and neuronal subtypes dependent on Lef1 can change during evolution while maintaining a common behavioral output. While transcriptional networks can undergo rapid rewiring at the level of enhancer binding sites during yeast, insect and mammalian evolution [54,55], the direct transcriptional targets of Lef1 mediating hypothalamic neurogenesis are still unknown. We have identified Tcf/Lef consensus binding sites in zebrafish and mouse Crhbp and Pmch loci, but it remains important to determine whether these 2 genes are direct targets of Lef1, or are instead lost as a secondary result of neurogenesis defects in mutants. In either case, it will also be useful to understand the circuitry of Lef1-dependent neurons. While the targets of Crhbp+ neurons in Drosophila and zebrafish are unknown, the projections of Pmch+ neurons in the hypothalamus of mice and other mammals are well characterized, and the regulation of these circuits by Lef1 in these species may be linked to anatomical and functional expansion of target brain regions such as the cortex [36]. Importantly, the coordinated expression of CRHBP and PMCH in the human hypothalamus suggests that they may be co-expressed in a single neuronal cell type.
Loss of other genes important for hypothalamic neurogenesis has been shown to affect behavior [2]. Interestingly, mice lacking hypothalamic Dbx1 also exhibit a loss of Pmch+ neurons along with other populations [56]. In that study, Lef1-expressing hypothalamic nuclei were hypothesized to regulate innate behaviors outside the hypothalamic-pituitary-adrenal (HPA) axis, partly due to the observation of expanded Wnt activity in Dbx1 knockout animals. However, because our work demonstrates that Lef1 is in fact required for the genesis of Pmch+ neurons and for HPA-related behaviors, an alternative explanation is that Dbx1 functions in a parallel pathway to Lef1.
Together these results identify Wnt signaling as a link between brain development and function that allows essential behaviors to be maintained even as anatomical structures change through evolution. In addition, given the function for hypothalamic Wnt signaling in regulating postembryonic zebrafish neurogenesis [4], and the continuous expression of Lef1 in the hypothalamus of fish (S2D Fig) and mammals [16] throughout life, it would be interesting to test a possible contribution to adult behavior using temporal conditional knockout models. While Wnt signaling in the mammalian hippocampus and nucleus accumbens has been associated previously with anxiety and depression [57,58], our data demonstrate a novel requirement for pathway activity in a brain region that is highly conserved throughout the vertebrate lineage, and may prove useful for the diagnosis and treatment of hypothalamus-related anxiety disorders.
All experimental protocols were approved by the University of Utah Institutional Animal Care and Use Committee and were in accordance with the guidelines from the National Institutes of Health. Approval number: 16–09011. Zebrafish were euthanized by ice water immersion. Mice were euthanized by CO2 or ketamine/xylazine.
Zebrafish (Danio rerio) were bred and maintained in a 14:10 hour light/dark cycle as previously described [59]. Zebrafish per tank were fed with similar amount of food and treated by the staff who were blinded to the experiments. Wt strains were *AB. The following mutant and transgenic strains were used: lef1zd11 [4], Tg(top:GFP)w25 [7], Tg(dlx6a-1.4dlx5a-dlx6a:GFP)ot1 [60], Tg(h2afv:GFP)kca6 [61], Tg(th2:GFP-Aequorin)zd201 [8], p53e7 [62]. lef1-/- homozygous mutants were identified between 3 dpf and 10 dpf by DASPEI staining as described previously [15] and at or after 15 dpf by loss of caudal fin [4]; wt and heterozygous siblings were used as controls. All the zebrafish were from at least 1 single-pair breeding. Genotyping was done as described before for lef1zd11 [4] and p53e7 [63], except primers used for lef1zd11 (forward primer: 5ʹ-CACTCTCTCCAGCCCAACATT-3ʹ, reverse primer: 5ʹ-TGTTACTGTTGGGACTGATTTCTG-3ʹ).
Male and female C57BL/6J mice (Mus musculus) were group-housed with 2–5 mice per cage in a reverse 12 hour light/dark cycle with ad libitum access to food and water. Mice were 19–20 and 15–20 weeks old at the time of behavioral tests for male and female animals, respectively. Ai9 reporter RosatdTomato (line 007905) [22], Nkx2-1Cre (line 008661) [21], and TCF/Lef:H2B-GFP mice (line 013752) [23] were purchased from Jackson Laboratories. Lef1flox/flox mice were provided by HHX [20]. All strains were maintained on a C57BL/6J background except TCF/Lef:H2B-GFP mice, which were originally on a C57BL/6 × 129 background. Male Nkx2-1Cre/Cre;Lef1flox/+ and female Lef1flox/flox;RosatdTomato/tdTomato mice were used to generate conditional knockout (Lef1CKO: Nkx2-1Cre/+;Lef1flox/flox;RosatdTomato/+) and control (Lef1CON: Nkx2-1Cre/+;Lef1flox/+; RosatdTomato/+) offspring. Females breeders were maintained by inbreeding. Male breeders were maintained by interbreeding Nkx2-1Cre/Cre;Lef1+/+ and Nkx2-1Cre/Cre;Lef1flox/+ for no more than 5 generations to avoid potential artifacts caused by Cre homozygous inbreeding [64]. In occasional litters, Ai9 reporter expression was observed throughout the body of approximately 10% of experimental animals, consistent with published literature [21]; such animals were not used for experiments. All the mice were from at least 3 litters unless otherwise noted. Sex at E14.5 was determined by genotyping by Jarid 1c [65]. When generating experimental mice for body weight measurement and behavioral tests, each litter was culled to 8 pups at P0. Genotyping for RosatdTomato and TCF/Lef:H2B-GFP animals was done according to available Jackson Laboratory protocols for these strains. Genotyping for Nkx2-1Cre mice was done using primers for Cre recombinase detection (forward primer: 5ʹ-ATGCTTCTGTCCGTTTGCCG-3ʹ, reverse primer: 5ʹ-CCTGTTTTGCACGTTCACCG-3ʹ). Genotyping for Lef1flox mice was done using primers contributed by HHX (forward primer: 5ʹ-GCAGATATAGACACTAGCACC-3ʹ, reverse primer: 5ʹ-TCCACACAACTAACGGCTAC-3ʹ).
Canton-S wild-type and pan2 mutant (BL4759) Drosophila melanogaster strains were obtained from Bloomington Stock Center.
At the sphere stage, 10–50 blastula cells from donor embryos were transplanted using a glass micropipette into the dorsal side of shield stage host embryos, 20–40 degrees from the animal pole, representing the hypothalamus anlage [66]. Embryos were then raised to 5 dpf for immunohistochemistry. Donor and host embryos were retained for genotyping to identify lef1 mutants.
Four dpf zebrafish embryos were incubated in E3 media containing 10 mM BrdU (Sigma-Aldrich, St. Louis, MO) at 28.5°C for indicated time before being washed in E3 media for at least 3 times.
Embryos and larvae were fixed in 4% paraformaldehyde (PFA) for 3 hours at room temperature (RT) or overnight (O/N) at 4°C followed by brain dissection. Brains were either dehydrated in methanol and stored at −20°C, or immediately processed for immunohistochemistry. For 3 dpf embryos, 5% sucrose was included in the fixative to ease dissection. Brains were treated with 0.5 U dispase (Gibco #17105–041) in 2% PBST (PBS/2% Triton X-100) for 60 minutes at RT. For BrdU, PCNA, pH3 or Caspase-3 staining, brains were washed in water for 5 minutes twice, followed by incubation in 2 N HCl for 60 minutes at RT, followed by 2 more water washes. Brains were then blocked in 5% to 10% goat serum in 0.5% PBST for 60 minutes at RT. Embryos were incubated in primary antibodies in block O/N at 4°C and secondary antibodies and Hoechst 33342 (Life Technologies, H3570) in block O/N at 4°C before mounting in Fluoromount-G (SouthernBiotech, Birmingham, AL) with the ventral hypothalamus facing the coverslip. Primary antibodies were all used at 1:500 dilution except as noted: chicken anti-GFP (Aves Labs, GFP-1020), rabbit anti-GFP (Molecular Probes, A11122), mouse anti-HuC/D (Molecular Probes, A21271), rabbit anti-5-HT (ImmunoStar, 541016), rabbit anti-pH3 (1:400, Cell Signaling, 9713), rabbit anti-active Caspase-3 (BD Pharmingen, 559565), rabbit anti-BLBP (Abcam, ab32432), mouse anti-PCNA (Sigma, P8825), and chicken anti-BrdU (ICL, CBDU-65A-Z). Secondary antibodies were all used at 1:500 dilution: goat anti-mouse Alexa Fluor 448 (Invitrogen, A11001), goat anti-rabbit Alexa Fluor 488 (Invitrogen, A11008), donkey anti-chicken Alexa Fluor 488 (Jackson ImmunoResearch, 703-545-155), goat anti-rabbit cy3 (Jackson ImmunoResearch, 111-165-003), goat anti-mouse cy3 (Jackson ImmunoResearch, 115-165-003), goat anti-mouse Alexa Fluor 647 (Invitrogen, A21235), goat anti-rabbit Alexa Fluor 647 (Invitrogen, A21244), and goat anti-chicken Alexa Fluor 647 (Invitrogen, A21449). Hoechst 33342 (1:10,000) was used to stain nuclei. All the primary antibodies were validated previously [4,67].
E14.5 embryo heads were dissected in PBS and fixed in 4% PFA at RT for 1.5 hours or O/N at 4°C. Brains were dissected and cryoprotected in 15% and then 30% sucrose, embedded in OCT, and stored at −80°C. Brains were cryosectioned at a thickness of 16 μm, air dried and stored at −80°C. Air-dried sections were then washed in PTW (PBS+0.1% Tween 20) 3 times, followed by permeabilization in 0.25% PBST for 5 minutes and blocking in 10% goat serum in PTW for 60 minutes. Sections were incubated in primary antibodies in blocking solution O/N at 4°C and secondary antibodies in blocking solution for 2 hours at RT, followed by Hoechst 33342 stain for 10 minutes at RT before mounting in Fluoromount-G. Antibodies used were as described above except rabbit anti-LEF1 (1:200, Cell Signaling, 2230), goat anti-PMCH (1:500, Santa Cruz, sc14509) and donkey anti-goat Alexa Fluor 647 (1:400, Invitrogen, A21447). All primary antibodies were validated by absence of staining in Lef1CKO animals. For HuC/D staining, incubation for 30 minutes in 0.5 U dispase was performed in 0.25% PBST.
Drosophila immunohistochemistry was performed as previously described [68] except that a fluorescent secondary antibody was used. Antibodies used were as described above except mouse anti-FasII (1:5, DSHB, 1D4), which was validated previously [32].
In situ hybridization probes were made by a clone-free method as described previously [69,70], with DNA templates purified using Zymo Research DNA Clean & Concentrator-5 kit. Primers were designed by Primer-BLAST [71] except for mouse genes with primer sequences available from the Allen Brain Atlas (ABA) [16] or GenePaint Atlas [26]. A full list of primers used to make probes is in S9 Table. cDNA made from 3 dpf zebrafish embryos, P2, and P60 mouse hypothalamus, and adult Drosophila (gift from C. Thummel) was used as the initial template for PCR to generate T7 promoter-containing DNA. RNA probes for zebrafish lef1 [72] and axin2 [73] were previously described. The RNA probe for Drosophila pan was generated from the Drosophila Gene Collection T7 promoter-containing cDNA GM04312 [74].
Zebrafish whole mount in situ hybridization was performed as described previously [75] except that 15 dpf and adult zebrafish were fixed in 4% PFA O/N at 4°C followed by washing in PBS and brain dissection. All tissues were treated for 30 minutes with 10 μg/ml Proteinase K. Pigmented embryos were bleached in 1% H2O2/5% formamide/0.5× SSC O/N at RT after in situ hybridization. 3 dpf embryos and postembryonic brains were imaged in 100% glycerol and PBS, respectively. For automated whole mount in situ hybridization, all steps following probe hybridization and before color reaction were performed using a BioLane HTI (Intavis, Chicago, IL).
Twenty-five μm brain cryosections were collected and post-fixed as previously described [76] (http://developingmouse.brain-map.org/docs/Overview.pdf). In situ hybridization was then performed as described [77].
Drosophila whole mount in situ hybridization was performed as described previously [68].
Zebrafish from a single home tank were anesthetized using tricaine (Sigma-Aldrich, St. Louis, MO) in shallow water. Images were acquired of immobilized, non-overlapping fish with a ruler for scale. Body length was calculated by measuring the distance between the mouth and the anterior edge of the tail fin, using ImageJ.
Five fish from lef1+/- incrosses were raised per tank starting at 5 dpf. lef1 mutants and controls were separated at 15 dpf. Novel tank diving tests [13] were performed on 16 dpf larvae during the early afternoon of the same days, before lef1 mutants start to display surfacing behavior at 20 dpf. Novel rectangular tanks (16.6 cm × 9.5 cm × 12.3 cm) were illuminated by a centered white light, and videos were acquired with a mounted Nokia Lumia 640 phone 1080p camera. For each experiment, single mutant and control larvae were netted and then removed simultaneously from their home cages and transferred to novel tanks with identical water volume. The order of netting mutant and control fish was rotated between trials. Videos were viewed in MPlayerX to manually analyze the latency of larvae to enter the upper half of the tank after initial sinking. Videos were then imported and analyzed using Ethovision XT version 11.5 (Noldus, Leesburg, VA) during the initial exploration phase, with a tracking period of 2 minutes beginning 1 minute after release into the novel tank to decrease water agitation resulting from netting. Videos were also analyzed after the initial exploration phase with a tracking period during the 4 to 6 minute interval. Tracks were analyzed for distance travelled, time in upper half of the tank and time of immobility.
All pups were weaned at P21 immediately following the first weighing. Pups weighing less than 6.5 g were excluded from analysis. All mice were weighed during the morning of the same days of the following weeks.
Group-housed mice were allowed to acclimate to the animal facility for behavioral tests 9 days after an on-campus transfer. Each mouse was handled daily for 2 minutes, during midmorning for 7 days before commencement of behavioral testing using the cupped hand method [78]. To avoid behavioral variation caused by female estrous cycle [79], a vaginal lavage procedure was done after daily handling for estrous phase evaluation for 7 days, as previously reported [80]. Female mice in their proestrus or estrus phases were collectively grouped as “Estrus” and females in their metestrus and diestrus phases were collectively grouped as “Diestrus.” All mice were acclimated to the behavior room for 1 hour under red light (69 lux) before commencement of tests. Open field and EPM behavioral tests were performed in order, once daily for 2 days, from 9 am to 5 pm. The experimenter was blinded to genotype.
Each mouse was placed in a circular plexiglass chamber (4.5” diameter × 3” height) located inside an illuminated (330 lux) circular open field arena (110 cm diameter) and allowed to acclimate for 1 minute to decrease movement bias resulting from experimenter handling. After 1 minute, the plexiglass chamber was removed from the arena, and the mouse was allowed to freely explore the arena for 10 minutes. Movement was video recorded and analyzed using Ethovision version 9 (Noldus, Leesburg, VA).
The EPM apparatus was elevated 60 cm from the floor, having 2 open arms (35 cm × 5 cm) and 2 closed arms (35 cm × 16 cm) connected by a central platform (5 cm × 5 cm). The EPM was illuminated by a white light (205 lux) at the center platform. Each mouse was placed in a rectangular opaque white plexiglass chamber (2” × 3” × 5”) located on the center platform, and allowed to acclimate for 1 minute before commencement of the test. The white chamber was removed and the mouse was allowed to freely explore the EPM for 5 minutes. Behavior was video recorded and analyzed using Ethovision version 9 (Noldus, Leesburg, VA).
Embryos were fixed for 1.5 hours in 4% PFA/5% sucrose in PBS at RT, followed by whole hypothalamus dissection with super-fine forceps (FST, 11252–00). For each biological replicate, 28 to 38 dissected hypothalami were pooled for lef1 mutant and control samples from at least 1 single-pair breeding. RNA was extracted using a RecoverAll Total Nucleic Acid Isolation Kit for FFPE (Ambion, AM1975) according to the manufacturer’s instructions. Three biological replicates were obtained on different days from offspring of different breedings. A total of 300 ng RNA per sample was submitted to the High Throughput Genomic Core at the University of Utah for RNA quality control by High Sensitivity R6K ScreenTape, RNA concentration by vacuum drying, cDNA library prep by Illumina TruSeq Stranded RNA Kit with Ribo-Zero Gold and sequencing by HiSeq 50 Cycle Single-Read Sequencing version 3. RNA-seq was analyzed by the Bioinformatics Core at the University of Utah. A transcriptome reference was created by combining GRCz10 chromosome sequences with Ensembl build 84 splice junction sequences generated with USeq (version 8.8.8) MakeTranscriptome. RNA-seq reads were mapped to the GRCz10 zebrafish transcriptome reference using Novoalign (version 2.08.03). Splice junction alignments were converted back to genomic space using USeq SamTranscriptomeParser. USeq DefinedRegionDifferentialSeq was used to generate per gene read counts, which were used in DESeq2 to determine differential expression. RNA-seq graph in Fig 2A was made by IPython Notebook with package NetworkX.
E14.5 and P22 nonweaned male Lef1CON and Lef1CKO hypothalami were dissected using a fluorescent microscope in ice-cold PBS, while tail tissue was retained for genotyping. E14.5 tissues were immediately immersed in RNAlater (Thermo Fisher, Waltham, MA) and stored at 4°C for up to 7 days until RNA extraction. P22 tissues dissected from at least 2 litters were immediately homogenized in TRIzol (Thermo Fisher, Waltham, MA) and stored at −80°C. Three biological replicates were prepared from either 5 pooled hypothalami (E14.5) or a single hypothalamus (P22) from Lef1CON and Lef1CKO mice, and RNA was extracted on the same day using TRIzol followed by purification with an RNeasy Mini Kit (Qiagen, Hilden, Germany) and on-column DNase digestion (Sigma-Aldrich, St. Louis, MO). One μg of RNA per sample was submitted to the High Throughput Genomic Core at the University of Utah for RNA quality control with Agilent RNA ScreenTape, cDNA library prep with Illumina TruSeq Stranded RNA Kit with Ribo-Zero Gold, and sequencing using HiSeq 50 Cycle Single-Read Sequencing version 4. RNA-seq reads were mapped to GRCm38. Differential gene expression analysis and graph plotting were carried out using the same methods as for zebrafish RNA-seq.
Three biological replicates of RNA from male and female mice were prepared as described above for RNA-seq. Two and a half μg RNA was used for cDNA synthesis with a SuperScript III Reverse Transcriptase kit (Invitrogen, Carlsbad, CA). qPCR was performed in triplicate using Platinum SYBR Green master mix (Invitrogen, Carlsbad, CA) on 96-well CFX Connect (Bio-Rad, Hercules, CA) plates or 384-well QuantStudio 12K Flex (Life Technologies, Durham, NC) plates at the Genomics Core at the University of Utah, according to manufacturer’s instructions. Gapdh was used to normalize quantification, and reverse transcriptase was omitted for controls. qPCR analysis was performed with the ΔΔCt method to determine relative expression change [81]. Dissociation curve analysis was performed to confirm the specificity of amplicons. qPCR primers were designed from PrimerBank [82] as follows (forward primer first, reverse primer second, in 5ʹ to 3ʹ orientation with PrimerBank ID in the parentheses), Pmch (12861395a1): GTCTGGCTGTAAAACCTTACCTC, CCTGAGCATGTCAAAATCTCTCC; Tacr3 (10946720a1): CTGGGCTTGCCAGTGACAT, CGCTTGTGGGCCAAGATGAT; Crhbp (162287189c2): CTTACCCTCGGACACTTGCAT, GGTCTGCTAAGGGCATCATCT.
Fluorescent images of dissected zebrafish and mouse brains were obtained with an Olympus FV1000 confocal microscope at the Cell Imaging Core at the University of Utah. Z-stack images were all maximum intensity z-projections of 3 μm slices; single- or double-labeled cells were manually counted in FV1000 ASW 4.2 Viewer. All the zebrafish and mouse in situ hybridization images were obtained with an Olympus SZX16 dissecting microscope except those in Fig 5E, S2C Fig and S6B Fig, which were obtained with an Olympus BX51WI compound microscope. Two months post-fertilization (mpf) zebrafish images (S3A and S3B Fig) were acquired using a Leica MZ16 microscope. Drosophila in situ hybridization images were obtained with a Zeiss Axioskop.
IPA (QIAGEN, Redwood City, CA) was performed with 129 mouse orthologs of the 138 zebrafish protein-coding genes identified from RNA-seq with AdjP <0.1 (S4 Table). Analysis was performed by the Bioinformatics Core at the University of Utah according to QIAGEN's instructions and “diseases and functions” were extracted from the software (S3 Table).
Publically available GTEx raw datasets were downloaded from www.gtexportal.org in April 2017 as a single file: GTEx_Analysis_v6p_RNA-seq_RNA-SeQCv1.1.8_gene_rpkm.gct.gz. Ninety-six hypothalamic samples were identified according to their specific strong PMCH expression, and extracted into S7 Table by IPython Notebook with packages gzip and xlwt. Pearson correlation was calculated by gene reads per kilobase of transcript per million mapped reads (RPKM) using IPython Notebook with function scipy.stats.stats.pearsonr, followed by result writing into S8 Table by IPython Notebook with package xlwt. The same Pearson correlation r values were confirmed using Excel’s CORREL function. A similar correlation result was obtained when searching for the top 200 correlated genes by Pearson on GeneNetwork (www.genenetwork.org) in April 2017. Several differences are noted between our analyses and GeneNetwork’s analyses. First, GeneNetwork imported an older version of GTEx’s datasets (GTEXv5 Human Brain Hypothalamus RefSeq [Sep15] RPKM log2). Second, GeneNetwork calculated Pearson correlation using RPKM log2 rather than RPKM in our case. Third, GeneNetwork calculated Pearson’s sample correlation across a population, with an adjustment across the genome, and also based on the number of the top correlated genes requested by the users; in our case, we calculated Pearson correlation between 2 genes, and simply ranked all the genes by their Pearson’s r values calculated for the gene of interest. Lastly, GeneNetwork’s imported older GTEx datasets had 102 hypothalamic samples, 6 among which were left out in current GTEx’s server. The complete overlapping of the 96 samples further confirmed our successful extraction of hypothalamic datasets from the GTEx project.
No statistical methods were used to predetermine sample size. For behavioral assays, sample size was determined based on accepted practice. The experiments were not randomized. Due to visible phenotypes, the investigators were not blinded to outcome assessment except for whole mount in situ hybridization of zebrafish lef1+/- incrosses, Drosophila pan+/- incrosses, and mouse body weight and behavioral assays. Two-tailed unpaired Student t tests were performed for all statistical analysis, except mouse body weight (2-way ANOVA with repeated measures), using GraphPad Prism software version 6. Outliers were identified by Grubbs’ test for behavioral assays with significance assigned at P < 0.05 (alpha = 0.01). All the criteria for excluding data points were established prior to data collection.
|
10.1371/journal.pcbi.1000471 | Temporal Variability and Social Heterogeneity in Disease Transmission: The Case of SARS in Hong Kong | The extent to which self-adopted or intervention-related changes in behaviors affect the course of epidemics remains a key issue for outbreak control. This study attempted to quantify the effect of such changes on the risk of infection in different settings, i.e., the community and hospitals. The 2002–2003 severe acute respiratory syndrome (SARS) outbreak in Hong Kong, where 27% of cases were healthcare workers, was used as an example. A stochastic compartmental SEIR (susceptible-exposed-infectious-removed) model was used: the population was split into healthcare workers, hospitalized people and general population. Super spreading events (SSEs) were taken into account in the model. The temporal evolutions of the daily effective contact rates in the community and hospitals were modeled with smooth functions. Data augmentation techniques and Markov chain Monte Carlo (MCMC) methods were applied to estimate SARS epidemiological parameters. In particular, estimates of daily reproduction numbers were provided for each subpopulation. The average duration of the SARS infectious period was estimated to be 9.3 days (±0.3 days). The model was able to disentangle the impact of the two SSEs from background transmission rates. The effective contact rates, which were estimated on a daily basis, decreased with time, reaching zero inside hospitals. This observation suggests that public health measures and possible changes in individual behaviors effectively reduced transmission, especially in hospitals. The temporal patterns of reproduction numbers were similar for healthcare workers and the general population, indicating that on average, an infectious healthcare worker did not infect more people than any other infectious person. We provide a general method to estimate time dependence of parameters in structured epidemic models, which enables investigation of the impact of control measures and behavioral changes in different settings.
| Recent epidemics have shown that healthcare workers may be overrepresented among cases and how critical it is to protect them. For example, during the 2002–2003 severe acute respiratory syndrome (SARS) epidemics in Hong Kong, 27%of cases were healthcare workers when they were <1% of the population. Better means of protection require understanding how healthcare workers were infected and assessing their role in disease transmission. Here, we describe a method for estimating the temporal profile of the risk of infection and probability of transmission in the community and hospitals. The 2002–2003 SARS outbreak in Hong Kong is used as an example. For the SARS epidemic, we show that the risk of infection in the community and hospitals decreased with time down to zero in hospitals but remained larger in the community. This observation suggests that public health measures and behavioural changes most effectively reduced transmission in hospitals. Besides, we find that the large number of cases observed among healthcare workers is more likely a result of large and sustained exposure to hospitalized cases than to transmission among healthcare workers. These results are of interest to design control measures in the event of an influenza pandemic.
| Emerging infectious diseases have been defined as, “infections that have newly appeared in a population or have existed previously but are rapidly increasing in incidence or geographic range. [1]” Several features may make them particularly threatening. First, recognizing the disease can be difficult when the first cases appear, especially when the symptoms are non-specific. Second, no vaccine or specific treatment may be known initially. Moreover, heterogeneities in disease transmission may create high-risk groups, such as healthcare workers [2]–[5] and high-risk geographical areas, thereby dramatically enhancing the impact of the outbreak [6].
The 2003 severe acute respiratory syndrome (SARS) outbreak in Hong Kong is remarkably illustrative of the above issues: symptoms were similar to pneumonia [7]; the incubation period was long enough for local and international transmission to occur [8]; no vaccine or treatment was available; as much as 21% of cases worldwide were healthcare workers [9]. The outbreak also demonstrated the possible existence of super-spreading events (SSEs) [10], during which a few infectious individuals contaminated a high number of secondary cases. Hong Kong had two SSEs: the first occurred in Hospital X around March 3 and led to about 125 cases [11]; the second occurred in Housing Estate Y on March 19, and led to over 300 cases [12],[13]. Despite its particularly threatening features, the outbreak was brought under control.
In this context, once the epidemic is detected, spontaneous changes in behavior will occur, and non-pharmacological measures are usually initiated to control the outbreak. The resulting effects of these two phenomena on disease transmission is not easily quantified.
The effective contact rate, which reflects the combined influences of social proximity (the number of contacts per time unit) and the probability of infection through each contact, is an essential determinant of disease spread. Our aim was to estimate the temporal variation of this parameter in the community and hospitals, over the course of the outbreak.
Previously published mathematical models of parameter estimation addressed the issues of temporal variability [12],[14] or social heterogeneity [2],[15]. Here we present an approach that deals with both issues, together with the occurrence of SSEs. Then the method is applied to the 2003 SARS epidemic in Hong Kong (SARSID database [13]).
Among the 1755 patients admitted to Hong Kong hospitals in 2003 for suspected SARS, 1467 serologically confirmed SARS cases were retained for analysis. For each case, occupation, date of symptom onset, date of hospital admission, duration of hospital stay and discharge status (dead or alive) were recorded. Durations of hospital stay were missing for 12 cases and imputed to 100 days.
The epidemic process was cast into a discrete time stochastic susceptible-exposed-infectious-removed (SEIR) compartmental model, designed to reflect a two-way classification of individuals according to disease status and ‘social’ category (Figure 1).
The latter was defined in three categories: hospitalized patients (hp), healthcare workers (hw), and the general population (gp).
According to these three social categories, SARS cases were qualified: nosocomial when the patient had been hospitalized for ≥5 days before symptom onset ; healthcare workers when the subjects were indeed healthcare workers and not nosocomial ; or general population, all others . Their corresponding epidemic curves are shown in Figure 2.
Disease status was described in five compartments: susceptible (S), exposed (E), infectious non-hospitalized (I), infectious hospitalized (H), and removed (R). Individuals are initially susceptible to the disease and infected through contact with infectious subjects. Once infected, individuals are first exposed (infected, non-infectious) and then become infectious. The infectious stage is defined as the period of time during which infectious individuals can transmit the disease through contact with susceptibles. Finally, the infectious individuals are removed, either through recovery or death. Quarantine or isolation was not documented in the database, and was not specifically described: possibly isolated infectious individuals remain in stage or , and quarantined contacts remain in stage .
Thus, depending on social category, susceptible individuals may be in compartments (general population), (healthcare workers), or (hospitalized patients); similarly, exposed and recovered individuals may be in compartments , or , and , or , respectively; while infectious subjects are in compartments or before hospitalization, and in compartments , or once hospitalized.
The size of the Hong Kong population () was obtained from local census data (http://www.info.gov.hk/info/hkbrief/eng/living2.htm). The number of hospitalized patients () equaled the number of hospital beds in Hong Kong (http://www.info.gov.hk/info/hkbrief/eng/living2.htm). The number of healthcare workers () was derived from the healthcare worker-to-bed ratio in the Hospital X [13]. , and were assumed to be constant throughout the epidemic. Under this steady-state assumption, transitions between compartments , , and did not have to be included explicitly in the model.
The model assumes that there is no direct contact between hospitalized individuals and non hospitalized individuals from the general population. In particular, susceptible individuals in the general population ( compartment) cannot be infected by infectious hospitalized SARS cases (, , and compartments), and susceptible hospitalized patients ( compartment) cannot be infected by infectious not-yet-hospitalized cases from the general population ( compartment).
In the following, 1{.} denotes the indicator function, defined by if is true, and 0 otherwise.
For each Hong Kong inhabitant , let be the time of symptom onset, the day of hospital admission, the day of hospital discharge, the day of death ( if the case did not die from SARS), and the social category ( if , if and inhabitant is a healthcare worker, and otherwise). For all inhabitants who were not infected by SARS, we let .
For each individual , let . The observed data were augmented with , where , , and correspond to the dates of transition into the , , and states respectively ; is the date of death, and is the social category for case (, or ).
Letting and , the joint density of , , and of the vector of unknown parameters is written as the following product:where , and is a prior distribution for .
As defined by Auranen et al. [17], , and refer to the observation level, the transmission level and the prior level respectively.
The observation level ensures that the observed data are consistent with the augmented data.
During the SARS outbreak, few cases were reportedly infected by asymptomatic persons, but cases rapidly became infectious after symptom onset [12],[18],[19]. Therefore, for each case , the onset of symptom was considered acceptable if .
The day of hospital admission was consistent with the augmented data if when the case was infectious prior to hospitalization () when the case was infectious only after hospitalization () when the case was not infectious anymore at the time of hospitalization ().
It was also assumed that the infectious period did not outlast hospital discharge, that is
The date of death was .
Finally, the professional category was acceptable if .
Hence:
The transmission level describes SARS transmission, assuming and are known, conditional on the day of infection of the first case.
A deterministic latent period of 5 days was assumed for all cases ( for such that ) [13].
The duration of the infectious period () for SARS cases was gamma-distributed, with mean and variance . We let and denote its density and cumulative distribution function respectively. For SARS patients dead on discharge, the infectious period was considered censored by death. Since the infectious period was defined as the period during which infectious cases can transmit the disease through contact with susceptibles, its distribution was assumed to remain the same over the course of the epidemic.
The specific stochastic infection rates on day for susceptible individuals in compartments , , and are: , and , where and denote the numbers of individuals in compartments and , respectively; and are the daily effective contact rates in the community and hospitals, respectively; and are temporary level shift interventions [4] reflecting the increment of infectiousness during the Hospital X and Housing Estate Y SSEs, i.e. from days and to days and .
This leads to:where ; (), and () is the incidence in on day .
The vector comprised about 228 unknown parameters, the epidemic lasting about days.
For all model parameters except the effective contact rates, independent prior distributions were chosen. For the time of start of SSEs, the prior distributions were informative (see Table 1). The effective contact rates and were modeled as second-order Gaussian random walks, on the log scale, with flat exponential priors on the first two states of the random walk. In this approach, the respective variances and of innovations correspond to the smoothing parameters of cubic smoothing splines [21]; smaller values of and are associated with smoother trajectories. For the two precision parameters and , exponential hyperpriors with mean were selected. A sensitivity analysis of the hyperparameter value was performed (see Text S1).
A Markov chain Monte Carlo (MCMC) method was used to sample the joint posterior distribution [22],[23]. More details on the sampler are provided in Text S2.
From the joint posterior distribution of the parameters, a number of meaningful epidemiological quantities, such as daily case-reproduction numbers [24] in each category (see Text S3), could be derived. In particular, the number of cases generated by each SSE could be estimated.
Estimates of the days of SSE starts and ends, increments (, ), and the number of SSE cases in Hospital X and Housing Estate Y are shown in Table 2. Despite the somewhat shorter SSE duration for Housing Estate Y than for Hospital X, 2.5 times more cases occurred in Housing Estate Y than Hospital X.
The estimated mean of the infectious period was 9.3 days (95% credible interval: (8.6–9.9)), with an estimated standard deviation of 2.3 days (95% credible interval: (1.8–2.9)). The proportion of the infectious period spent in the community decreased continuously with time (>60% at the beginning, <20% as early as early April). Toward the end of the epidemic, >95% of the infectious period was spent inside hospitals (see Figure 3).
The daily effective contact rates in the community () and hospitals () exhibited progressive a decrease in time, as shown in Figure 4. However, while the contact rate was almost 0 by late March inside hospitals, it remained >0.17 in the community.
The case-reproduction number first increased to 5.1 in the general population in late February and to 3.0 for healthcare workers in early March (see Text S3). It then decreased until the end of the epidemic. The case-reproduction numbers was <1 on March 13 for healthcare workers and on March 20 for the general population. Among nosocomial cases, the case-reproduction number was always <1, with a maximum value of 0.2 on March 14.
The model's ability to reproduce the main features of the epidemic was checked by simulating 5000 epidemics with parameters sampled from the estimated joint posterior distribution, as described in Text S4. The size and duration of simulated epidemics, as well as cases breakdown in categories (, , ) mirrored the Hong Kong epidemic (see Figure 2).
We also simulated 100 epidemics with a single set of parameters, sampled from the posterior distribution. Then, the estimation procedure was applied to each simulated epidemic in order to reestimate the parameters. The original parameters were in the estimated corresponding 95% credible interval in 87% of cases.
To rapidly and economically design and assess control measures for epidemics in modern societies, added insight into the dynamics of disease transmission is needed. These dynamics are conveniently summarized by critical, albeit non-observable, characteristics, such as the duration of the infectious period and effective contact rates. Estimation of these parameters from the observed data requires the development of mathematical models. Herein, we presented a model for epidemics that provides for social heterogeneity and time variability of transmission parameters. As a working example, the model was applied to the 2003 SARS epidemic in Hong Kong.
The effect of interventions and/or changes in behavior during the 2003 SARS outbreak may be modelled as time varying contact rates [12],[15],[25] or involve shortening of the infectious period [19]. Here, we adopted the first view. To assess if the data supported this choice, a model was fit where, in addition to time varying contact rates, we allowed the mean infectious period to change over three consecutive periods. The three posterior means were 9.5 days (before March 20), 9.2 (March 21 to April 9) and 10 days (after April 10), indicating that the time varying contact rates alone model the data adequately.
While the duration of the infectious period is an obvious determinant of disease transmission, no estimate has been available for SARS. The distribution of the viral load was found to peak 8–10 days after symptom onset [13],[18],[26],[27]. Here, assuming that the infectious period started between 1 day before and 4 days after symptom onset, it was estimated to extend over an average period of 9.3 days. We also found that the proportion of time infectious people spent outside hospitals decreased during the outbreak and was <5% at the very end, in agreement with Anderson et al. [18] and Leung et al. [13] who showed that the time from symptom onset to admission was shorter at the end of the epidemic.
One of the most striking features of the Hong Kong SARS epidemic was the occurrence of two SSEs. By definition, SSEs correspond to exceptional circumstances that are usually limited to well-circumscribed areas, such as Hospital X and Housing Estate Y, and last for only a few days [10]. In this respect, the very high contact rates generated by the SSEs were modeled as ‘innovation outliers’ [28], to avoid spurious overestimation of contact rates among the Hong Kong population.
Whether SSEs are a result of a few particularly highly infectious cases (excreting much virus and/or highly connected socially), or of particular environmental circumstances, or maybe both, remains unclear [16],[29],[30]. In our model, the force of infection associated with each SSE was independent on the number of currently infectious cases. The duration of SSEs was estimated independently for each SSE, and was independent on the duration of the infectious period. Therefore, our model was consistent with all possible causes of SSEs: one or several super-spreaders, or particular environmental circumstances, etc.
The level shift interventions [20] that were superimposed on the process describing the time evolution of the infection rates differed significantly from zero. Taking into account only serologically confirmed cases, we estimated that the Hospital X SSE began on March 1st, lasted 11 days and was responsible for 94 cases; and that the Housing Estate Y SSE began on March 18, lasted 6 days and caused 235 cases. Previous studies investigating SSEs in Hong Kong used all cases. By contact tracing, Lee et al. [11] found that the Hospital X SSE started on March 4 and involved 125 cases; the Housing Estate Y SSE had been estimated to start on March 19 [13] and to involve 312–330 [13] or 331 [12] cases.
Effective contact rates were estimated on a daily basis, in the community and hospitals. Both rates tended to decline, probably reflecting the effect of control measures (listed in Figure 4 [31],[32]) or self-adopted behavioral changes. The measures seem to have been particularly effective in hospitals, where the effective contact rate was 0 by late March, whereas the risk in the community did not decrease as sharply. In both settings, the effective contact rate was almost constant after late March, when no more control measures were introduced.
Others who studied the dependence of disease transmission on time reported reproduction numbers rather than effective contact rates [12],[14]. While the daily effective contact rates are sensitive to short-term day-to-day variations in transmission, the reproduction numbers reflect the integrated influences of the temporal evolution of effective contact rates, the infectious period duration and other factors, such as time spent in the community before hospitalization. Here, estimates of daily reproduction numbers were obtained for each social category. Notably, unlike Cauchemez et al. [14], it was not necessary to assume prior knowledge or constancy of the generation interval. The reproduction numbers showed a trend similar to the effective contact rates, with a clearly decreasing trend over time, suggesting that the epidemic was under control as early as mid-March (see Figure in Text S3). Moreover, the temporal patterns for the general population and healthcare workers were similar, with the reproduction numbers being higher for the general population, thereby indicating that on average, an infectious healthcare worker did not infect more people than any other infectious person. The reproduction numbers for nosocomial cases were much lower, either because they had fewer contacts or because the people they were in contact with were protected (typically healthcare workers wearing masks).
Our estimation procedure, applied to a set of 100 simulated epidemics, showed that in 87% of cases, the parameters used for simulation were inside the corresponding posterior 95% credible intervals. While most parameters were well estimated, the procedure tended to overestimate the duration of each SSE, while simultaneously underestimating its strength. The number of people affected by each SSE (i.e. population×duration×strength) was therefore correct, but its extent in time less robust. Ignoring the 17 days corresponding to both SSEs, 98% of the remaining parameters used for simulation were inside the posterior corresponding 95% credible intervals, indicating very little bias in our estimation procedure.
Herein, we described an approach to estimate the role of time variability and social heterogeneity in epidemic dynamics. Our model's simplifying assumptions such as the fixed duration of the latency period or the constant probability of transmission throughout the infectious period of cases, can be relaxed at the price of increasing complexity. Similarly, a more detailed model taking into account household transmission, and transmission inside and between hospitals, rather than assuming homogeneous mixing in the community and in hospitals, could be implemented, at the cost of a dramatic increase in the number of model parameters. More generally, the model can be easily accommodated to fit the specificities of any transmissible disease.
|
10.1371/journal.pcbi.1003941 | Dynamics from Seconds to Hours in Hodgkin-Huxley Model with Time-Dependent Ion Concentrations and Buffer Reservoirs | The classical Hodgkin-Huxley (HH) model neglects the time-dependence of ion concentrations in spiking dynamics. The dynamics is therefore limited to a time scale of milliseconds, which is determined by the membrane capacitance multiplied by the resistance of the ion channels, and by the gating time constants. We study slow dynamics in an extended HH framework that includes time-dependent ion concentrations, pumps, and buffers. Fluxes across the neuronal membrane change intra- and extracellular ion concentrations, whereby the latter can also change through contact to reservoirs in the surroundings. Ion gain and loss of the system is identified as a bifurcation parameter whose essential importance was not realized in earlier studies. Our systematic study of the bifurcation structure and thus the phase space structure helps to understand activation and inhibition of a new excitability in ion homeostasis which emerges in such extended models. Also modulatory mechanisms that regulate the spiking rate can be explained by bifurcations. The dynamics on three distinct slow times scales is determined by the cell volume-to-surface-area ratio and the membrane permeability (seconds), the buffer time constants (tens of seconds), and the slower backward buffering (minutes to hours). The modulatory dynamics and the newly emerging excitable dynamics corresponds to pathological conditions observed in epileptiform burst activity, and spreading depression in migraine aura and stroke, respectively.
| The classical theory by Hodgkin and Huxley (HH) describes nerve impulses (spikes) that manifest communication between nerve cells. The underlying mechanism of a single spike is excitability, i.e., a small disturbance triggers a large excursion that reverts without further input to the original state. A spike lasts a 1/1000 second and, even though during this period ions are exchanged across the nerve cell membrane, the change in the corresponding ion concentrations can become significant only in series of such spikes. Under certain pathological conditions, changes in ion concentrations become massive and last minutes to hours before they recover. This establishes a new type of excitability underlying communication failure between nerve cells during migraine and stroke. To clarify this mechanism and to recognize the relevant factors that determine the slow time scales of ion changes, we use an extended version of the classical HH theory. We identify one variable of particular importance, the potassium ion gain or loss through some reservoirs provided by the nerve cell surroundings. We suggest to describe the new excitability as a sequence of two fast processes with constant total ion content separated by two slow processes of ion clearance (loss) and re-uptake (gain).
| In this paper we study ion dynamics in ion-based neuron models. In comparison to classical HH type membrane models this introduces dynamics on much slower time scales. While spiking activity is in the order of milliseconds, the time scales of ion dynamics range from seconds to minutes and even hours depending on the process (transmembrane fluxes, glial buffering, backward buffering). The slow dynamics leads to new phenomena. Slow burst modulation as in seizure-like activity (SLA) emerges from moderate changes in the ion concentrations. Phase space excursions with large changes in the ionic variables establish a new type of ionic excitability as observed in cortical spreading depression (SD) during stroke and in migraine with aura [1], [2]. Such newly emerging dynamics can be understood from the phase space structure of the ion-based models.
Mathematical models of neural ion dynamics can be divided into two classes. On the one hand the discovery of SD by Leão in 1944 [3]—a severe perturbation of neural ion homeostasis associated with huge changes in the potassium, sodium and chloride ion concentrations in the extracellular space (ECS) [4] that spreads through the tissue—has attracted many modelling approaches dealing with the propagation of large ion concentration variations in tissue. In 1963 Grafstein described spatial potassium dynamics during SD in a reaction-diffusion framework with a phenomenological cubic rate function for the local potassium release by the neurons [5]. Reshodko and Burés proposed an even simpler cellular automata model for SD propagation [6]. In 1978 Tuckwell and Miura developed a SD model that is amenable to a more direct interpretation in terms of biophysical quantities [7]. It contains ion movements across the neural membrane and ion diffusion in the ECS. In more recent studies Dahlem et al. suggested certain refinements of the spatial coupling mechanisms, e.g., the inclusion of nonlocal and time-delayed feedback terms to explain very specific patterns of SD propagation in pathological situations like migraine with aura and stroke [8], [9].
On the other hand single cell ion dynamics were studied in HH-like membrane models that were extended to include ion changes in the intracellular space (ICS) and the ECS since the 1980s. While the first extensions of this type were developed for cardiac cells by DiFranceso and Noble [10], [11], the first cortical model in this spirit was developed by Kager, Wadman and Somjen (KWS) [12] only in 2000. Their model contains abundant physiological detail in terms of morphology and ion channels, and was in fact designed for seizure-like activity (SLA) and local SD dynamics. It succeeded spectacularly in reproducing the experimentally known phenomenology. An even more detailed model was proposed by Shapiro at the same time [13] who—like Yao, Huang and Miura for KWS [14]—also investigated SD propagation with a spatial continuum ansatz. Another model of SD investigated Ca transmission along an astrocyte lane [15], where glutamate released from neurons that acts on metabotropic receptors of astrocytes determines the characteristics.
HH-like models of intermediate complexity were developed by Fröhlich, Bazhenov et al. to describe potassium dynamics during epileptiform bursting [16]–[18]. The simplest HH-like model of cortical ion dynamics was developed by Barreto, Cressman et al. [19]–[21] who describe the effect of ion dynamics in epileptiform bursting modulation in a single compartment model that is based on the classical HH ion channels. Interestingly, in none of these models, which are considerably simpler than, for example, Shapiro's model and the KWS model, extreme ion dynamics like in SD or stroke was studied. To our knowledge the only exception is a study by Zandt et al. who describe in the framework of Cressman et al. what they call the “wave of death” that follows the anoxic depolarization after decapitation as measured in experiments with rats [22].
In this study we systematically analyze the entire phase space of such local ion-based neuron models containing the full dynamical repertoire ranging from fast action potentials to slow changes in ion concentrations. We start with the simplest possible model for SD dynamics—a variation of the Barreto, Cressman et al. model—and reproduce most of the results for the KWS model. Our analysis covers SLA and SD.
Three situations should be distinguished: isolated, closed, and open systems, which is reminiscent of a thermodynamic viewpoint (see Fig. 1). An isolated system without transfer of metabolic energy for the ATPase-driven pumps will attain its thermodynamic equilibrium, i.e., its Donnan equilibrium. A closed neuron system with functioning pumps but without ion regulation by glial cells or the vascular system is generally bistable [23]. There is a stable state of free energy-starvation (FES) that is close to the Donnan equilibrium and coexists with the physiological resting state. The ion pumps cannot recover the physiological resting state from FES.
We will now develop a novel phase space perspective on the dynamics in open neuron systems. We describe the first slow-fast decomposition of local SD dynamics, in which the ion gain and loss through external reservoirs is identified as the crucial quantity whose essential importance was not realized in earlier studies. Treating this slow variable as a parameter allows us to derive thresholds for SD ignition and the abrupt, subsequent repolarization of the membrane in a bifurcation analysis for the first time. Moreover we analyze oscillatory dynamics in open systems and thereby relate SLA and SD to different so-called torus bifurcations. This categorizes SLA and SD as genuinely different though they are ‘sibling’ dynamics as they both bifurcate from the same ‘parent’ limit cycle in a supercritical and subcritical manner, respectively, which also explains the all-or-none nature of SD. In contrast, SLA is gradual.
Local ion dynamics of neurons has been studied in models of various complexity. Reduced model types consist of an electrically excitable membrane containing gated ion channels and ion concentrations in an intra- and an extracellular compartment [19]–[22]. Transmembrane currents must be converted to ion fluxes that lead to changes in the compartmental ion concentrations. Such an extension requires ion pumps to prevent the differences between ICS and ECS ion concentrations that are present under physiological resting conditions from depleting.
We consider a model containing sodium, potassium and chloride ions. The simulation code is available from ModelDB [24], the accession number is 167714. The HH-like membrane dynamics is described by the membrane potential and the potassium activation variable . The sodium activation is approximated adiabatically and the sodium inactivation follows from an assumed functional relation between and . The ICS and ECS concentrations of sodium, potassium and chloride ions are denoted by , and , respectively.
In a closed system mass conservation holds, i.e., (1)with and the ICS/ECS volumes . Together with the electroneutrality of ion fluxes across the membrane, i.e., (2)only two of the six ion concentrations are independent dynamical variables. The full list of rate equations then reads (3)(4)(5)(6)
They are complemented by six constraints on gating variables and ion concentrations: (7)(8)(9)(10)(11)(12)
Superscript 0 indicates ion concentrations in the physiological resting state. Unless otherwise stated and are used as initial conditions in the simulations. Constrained ion concentrations (Eqs. (7)–(10)) then also take their physiological resting state values. These ion concentrations, the membrane capacitance , the gating time scale parameter , the conversion factor from currents to ion fluxes, and the ICS and ECS volumes are listed in Tab. 1. The conversion factor is an expression of the membrane surface area and Faraday's constant (both given in Tab. 1, too): (13)We remark that all parameters in Tab. 1 are given in typical units of the respective quantities. The numerical values in these units can directly be used for simulations. Time is then given in msec, the membrane potential in mV and ion concentrations in .
The electroneutrality of the total transmembrane ion flux as expressed in Eqs. (2) and (7) is a consequence of the large time scale separation between the membrane dynamics and the ion dynamics (cf. Ref. [23] and the below discussion of time scales). This constraint is the reason why the thermodynamic equilibrium of the system must be understood as a Donnan equilibrium. This is the electrochemical equilibrium of a system with a membrane that is impermeable to some charged particles, which can be reached in an electroneutral fashion, i.e., without separating charges. We do not include this impermeant matter explicitly, because it does not influence the dynamics as long as osmosis is not considered. One should however keep in mind that the initial ion concentrations in Tab. 1 do not imply zero charge in the ICS or ECS and hence impermeant matter to compensate for this must be present.
The gating functions , and are given by (14)(15)(16)Here and are the asymptotic values and is potassium activation time scale. They are expressed in terms of the Hodgkin-Huxley exponential functions [19]–[21] (17)(18)(19)(20)
The three ion currents are (21)(22)(23)They are given in terms of the leak and gated conductances (with ) and the Nernst potentials which are computed from the (dynamical) ion concentrations : (24) denotes the valence of the particular ion species.
The pump current modelling the ATPase-driven exchange of intracellular sodium with extracellular potassium at a -ratio is given by (25)where is the maximal pump rate [21]. The pump current increases with and . The values for the conductances and pump rate are also given in Tab. 1. Let us remark that in comparisons with Ref. [23], we have mildly increased the maximal pump rate and decreased the chloride conductance to obtain a SD threshold in agreement with experiments (see Sect. Results).
Eqs. (3)–(12) describe a closed system in which ion pumps are the only mechanism maintaining ion homeostasis and in which mass conservation holds for each ion species. A remark on terminology is due at this point: a ‘closed’ system refers exclusively to the conservation of the ion species that we model. We do not directly model other mass transfer that occurs in real neural systems. Yet it is indirectly included. The ion pumps use energy released by hydrolysis of ATP, a molecule whose components (glucose and oxygen or lactate) therefore have to pass the system boundaries. In thermodynamics, it is customary to call systems that exchange energy but not matter with their environment closed. Since ATP is in this framework only considered as an energy source, we can describe the system as closed, if ions cannot be transferred across its boundaries.
As mentioned above the closed system is bistable. Superthreshold stimulations cause a transition from physiological resting conditions to FES. To resolve this and change the behaviour to local SD dynamics it is necessary to include further regulation mechanisms [23]. Since SD is in particular characterized by an extreme elevation of potassium in the ECS we will only discuss potassium regulation.
If ECS potassium ions are subject to a regulation mechanism which is independent of the membrane dynamics, then the symmetry between ICS and ECS potassium dynamics is broken and Eq. (9) for the potassium conservation does not hold. Let us represent changes of the potassium content of the system by a variable which is defined by the following relation: (26)
Changes of the potassium content, i.e., changes of , can be of different physiological origin. If glial buffering is at work the potassium content will be reduced by the amount of buffered potassium . An initial potassium elevation simply leads to an accordingly increased : (27)For the coupling to an extracellular potassium bath or to the vasculature is a measure for the amount of potassium that has diffused into (positive ) or out of (negative ) the system.
We are going to discuss two regulation schemes—coupling to an extracellular bath and glial buffering. They could be implemented simultaneously, but for our purpose it will suffice to apply only one scheme at a time. In the second subsection of Sect. Results, the dynamics of is given by glial buffering, while in the third subsection we will discuss the oscillatory regimes one finds for bath coupling with elevated bath concentrations. To implement glial buffering we assume a phenomenological chemical reaction of the following type [12], [25]: (28)The buffer concentration is denoted by . We are using the buffer model from Ref. [12] in which the potassium-dependent buffering rate is given as (29)The parameter is normally assumed to have the same numerical value as the constant backward buffering rate which is hence an overall parameter for the buffering strength. However, the parameters should be denoted differently as they have different units (cf. Tab. 1). This chemical reaction scheme together with the mass conservation constraint (30)where is the initial buffer concentration, leads to the following differential equation for : (31)Eq. (27) the implies the following rate equation for (32)where and are given by Eqs. (27) and (26), respectively.
To model the coupling to a potassium bath one normally includes an explicit rate equation for the ECS potassium concentration (33)where the diffusive coupling flux (34)is defined by its coupling strength and the potassium bath concentration . Eq. (26) implies that Eq. (33) can be rewritten in terms of as follows: (35)Note that we have chosen to formulate ion regulation in terms of rather than which would be completely equivalent. This is crucial, because the dynamics of happens on a time scale that is only defined by the buffering or the diffusive process, while dynamics involves transmembrane fluxes and reservoir coupling dynamics at different time scales (cf. the last paragraph of this section). This can be seen from Eq. (33).
Both regulation schemes—glial buffering given by Eq. (32) and coupling to a bath with a physiological bath concentration as in Eq. (35)—can be used to change the system dynamics from bistable to ionically excitable, i.e., excitable with large excursions in the ionic variables. Like all other system parameters the regulation parameters and are given in Tab. 1. They have been adjusted so that the duration of the depolarized phase is in agreement with experimental data on spreading depression.
Note that the parameters we have chosen are up to almost one order of magnitude lower than intact brain values like the ones used in Refs. [12], [25]–[27]. While this does not affect the general time scale separation between glial or vascular ion regulation and ion fluxes across the cellular membrane, the duration of SD depends crucially on these parameters. However, during SD oxygen deprivation will weaken glial buffering, and the swelling of glial cells and blood vessel constriction will restrict diffusion to the vasculature. Such processes can be included to ion-based neuron models and make ion regulation during SD much slower [12], [25]–[27].
For our purpose it is however sufficient to assume smaller values from the beginning. We remark that the ion regulation schemes in our model only refer to vascular coupling and glial buffering. Lateral ion movement between the ECS of nearby neurons is a different diffusive process that determines the velocity of a travelling SD wave in tissue. This is not described in our framework. In the following section we will demonstrate in detail how can be understood as the inhibitory variable of this excitation process.
The above presented model is indeed the simplest ion-based neuron model that exhibits local SD dynamics. Model simplicity is an appealing feature in its own right, but one might doubt the physiological relevance of such a reduced model. Our hypothesis is that it captures very general dynamical features of neuronal ion dynamics, and to confirm this we will compare the results obtained with the reduced model to the physiologically much more detailed KWS model [12]. This detailed model contains five different gated ion channels (transient and persistent sodium, delayed rectifier and transient potassium, and NMDA receptor gated currents) and has been used intensively to study SD and SLA. In fact, one modification is required so that we can replicate the results obtained from the reduced model. The KWS model contains an unphysical so-called ‘fixed leak’ current (36)that has a constant reversal potential of mV and no associated ion species. This current only enters the rate equation for the membrane potential .
The effect on the model dynamics is dramatic. To see this note that the electroneutrality constraint Eq. (8) reflects a model degeneracy(37)that occurs when is modelled explicitly with (for details see Ref. [23]). With a fixed leak current Eq. (37) becomes (38)which implies that mV is a necessary fixed point condition for the system.
In other words, the type of bistability with a second depolarized fixed point that we normally find in closed systems is ruled out by this unphysical current. If we, however, replace it with a chloride leak current as in our model (cf. Eqs. (6) and (23)), i.e., a current with a dynamically adjusting reversal potential by virtue of Eq. (24), we find the same type of bistability for the closed system and monostability for the system that is buffered or coupled to a potassium bath. The morphological parameters (compartmental volumes and membrane surface area ) are the same as for the reduced model.
In fact in Ref. [14] the KWS model was used without additional ion regulation for a reaction-diffusion study of SD, and the only recovery mechanism of the local system seems to be this unphysical current. Theoretically SD could be a travelling wave in a reaction-diffusion system with bistable local dynamics, but unpublished results show that the propagation properties in the bistable system are dramatically different from standard SD dynamics with wave fronts and backs travelling at different velocities. We hence suppose that a local potassium clearing mechanism is crucially involved in SD.
We conclude this section with a discussion of the time scales of the model. To this end, it is helpful to keep in mind that the phenomenon of excitability requires a separation of time scales. We have electrical and ionic excitability and these dynamics themselves are separated by no fewer than three orders of magnitude.
Dynamics of happens on a scale that is faster than milliseconds. This follows from the gating time scale which is given explicitly in Eq. (15) and the time scale of of which can be computed from the membrane capacitance (given in Tab. 1) and the resistance of the ion channels (for details see Ref. [28]): (39)with (40)If we approximate the products of gating variables in the above expression with 0.1 this gives . Dynamics of happens on a scale in the order of milliseconds.
The time scale of ion dynamics is more explicit in the Goldman-Hodgkin-Katz (GHK) formalism than in the Nernst formalism used in this paper. The Nernst currents in Eqs. (21)–(23) are an approximation of the physically more accurate GHK currents, but in Ref. [23] we have shown that ion dynamics of GHK models and Nernst models are very similar. That is why the latter may be used for studies like this. For time scale considerations, however, we will now switch to the GHK description. The GHK current of ions with concentrations across a membrane is given by (41)where is the permeability of the membrane to the considered ion species and is the dimensionless membrane potential with (42)This expression contains the ideal gas constant , the temperature , ion valence and Faraday's constant . If we now write down the GHK analogue of the ion rate Eqs. (5) and (6) we obtain (43)For the conversion factor we have inserted the expression Eq. (13). The fraction term is of the order of the ion concentrations, is a dimensionless quantity and hence of order one. With the ion dynamics time scale (44)we can thus group the parameters as follows (45)Permeabilities of ion channels can be found in Refs [14], [23], [29]. Similar as for the resistance the permeability of a gated channel involves a product of gating variables. Approximating such terms again with 0.1 a typical value for the permeability is . Together with the values for the membrane surface area and the cell volume from Tab. 1 the time scale of transmembrane ion dynamics is .
The slowest time scales are related to potassium regulation, i.e., to dynamics. The glia scheme from Eq. (28) and Eq. (32) contains a forward buffering process that reduces at a time scale (46)and a backward buffering process with time scale (47)With the parameters from Tab. 1 this leads to and . So backward buffering is much slower. This is an important property, because in the following section we will see that recovery from FES requires a strong reduction of the potassium content. If buffering and backward buffering would happen on the same time scale the required potassium reduction would not be possible. Backward buffering could well happen at a considerably faster scale than Eq. (47), but as soon as is comparable to the buffer cannot re-establish physiological conditions after FES.
The glia scheme here is phenomenological. A more biophysically detailed model would describe a glial cell as a third compartment. An elevation of ECS potassium leads to glial uptake. Spatial buffering, i.e., the fast transfer of potassium ions between glial cells with elevated concentrations to regions of lower concentrations maintains an almost constant potassium level in the glial cells. In SD potassium in the ECS is strongly elevated during an about 80 sec lasting phase of FES and is continuously cleared during this time. After 80 sec the concentration quickly recoveres to even slightly less than the normal physiological level. Still there is a huge potassium deficit in the system and what we call backward buffering, i.e., the release of potassium from the glial cells, sets in. It is much slower than the uptake, because it is driven by a far smaller deviation of the potassium concentration from normal physiological resting level.
Similar to the above explanation of slow backward buffering in the glia scheme, an extremely slow backward time scale follows naturally in diffusive coupling. For diffusion the potassium content is reduced at a time scale (48)if extracellular potassium is greater than . Backward diffusion, however, only occurs in the final recovery phase that sets in after the neuron has returned from the transient FES state and is repolarized. While is still far from the resting state level, is comparable to normal physiological conditions (see the below bifurcation diagrams in Figs. 2b and 3b) and hence the driving force during the final recovery phase is very small for a bath concentration close to the resting state level. Consequently backward diffusion is much slower than forward diffusion.
Note that this argument for different slow regulation time scales relies exclusively on the almost constant values of the ECS potassium concentration along the physiological fixed point branch (see Figs. 2b and 3b). It is not a feature of the particular regulation scheme we apply.
The results are presented in three parts that describe (i) the stability of closed models, where we treat the change of the potassium content as a bifurcation parameter, (ii) open models, i.e., becomes a dynamical variable, with glial buffering and (iii) oscillations in ion concentrations in open models for bath coupling with the bath concentration as a bifurcation parameter.
At first we will not treat the change of the potassium content as a dynamical variable, but as a parameter whose influence on the system's stability we investigate. So the model we consider is defined by the rate Eqs. (3)–(6) and the constraint Eqs. (7), (8), (10)–(12) and (26). Its stability will be important for the full system with dynamical ion exchange between the neuron and a bath or glial reservoir to be discussed in the next two subsections. The phenomenon of ionic excitability as in SD only occurs for dynamical . We will see that a slow-fast decomposition of ionic excitability is possible. The fast ion dynamics is governed by the transmembrane dynamics that we discuss now and happens at the time scale . The dynamics of is much slower ( and ). Fast ion dynamics of the full system can hence be understood by assuming as a parameter that determines the level at which fast (transmembrane) ion dynamics occurs. This implies a direct physiological relevance of the closed system bifurcation structure with respect to potassium content variation for transition thresholds in the full (open) system.
The bifurcation diagram of the reduced model is presented in Fig. 2. It is shown in the -plane (Fig. 2a) and in the -plane (Fig. 2b) to display membrane and ion dynamics, respectively. A pair of arrows pointing in the direction of extracellular potassium changes only due to fluxes across the membrane (vertical ‘m’ direction) and only due to exchange with a reservoir (diagonal ‘r’ direction) is added to Fig. 2b.
The fixed point continuation yields a branch (black line) where fully stable sections are solid and unstable sections are dashed. Stability changes occur in saddle-node bifurcations (also called limit point bifurcation, LP) and Hopf bifurcations (HB). In a LP the stability changes in one direction (zero-eigenvalue bifurcation), in a HB it changes in two directions and a limit cycle is created (complex eigenvalue bifurcation). A limit cycle is usually represented by the maximal and minimal value of the dynamical variables. However, the oscillation amplitude of the ionic variables is almost zero for the limit cycles in our model. Maximal and minimal values cannot be distinguished on the figure scale. Hence in the -plane the limit cycle continuation appears only as a single line (green). Stability changes of limit cycles occur in saddle-node bifurcations of limit cycles (LP). The limit cycles in our model disappear in homoclinic bifurcations. In this bifurcation a limit cycle collides with a saddle. When it reaches the saddle it becomes a homoclinic cycle of infinite period.
As a reference point the initial physiological condition is marked by a black square. We will call the entire stable fixed point branch that contains this point the physiological branch , because the conditions are comparable to the normal functioning physiological state—in particular, action potential dynamics is possible when the system is on this branch.
Let us discuss the bifurcation diagram starting from this reference point and follow the fixed point curve in the right direction, i.e., for increasing . The physiological fixed point loses its stability in the first (supercritical) Hopf bifurcation (HB1) at mM. The extracellular potassium concentration is then at mM. In other word, much of the added potassium has been taken up by the cell.
The limit cycle associated with HB1 loses its stability in a period-doubling bifurcation (PD) and remains unstable. Finally it disappears in a homoclinic bifurcation shortly after its creation (cf. right inset in Fig. 2a). The stable limit cycle emanating from the PD point becomes unstable in a and vanishes in a homoclinic bifurcation, too. The parameter range of these bifurcations is extremely small (). Such fine parameter scales will not play a role for the interpretation of ion dynamics. Ion concentrations are stationary and physiological up to , but for practical purposes it is irrelevant if we identify or as the end of the physiological branch .
The first HB is followed by four more bifurcations (LP1, HB2, LP2, HB3) that all neither restore the fixed point stability nor create any stable limit cycles. The limit cycles for HB2 and HB3 are hence not plotted either. It is only the fourth Hopf bifurcation (HB4) at mM in which the fixed point becomes stable again and in which a stable limit cycle is created. The limit cycle branch loses its stability in LP1 and regains it in LP2. It becomes unstable again and even more unstable in LP3 and LP4. Shortly after that (not resolved on the scales in Fig. 2) it ends in a homoclinic bifurcation with the saddle between HB1 and LP2. At HB4 the stable free energy-starved branch begins. It is generally characterized by a strong increase in the ECS potassium compared to physiological resting conditions (Fig. 2b), and a significant membrane depolarization (Fig. 2a). Corresponding to the extracellular elevation intracellular potassium is significantly lowered. This goes along with inverse changes of the compartmental sodium concentrations (all not shown). is hence characterized by largely reduced ion gradients and strong membrane depolarization. In fact, at this membrane potential the sodium channels are inactivated which is normally called depolarization block in HH-like membrane models without ion dynamics. Depolarization block is, however, only one feature of FES. The closeness of FES to the thermodynamic equilibrium of the system is more importantly manifested in the reduced ion gradients. On no more bifurcations occur and it remains stable for increasing .
The interpretation of this bifurcation diagram should be as follows. The end of defines the maximal potassium content compatible with a physiological state of a neuron. For larger it will be inevitably driven to the FES. In other words the end of marks the threshold value for a slow, gradual elevation of the potassium content to cause the transition from physiological resting conditions to FES. In a buffered system it is the threshold for SD ignition. On the other hand stable FES-like conditions require a minimal potassium content which marks the end of . It is given by mM. Below this value the only stable fixed point is physiological. Again there is a narrow range, namely between and mM, in which stable oscillations can occur.
When glial buffering is at work the end of defines the threshold for potassium buffering, i.e., for the potassium reduction that is required to return from FES to physiological conditions (cf. Eq. (27)). In the second subsection of Sect. Results, we will see that this is exactly how ion regulation facilitates recovery in SD models.
There is another way the bifurcation diagram in Fig. 2b can be read. As we have remarked above the limit cycles of the model are characterized by large oscillation amplitudes in the membrane variables (not shown) and , but almost constant ionic variables , and (only shown). So Fig. 2b tells us which extracellular potassium concentrations can possibly be stable and which ones cannot. Values below the end of at mM, values between mM and mM and finally concentrations in the range of starting at mM can be stable. Any other extracellular potassium concentration is unstable and the system will evolve towards a stable ion configuration that is present in the phase space. The highest stable potassium concentration below FES values is . If potassium in the ECS is increased instantaneously, this value indicates the threshold for SD ignition or the transition to FES.
Performing the same type of bifurcation analysis with the physiologically more detailed model from Kager et al. [12], [14] (cf. last paragraph of Sect. Models) leads to the diagram in Fig. 3. It has been shown before that also in this model there is stable FES [23]. We do not find the same bifurcations as in the reduced model, but only two LPs and one HB. However, the physiological implications are very similar. Like in the reduced model there is an upper limit of the potassium content for stable physiological conditions ( mM) and a lower limit for stable FES ( mM). Also the downward snaking and the stability changes of the limit cycle that starts from HB1 are very similar to Fig. 2. This leads to the same type of conclusion concerning possible stable extracellular potassium concentrations. While numerical values of the stability limits in terms of are specific to each model, the topological similarity of the bifurcation diagrams suggests a generality of results: there is a stable physiological branch that ends at some maximal value of the potassium content. Beyond this point the neuron cannot maintain physiological conditions, but will face FES. On the other hand the stable FES branch ends for a sufficiently reduced potassium content the neuron will return to physiological conditions.
The new bifurcation diagrams presented in this section confirm our results from Ref. [23]: Neuron models whose ionic homeostasis is only provided by ATPase-driven pumps, but without diffusive coupling or glial buffering, will have a highly unphysiological fixed point that is characterized by free energy-starvation and membrane depolarization. However, the presented bifurcation diagrams here contain additional information of great importance. Using the new bifurcation parameter crucially extends our results from Ref. [23] by uncovering the threshold concentrations in extracellular potassium concentration. These are completely novel insights.
In the next subsection the bifurcation diagrams of the unbuffered (closed) systems shall facilitate a phase space understanding of the activation and inhibition process of ionic excitability as observed in SD in the buffered (open) systems. We are aiming for an interpretation of ionic excitability where neuronal discharge and recovery are fast dynamics that are governed by the bistable structure discussed above, whereas additional ion regulation takes the role of slowly changing .
However, only the gated ion dynamics, i.e., dynamics of sodium and potassium is fast compared to that of , chloride is similarly slow. Due to the enforcement of electroneutrality this means that the overall concentration of positively charged ions in the ICS, i.e., the sum of sodium and potassium ion concentrations changes on the same slow time scale as the chloride concentration.
To describe this slow process not dynamically but—like —in terms of a parameter we simply investigate the stability for a given distribution of non-dynamic, i.e., impermeant chloride. To determine this stability we set the chloride current to zero and vary in a certain range (from 8 to 32 mM for the reduced model, and from 9 to 33 mM for the detailed model). This affects the system only through the electroneutrality constraint Eq. (7) which sets the intracellular charge concentration to be shared by sodium and potassium.
For each value of we perform a fixed point continuation as in Figs. 2 and 3 which yields similarly folded s-shaped curves. The result is shown in Fig. 4. For our analysis of SD it is only relevant where ends. That is why the plot does not contain the whole fixed point curve, but only and a part of the unstable branch for a selection of values. As a reference the diagrams also contain the fixed point curves from Figs. 2 and 3 which include chloride dynamics. The FES branches in Fig. 4 end in Hopf bifurcations. The bifurcation points for different chloride concentrations yield the blue Hopf line. It marks the threshold for recovery from FES when dynamics of chloride and is slow.
In the previous subsection we have analyzed the phase space structure of ion-based neuron models without contact to a reservoir, i.e., without glial buffering or diffusive coupling. These models have only transmembrane ion dynamics and obey mass conservation of each ion species. Hence they describe a closed system. The bistability of a physiological state and FES that we found in these closed models is not experimentally observed, because real neurons are always open systems not merely in the sense that they consume energy—a necessary prerequisite for being far from thermodynamic equilibrium—but they also can lose or gain ions through reservoirs or buffers. We will now include glial buffering and show how it facilitates recovery from FES, a condition which in contrast to the physiological state is close to a thermodynamic equilibrium, namely the Donnan equilibrium (cf. Ref. [23]).
When glial buffering is at work, becomes a dynamical variable whose dynamics is given by the buffering rate Eq. (32). In a previous subsection we have explained that the bifurcation diagrams in Figs. 2 and 3 imply thresholds for an elevation of extracellular potassium to trigger the transition from physiological resting conditions to FES. This is in agreement with computational and experimental SD studies in which high extracellular potassium concentrations are often used to trigger SD. Another physiologically relevant way of SD ignition is the disturbance or temporary interruption of ion pump activity. As we have shown in Ref. [23] there is a minimal pump rate required for normal physiological conditions in a neuron. Below this rate the neuron will go into a FES state and remain in that state even when the pump activity is back to normal.
For the simulations in Fig. 5 we have interrupted the pump activity for about 10 sec in the reduced model, and we have elevated the extracellular potassium concentration by mM in the detailed model to trigger SD. Both stimulation types work for both models, but only the two examples are shown. The phase of pump interruption (Fig. 5a and 5c) is indicated by the shaded region in the plots, the time of potassium elevation is marked by the vertical grey line. The dynamics of the two models is very similar: in response to the stimulation the neuron strongly depolarizes and remains in that depolarized state for about 70 sec (Fig. 5a and 5b). After that the neurons repolarize abruptly and asymptotically return to their initial state. In addition to the membrane potential (black curve) the potential plots also contain the Nernst potentials for sodium (red line), potassium (blue line) and chloride (green line) that change with the ion concentrations according to the definition of the Nernst potentials in Eq. (24). In Fig. 5c and 4d we see that the potential dynamics goes along with great changes in the ion concentrations. In particular, extracellular potassium is strongly increased in the depolarized phase. These conditions are very similar to the type of FES states discussed in the previous subsection. The recovery of ion concentrations sets in with the abrupt repolarization, but it is a very slow asymptotic process that is not shown in Fig. 5.
In both models the neuron is capable of producing spiking activity again right after the repolarization. All these aspects of ion dynamics during SD are well-known from several studies [12], [14]. We remark that the time series are almost identical if glial buffering is replaced by the coupling to a potassium bath. Both, the strength of glial buffering and of diffusive coupling have been adjusted so that the depolarized phase lasts about 70 sec which is the experimentally determined time. We will focus on bath coupling in last subsection of Sec. Results. If neither buffering nor a potassium bath is included the neuron does not repolarize (for time series plots of terminal transitions to FES see Ref. [23]).
The time series in Fig. 5 are useful to confirm that the neuron models we investigate have the desired phenomenology and indeed show SD-like dynamics. Yet the nature of the different phases of this ionic excitation process—the fast depolarization, the prolonged FES phase and the abrupt repolarization—remains enigmatic [12], [14], [29], [30]. In a phase space plot the picture becomes much clearer and the entire process can be directly related to the two stable branches, and , that we found for the closed and therefore pure transmembrane models in the previous subsection. In Fig. 6 the time series from Fig. 5 for a simulation time of 50 min are shown in the - and the -plane. The parts of the trajectories during the stimulation (pump interruption and potassium elevation) are dashed. In the chosen planes vertical lines belong to dynamics of constant potassium contents that can be understood in terms of the models we analyzed in the previous subsection. That is why Fig. 6 contains the fixed point curves from Fig. 4 as shaded lines as a guide to the eye. In Fig. 6c and 6d buffering dynamics is diagonal as indicated by the pair of arrows added to the plot.
For both trajectories the stimulation is followed by a vertical activation process that leads to the transition from to . The verticality means that this is a process almost purely due to transmembrane dynamics. It is governed by the bistable phase space structure that we discussed in the previous section and also in Ref. [23]. Buffering dynamics is too slow to inhibit the activation. The types of stimulation we applied are related to bifurcations of the transmembrane system: the potassium elevation is beyond the end of which is marked by the first Hopf bifurcation (HB1) in Fig. 2. The interruption of pump activity means that we go below a pump rate threshold that is defined by a saddle-node bifurcation (cf. Ref. [23]). More generally, to initiate an ionic excitation it is necessary to stimulate the system until it enters the basin of attraction—derived in the unbuffered system—of the FES state. The activation is followed by a phase of both, slow transient transmembrane dynamics mostly due to chloride, and potassium buffering. It is the latter that bends the trajectories in the diagonal direction so that they go along the FES branches from Fig. 4. The trajectories slowly approach the repolarization threshold given by the Hopf line. The duration of this FES phase is determined by how long it takes the system to reach the Hopf line.
This process is a mixture of buffering and transient transmembrane dynamics for the reduced model and more buffering-dominated in the detailed model. The duration of the FES phase is consequently a result of both types of dynamics. However, the main insight we gain from this plot is: glial buffering is the necessary inhibitory mechanism that takes the system to the Hopf line so that it can repolarize. We remark that the time series and phase space plots for bath coupling instead of buffering are almost identical and the same interpretation holds. The more general conclusion is then: ion dynamics beyond transmembrane processes is necessary to take the system to the Hopf line so that it can repolarize. This can, of course, be a combination of bath coupling and buffering. When the Hopf line is reached that neuron repolarizes abruptly which is the second almost purely vertical process. The repolarization is followed by slow asymptotic recovery dynamics of ion concentrations that takes the neuron back to the initial state which is at mM. The neuron regains the electrical excitability that is lost during FES already right after the repolarization. So the system is back to physiological function long before the ion gradients are fully restored.
Let us summarize the results from this subsection. By relating the SD time series from Fig. 5 to the bifurcation structure of the unbuffered models from the first subsection of Sect. Results and in particular to the two stable branches and we have succeeded to understand ionic excitability as a sequence of different dynamical phases. The initial depolarization and the later repolarization are membrane-mediated fast processes that obey the bistable dynamics of unbuffered systems. The FES phase is buffering-dominated and lasts until buffering has taken the system to a well-defined repolarization threshold. The recovery phase is dominated by backward buffering. The full excursion time is the sum of the durations of each phase. For the de- and repolarization process this duration mainly depends on the time scale of the transmembrane dynamics and is hence comparably short. The duration of the FES phase is a result of both, the transient transmembrane dynamics and glial ion regulation at a much slower time scale. The final recovery phase is mainly backward buffering dominated which is the slowest process. Hence the duration of an SD excursion is mainly determined by the slow buffering and backward buffering time scales. This conclusion that relies on our novel understanding of the different thresholds involved in SD is in fact in agreement with recent experimental data suggesting vascular clearance of extracellular potassium as the central recovery mechanism in SD [31], [32].
The dynamics of excitable systems can often be changed to self-sustained oscillations by a suitable parameter variation. The type of bifurcation that leads to the oscillations and the shape of the limit cycle in the oscillatory regime determine excitation properties like threshold sharpness and latency [28]. The oscillatory dynamics that is related to ionic excitability can be obtained for bath coupling with an elevated bath concentration . So in this section we replace the buffering dynamics for with the diffusive coupling given by Eq. (35). This coupling is used in experimental in-vitro studies of SD [33] and has also been applied in computational models that are very similar to our reduced one [19]–[21].
Depending on the level of the bath concentration, we find three qualitatively different types of oscillatory dynamics that are shown in Fig. 7. The top row (a) shows the time series of seizure-like activity for . It is characterized by repetitive bursting and low amplitude ion oscillations. The other types of oscillatory dynamics are tonic firing at with almost constant ion concentrations (Fig. 7b) and periodic SD at with large ionic amplitudes (Fig. 7c). We see that SLA and periodic SD exhibit slow oscillations of the ion concentrations and fast spiking activity, which hints at the toroidal nature of these dynamics. Below we will relate SLA and periodic SD to torus bifurcations of the tonic firing limit cycle.
The examples in Fig. 7 show that our model contains a variety of physiologically distinct and clinically important dynamical regimes. A great richness of oscillatory dynamics, in fact, under the simultaneous variation of and the glial buffering strength has already been reported in Refs. [19]–[21] for a very similar model. In Ref. [19], [20] the authors even give a bifurcation analysis of ionic oscillations for elevation.
To investigate dynamical changes and the transitions between the dynamical regimes in our model we perform a similar bifurcation analysis and vary , too. Two important differences should be noted though. First, Ref. [19], [20] uses an approximation of the multi-time scale model in which the fast spiking dynamics is averaged over time, while our analysis does not rely on such an approximation. Second, our analysis covers a bigger range of values which allows us to compare SLA and SD, while Ref. [19], [20] exclusively deals with SLA.
Fig. 8 shows the bifurcation diagram for variation in the -plane and in the -plane. In addition to fixed points (black) and limit cycles (green) also quasiperiodic torus solutions (blue) are contained in the diagram. In comparison to Fig. 2 this model contains a new type of bifurcation, namely a torus bifurcation (TR). A torus bifurcation is a secondary Hopf bifurcation of the radius of a limit cycle in which an invariant torus is created. If this torus is stable, nearby trajectories will be asymptotically bound to its surface. However, we cannot follow such solutions with standard continuation techniques, because these require an algebraic formulation in terms of the oscillation period. This is not possible for torus solutions, because on a torus the motion is quasiperiodic, i.e., characterized by two incommensurate frequencies. We can hence only track the stable solutions by integrating the equations of motion and slowly varying . It is due to this numerically expensive method that in this section we will only analyze oscillatory dynamics of the reduced HH model with time-dependent ion concentrations.
The result of this bifurcation analysis in Fig. 8 shows us that there is a maximal level of the bath concentration compatible with physiological conditions. It is identified with the subcritical Hopf bifurcation HB1 in which the fixed point loses its stability. The related limit cycle is omitted, because it stays unstable and terminates in a homoclinic bifurcation with the unstable fixed point branch. The fixed point undergoes further bifurcations (LP1, LP2, HB2, HB3) which all leave it unstable and do not create stable limit cycles. It is in HB4 that the fixed point becomes stable again and also a stable limit cycle is created. This is the last fixed point bifurcation of the model.
The limit cycle that is created in HB4 changes its stability in several bifurcations. The physiologically most relevant ones are the four torus bifurcations. The bifurcation labels indicate the order of detection for the continuation that starts at HB4. Initially the limit cycle is characterized by fast low-amplitude oscillations. It becomes unstable in the subcritical torus bifurcation TR1. It regains and again loses its stability in the subcritical torus bifurcations TR2 and TR3. The last torus bifurcation, the restabilizing supercritical TR4, is directly followed by a PD after which no stable limit cycles exist any more. Again we have omitted in the diagram the unstable branch after PD and the limit cycle that is created in PD, which remains unstable.
Physiologically it is more intuitive to discuss the diagram for increasing starting from the initial physiological conditions marked by the black square. Normal physiological conditions become unstable at and above this value the neuron spikes continuously according to the stable limit cycle branch between PD and TR4. When is reached the dynamics changes from stationary spiking to seizure-like activity on an invariant torus. The beginning of SLA is hence due to a supercritical torus bifurcation and the related ionic oscillation sets in with finite period and zero amplitude. From on tonic spiking activity is stable again and there is a small -range of bistability between SLA and this tonic firing. As we mentioned above solutions on an invariant torus cannot be followed with normal continuation tools like AUTO, so only stable branches are detected. The details of the bifurcation scenario at TR3 are hence not totally clear, but we suspect that the unstable invariant torus that must exist near TR3 collides with the right end of the stable torus SLA-branch in a saddle-node bifurcation of tori. Tonic spiking then remains stable until TR2. This bifurcation is related to the period SD that already exist well below . In fact, the threshold value is in agreement with experiments [33]. Again the unstable torus near TR2 is not detected, but we suppose that a similar scenario as in TR3 occurs. The dynamics on the torus branch related to TR2 (and TR1 where it seems to end) is very different from the first torus branch. While the periods of the slow oscillations during SLA are 16–45 sec the ion oscillations of periodic SDs are much slower with periods of 350–550 sec.
Another crucial difference is obvious from Fig. 8b which shows the bifurcation diagram in the -plane. The fixed point is just a straight line, because the diffusive coupling Eq. (35) makes a necessary fixed point condition. The limit cycle is always extremely close to this line. On the chosen scale it cannot be distinguished from the fixed point and is hence not contained in the plot. Only the torus solutions of SD and SLA attain values that differ significantly from the regulation level. The ionic amplitudes of SD are one order of magnitude larger than those of SLA. This has to do with the fact that the peak of SD—as described above—must be understood as a metastable FES state that exists due to the bistability of the transmembrane dynamics. The dynamics of SLA is clearly of a different nature.
Note that the bifurcation diagram reveals a bistability of tonic firing and full-blown SD between the left end of the SD branch at about 11 mM and TR2. This means that there is no gradual increase in the ionic amplitudes that slowly leads to SD, but instead it implies that SD is a manifest all-or-none process.
In Fig. 9 we look at the same bifurcation diagram in the - and the -plane. While in Fig. 8 most of the ionic phase space structure is hidden, because for fixed points and limit cycles, the -presentation in Fig. 9a provides further insights into the ion dynamics. We see that the stable fixed point branch before HB1 has extracellular sodium concentrations close to the physiological value . The stable branch after HB4, however, has an extremely reduced extracellular sodium level and is indeed FES-like. The stable limit cycles between PD and TR4 and between TR3 and TR2, and also SLA are rather close to the physiological sodium level. On the other hand, periodic SD is an oscillation between FES and normal physiological conditions, which is an expected confirmation of the findings from the previous section.
Fig. 9b is useful in connecting the phase space structure of the bath coupled system to that of the transmembrane model of the first subsection of Sect. Results. If we interchange the - and the -axis in the diagram it looks very similar to Fig. 2b. The torus bifurcations TR1, TR2 and TR3 are very close to the limit point bifurcations , and of the transmembrane model. The fixed point curves are topologically identical.
This striking similarity has to do with the fact that the limit cycle in Fig. 2 has almost constant ion concentrations. We have pointed out in the first subsection of Sect. Results that Fig. 2 tells us which extracellular potassium concentrations are stable for pure transmembrane dynamics. Diffusive coupling with bath concentrations at such potassium levels leads to negligibly small values of (cf. Eq. (35)). Therefore the limit cycle is still present in the bath coupled model and also the stability changes can be related to those in the transmembrane model. Again this can be seen as a confirmation of the results from the previous section: the transmembrane phase space plays a central role for models that are coupled to external reservoirs. We can interpret the ionic oscillations from Fig. 7 and the bifurcations leading to them with respect to this phase space.
Last we consider the dynamics of SLA and periodic SD in a phase space projection. In Fig. 10 the trajectories for SLA and periodic SD are plotted in the -plane together with the underlying fixed point and limit cycles from the transmembrane model (cf. Fig. 4). The periodic SD trajectory has a very similar shape to the single SD excursion from Fig. 6 and is clearly guided by the stable fixed point branches and . On the other hand SLA is a qualitatively very different phenomenon. Rather than relating to the FES branch, it is an oscillation between physiological conditions and those stable limit cycles that exist for moderately elevated extracellular potassium concentrations. The ion concentrations remain far from FES. So SLA and SD are not only related to distinct bifurcations, though of similar toroidal nature and branching from the same limit cycle, but they are also located far from each other in the phase space. This completes our phase space analysis of local ion dynamics in open neuron systems.
In this paper we have analyzed dynamics at different time scales in a HH model that includes time-dependent ion concentrations. Such models are also called second generation Hodgkin-Huxley models. They exhibit two types of excitability, electrical and ionic excitability, which are based on fast and slow dynamics. The time scales of these types of excitability are themselves separated by four to five orders of magnitude. The dynamics ranges from high-frequency bursts of about 100 Hz with short interburst periods of the order of 10 msec (Fig. 7a) to the slow periodic SD with frequencies of about Hz and periods of about 7:30 min (Fig. 7c).
The slow SD dynamics in our model is classified as ultra-slow or near-DC (direct current) activity and cannot normally be observed by electroencephalography (EEG) recordings, because of artifacts due to the resistance of the dura (thick outermost layer of the meninges that surrounds the brain). However, recently subdural EEG recordings provided evidence that SDs occur in abundance in people with structural brain damage [1]. Indirect evidence was already provided earlier by functional magnetic resonance imaging (fMRI) [34] and a patient's symptom reports combined with fMRI [35] that SD also occurs in migraine with aura [2].
The slowest dynamics that can be accurately measured by EEG, i.e., the delta band, with frequencies about 0.5 to 4 Hz, has attracted modelling approaches much more than SD, which was doubted to occur in human brain until the first direct measurements were reported. It is interesting to compare the origin of slow time scales in such delta band models to our slow dynamics.
Models of the delta band essentially come in two types. On the one hand thalamo-cortical network and mean field models of HH neurons with fixed ion concentrations have been studied [36]. In this case, a slow time scale emerges because the cells are interconnected via synaptic connections using metabotropic receptors that are slow, because they act through second messengers. On the other hand, single neuron models with currents that are not contained in HH, namely a hyperpolarization-activated depolarizing current, -dependent sodium and potassium currents, and a persistent sodium current, were suggested. The interplay between these currents gives rise to oscillations at a frequency of about 2–3 Hz [37]. It is therefore hardly surprising that these currents, in particular the persistent sodium and the -dependent sodium and potassium currents, have also been proposed to play an essential role in SD [30], [38]. Furthermore, bursting as another example of slow modulating dynamics was studied in a pure conductance-based model with a dendritic and an axo-somatic compartment [16]. Also metabotropic receptors as modeled by Bennett et al. [15] and other cellular processes at appropriately slow time scales may play a role and contribute to the repolarization.
In contrast to those approaches our results show that dynamics in a HH framework with time-dependent ion concentrations and buffer reservoirs already range from seconds to hours even with the original set of voltage-gated ion currents. Time scales from milliseconds (membrane dynamics) to seconds (ion dynamics) and even minutes to hours (ion exchange with reservoirs) can be directly computed from the model parameters (cf. Sect. Models). The interplay of membrane dynamics, ion dynamics and coupling to external reservoirs (glia or vasculature) naturally leads to dynamics typical of SLA and SD.
In particular SD is explained by a bistability of neuronal ion dynamics that occurs in the absence of external reservoirs. The potassium gain or loss through reservoirs provided by an extracellular bath, the vasculature or the glial cells is identified as a bifurcation parameter whose essential importance was not realized in earlier studies (see Fig. 11). Using this bifurcation parameter and the extracellular potassium concentration as the order parameter, we obtain a folded fixed point curve with the two outer stable branches corresponding to states with normal physiological function, hence named physiological branch , and to states being free-energy starved ().
The definition of the bifurcation parameter implies that exchange with ion reservoirs happens along the diagonal direction labelled by ‘r’. Membrane-mediated dynamics is in the vertical ‘m’ direction. In the full system where the ion exchange is a dynamical variable our unconventional choice of variables, i.e. modelling instead of , makes it obvious that the time scales of diagonal and vertical dynamics is separated by at least two orders of magnitude. Slow dynamics is along and , and the fast dynamics describes the jumps between these branches. We remark that dynamics along is slower than along , because the branch is almost horizontal which leads to a very small gradient driving the diffusive coupling. Similarly the release of buffered potassium from the glial cells is only weakly driven (cf. the discussion of buffering time scales in Sect. Model).
In the closed system sufficiently strong stimulations lead to the transition from the physiological resting state located on to FES. In the full system with dynamical ion exchange with the reservoirs, physiological conditions are restored after a large phase space excursion to the the before stable FES state. We refer to this process as ionic excitability. In contrast to the electrical excitability of the membrane potential this process involves large changes in the ion concentrations. The entire phase space excursion of this excitation process can be explained through the specific transits between and along and .
We observe ion changes on three slow time scales. (i) Vertical transits between and caused by transmembrane dynamics in the order of seconds. The time scale is determined by the volume-surface-area ratio and the membrane permeability to the ions. (ii) Diagonal dynamics along in the order of tens of seconds caused by contact to ion reservoirs. This time scale is determined by buffer time constants or vascular coupling strength. (iii) Dynamics on again caused by contact to ion reservoirs, but at the slower backward buffering time scale in the order of minutes to hours determined by the slower backward rate of the buffer [12]. During this long refractory phase of ionic excitability the spiking dynamics based on electrical excitability—separated by seven orders of magnitude—seems fully functional.
The right end of and the left end of are marked by bifurcations that occur for an accordingly elevated or reduced potassium content. This is the first explanation of thresholds for local SD dynamics in terms of bifurcations. We remark, however, that for SD ignition the important question is not where ends, but instead where the basin of attraction of begins.
This new understanding of SD dynamics suggests a method to investigate the SD susceptibility of a given neuron model. One should consider the closed model without coupling to external reservoirs and check if shows the typical bistability between a physiological resting state and FES. We remark that unphysical so-called ‘fixed leak’ currents must be replaced by proper leak currents with associated leaking ions. Thresholds for the transition between and translate to thresholds for SD ignition and repolarization, i.e., recovery from FES in the full open model. Knowledge of the potassium reduction needed to reach the repolarization threshold and knowledge about the buffer capacity could then tell us if recovery from FES can be expected (such as in migraine with aura) or if the depolarization is terminal (such as in stroke).
Although our model does not contain all important processes involved in SD, our phase space explanation appears to be valid also for certain model extensions. For example, considering only diffusive regulation of potassium is physically inconsistent, but adding an analoguous regulation term for sodium turns out not to alter the dynamics qualitatively. Moreover osmosis-driven cell swelling—normally regarded as a key indicator of SD—is not included in our model, but can be added easily [13], [30], [39]. Unpublished results confirm that also with such cell swelling dynamics the fundamental bifurcation structure of Fig. 11 is preserved.
As a clinical application of our framework, we have linked a genetic defect, which affects the inactivation gate and which is present in a rare subtype of migraine with aura, to SD. Our simulations show that such mutations render neurons more vulnerable to SD [40]. The interesting point, however, is that on the level of the fast time scale the firing rate is decreased, which in a mean field approach (as done for the delta band) translates to decreased activity. This effect seemingly contradicts the increased SD susceptibility and hence illustrates the pitfalls in trying to neglect ion dynamics in the brain and to bridge the gap in time scales by population models.
|
10.1371/journal.ppat.1005091 | Inhibition of mTORC1 Enhances the Translation of Chikungunya Proteins via the Activation of the MnK/eIF4E Pathway | Chikungunya virus (CHIKV), the causative agent of a major epidemic spanning five continents, is a positive stranded mRNA virus that replicates using the cell’s cap-dependent translation machinery. Despite viral infection inhibiting mTOR, a metabolic sensor controls cap-dependent translation, viral proteins are efficiently translated. Rapalog treatment, silencing of mtor or raptor genes, but not rictor, further enhanced CHIKV infection in culture cells. Using biochemical assays and real time imaging, we demonstrate that this effect is independent of autophagy or type I interferon production. Providing in vivo evidence for the relevance of our findings, mice treated with mTORC1 inhibitors exhibited increased lethality and showed a higher sensitivity to CHIKV. A systematic evaluation of the viral life cycle indicated that inhibition of mTORC1 has a specific positive effect on viral proteins, enhancing viral replication by increasing the translation of both structural and nonstructural proteins. Molecular analysis defined a role for phosphatidylinositol-3 kinase (PI3K) and MAP kinase-activated protein kinase (MnKs) activation, leading to the hyper-phosphorylation of eIF4E. Finally, we demonstrated that in the context of CHIKV inhibition of mTORC1, viral replication is prioritized over host translation via a similar mechanism. Our study reveals an unexpected bypass pathway by which CHIKV protein translation overcomes viral induced mTORC1 inhibition.
| The ongoing chikungunya epidemic outbreak in the Caribbean, Central and South America highlights how poor is our understanding of CHIKV pathogenesis and the urgent need for new strategies that may limit CHIKV spread. Immunological studies have suggested that dissemination of infection is largely determined by early events of viral-host cell interactions. In our prior study, we investigated the role of type I interferon responses and the autophagy pathway as mediators of viral control. Here, we evaluated the role of mTOR, making the surprising discovery that inhibition of mTORC1 enhances viral protein translation independently of type I IFN and autophagy. While the inhibition of mTORC1 has no impact on viral binding or entry, we observed an increased translation of both structural and nonstructural viral proteins. Interestingly, the positive impact of mTORC1 inhibition is restricted to viral proteins, as compared to host cap-dependent protein translation that remains suppressed. Further analysis demonstrates that this bypass pathway is mediated the activation of PI3K and MnKs, which in turn hyper-phosphorylate eIF4E, a critical initiation protein for translation. Notably, CHIKV replication enables this pathway as a means to efficiently replicate. Thus, our study provides an unexpected role for mTORC1 in the control of CHIKV infection and highlights a new strategy by which the expression of CHIKV proteins can bypass and/or use the inhibition of mTORC1.
| Since 2005 there has been a recurrence of Chikungunya disease, with the initial outbreak occurring in the French territory La Réunion [1]. The epidemic has spread worldwide, with outbreaks in five continents [2,3]. Notably, in just nine months during, Chikungunya virus (CHIKV) spread to 22 countries in the Caribbean, Central and South America, resulting in hundreds of thousands of cases [4]. The treatment of CHIKV infections relies on symptomatic relief, as no effective anti-viral agents are available [3]. We therefore set out to investigate cellular pathways that regulate CHIKV replication and spread.
Like other alphaviruses, CHIKV contains a single positive stranded RNA genome of approximately 11.5 kB [5]. The genomic RNA is capped and polyadenylated, and encodes two open reading frames (ORFs). The 5' ORF encodes four nonstructural proteins that participate in genome replication [6]. It is expressed via cap-dependent translation as an nsP1–3 or nsP1–4 polyprotein that is cleaved by the nsP2-encoded protease. The structural proteins are encode by a single ORF within the subgenomic region and is also translated via a cap-dependent mechanism [7]. As observed for other alpahviruses (e.g., Sindbis), CHIKV infection induces several cell stress responses, which might be associated with pathogenesis [8]. In our previous work, we showed that cells infected by CHIKV exhibit phenotypic characteristics of oxidative stress, endoplasmic reticulum stress, interferon induction, autophagy and apoptosis [9]. Notably, these adaptations are due, in part, to modification of the regulator kinase mTOR.
The mammalian target of Rapamycin (mTOR) coordinates cellular catabolic and anabolic processes to promote growth, proliferation and survival signals [10]. mTOR elicits its pleiotropic functions in the context of two functionally distinct signaling complexes, termed mTOR complex 1 (mTORC1) and complex 2 (mTORC2). mTORC1, which contains mTOR, mLST8/GβL, Raptor and PRAS40, plays a key role in cap-dependent translation initiation by directly phosphorylating p70 S6 kinase 1 (S6K1) and eIF4E-binding protein 1 (4E-BP1), and is sensitive to Rapamycin [10]. S6K1 phosphorylate several proteins that are associated with mRNA translation or its control, including ribosomal protein S6 and eukaryotic initiation factor 4B (eIF4B) [10]. The 4E-BP1 are small phosphoproteins which bind to eIF4E at a site that overlaps its interaction site for eIF4G, preventing the formation of eIF4F complex essential for the initiation of capped mRNA [10]. mTORC2 shares mTOR and mLST8/GβL with mTORC1, but possesses three unique components, namely, rictor, mSin1 and PRR5/Protor [11]. Despite the presence of mTOR, mTORC2 is considerably less susceptible to Rapamycin inhibition [10].
As a master sensor of cellular homeostatic perturbations, several studies have investigated the relationship between mTOR activity and viral infection. Numerous viruses modify the activity of mTOR (or mTOR pathways) [12–14]. Regulation of mTOR induces virus-specific effects that are often with opposing action. For example, blocking of mTOR by Rapamycin or TORISEL (referred to be the class of drugs known as Rapalog) inhibit the replication of HCMV [12]; yet Rapalog treatment facilitates HEV replication [13]. During HCV infection of liver cells, activation of PI3K-PKB-mTOR mediates both viral supportive functions (e.g., prevention of apoptosis in HCV-infected cells), and the production of antiviral interferon [14,15].
Regarding CHIKV infection, we previously observed that transient inhibition of the mTORC1 pathway during the first hour of infection [9]. While we demonstrated that this transient inhibition of mTOR correlates with CHIKV-induced autophagy, the direct role of mTOR activity on CHIKV replication remained unknown. Herein, we report that inhibition of mTORC1 enhances CHIKV replication in a cell-intrinsic manner and promotes in vivo spread and worsening of disease. Moreover, we show that mTORC1 impacts CHIKV infection independently of type I IFN production and autophagy. Instead, its actions directly effect translation of viral proteins via the activation of the MnK/eIF4E pathway. These results reveals a role for mTORC1 as a host defense mechanism that limits CHIKV replication and highlights a new strategy by which the expression of CHIKV proteins can bypass the inhibition of mTORC1.
To determine the relationship between CHIKV infection and mTOR activity, mouse embryonic fibroblast (MEFs) cells were transfected with siRNA to suppress mTOR expression (Fig 1A), followed by infection with CHIKV (CHIKV-21, the La Réunion 2005 strain). Surprisingly, we observed a 3 fold increase in the number of infected cells as compared to control siRNA treatment (Fig 1B and 1C). Similar findings were found across different doses of input virus, following both the percentage of E2+ cells and extracellular viral load (Fig 1C and 1D). To confirm the impact of mTOR activity on CHIKV infection, we tested PP242, an ATP-competitive mTOR inhibitor that specifically targets the active site of mTOR kinase, and again observed enhanced CHIKV infection (S1A and S1B Fig). These results suggest that mTOR activity restricts CHIKV infection, and highlighted an important paradox. While viral mRNA employs cap-dependent translation to replicate [7], CHIKV infection is actually increased when mTOR is inhibited.
To ascertain the mTOR molecular complex responsible for these findings, we selectively inhibited mTORC1 and mTORC2 by silencing raptor or rictor, respectively (Fig 1E and 1F). Inhibition of gene expression of raptor, but not rictor, recapitulated the enhanced CHIKV infection (Fig 1G). Confirming these results, we observed enhanced E2+ cells, increased expression of CHIKV proteins and higher extracellular viral load when cells were treated with the mTORC1 inhibitors Rapamycin or TORISEL (Fig 1H–1K). Importantly, a similar outcome was observed in cells with reduced expression of rictor (S2 Fig), demonstrating that Rapalog affected CHIKV infection independently of mTORC2. Together, these data indicate that inhibition of mTORC1, but not mTORC2, enhances CHIKV infection in vitro.
Using a complementary approach, we next evaluated the impact of mTORC1 on CHIKV infection by enhancing mTORC1 activity. This was achieved by inhibiting the expression of tuberous sclerosis 2 (TSC2) gene, a physiologic inhibitor of mTORC1 (S3A Fig). When cells were treated with tsc2 siRNA, the percentage of E2+ cells was decreased, as compared to cells treated with si-control (S3B Fig). This was validated using real time microscopy, which indicated that reduced gene expression of tsc2 significantly restricted CHIKV propagation (S3C Fig). As an additional control for potential off-target effects, we demonstrated that Rapalog exposure overcame the inhibitory effect observed in tsc2 si-RNA treated cells (S3D Fig). These results provide evidence for mTORC1 as an antiviral mechanism in the context of CHIKV infection. Following these results, we were interested to examine the possible interplay between mTORC1 and two effector pathways that were previously shown to impact CHIKV replication: type I IFN production; and autophagy initiation.
CHIKV induces rapid production of type I interferon (IFN) [16]. As several groups have reported that mTOR regulate interferon expression and mRNA translation of IFN-stimulated genes (ISGs) [17,18], we tested if mTORC1 inhibition increased CHIKV infection in a type I IFN dependent manner. We investigated the impact of Rapalog treatment in MEF deficient for both interferon regulatory factor (IRF) 3 and IRF7, two proteins essential for the production of type I IFN in CHIKV infected MEFs [16]. Remarkably, we observed that even in absence of irf3 and irf7 (irf3-/- / irf7-/-), Rapamycin or TORISEL treatment resulted in increased CHIKV infection (S4A Fig). Similar results were obtained using interferon-α/β receptor (IFNAR) deficient MEF (S4A Fig).
We also analyzed the impact of mTORC1 inhibition on NF-κB-mediated inflammation. Indeed, Rapalog treatment did not perturb the expression levels of IκBα or the activation state of NF-κB (p-p65) (S4B Fig). Moreover, NF-κB-mediated cytokines were secreted at a similar level in untreated or TORISEL-treated cells, supporting that mTORC1 inhibition did not influence NF-κB pathways (S4C Fig). To assess other potential host response pathways, we queried whether the TORISEL-mediated enhancement of CHIKV infection is dependent on host cell transcription. Infected cells were exposed to TORISEL in the presence or absence of actinomycin D (ActD). Remarkably, the fold induction of CHIKV-GFP expression stimulated by TORISEL was unaffected by inhibition of transcription (S4D and S4E Fig). Based on these data, we excluded type I IFN or other host induced immune responses as the mechanism by which mTORC1 regulates CHIKV infection.
mTORC1 is a well-known inhibitor of macroautophagy (referred to as autophagy), a bulk degradation pathway that controls clearance and recycling of intercellular constituents for the maintenance of cellular survival [19]. Previously, we and others showed that CHIKV-induced autophagy serves as a mechanism of host defense by favoring cell survival and thereby restricting viral spread [9,20]. To test the hypothesis that mTORC1 enhances CHIKV infection in an autophagy dependent manner, we assessed viral infection in MEFs deficient for key autophagy genes Atg5 (autophagy-related gene 5) or Atg7. Interestingly, autophagy deficient cells still exhibited increased infection following Rapamycin or TORISEL treatment (S5A–S5D Fig). Expression of Atg5 and Atg7 genes were also silenced in human foreskin fibroblastic cell line (HFF), with results indicating that the mTOR phenotype is indeed independent of autophagy and that the effect is not species specific (S5E–S5G Fig). Of note, while segregated from the mTOR effect, we did confirm that inhibition of autophagy genes significantly increased CHIKV infection, as previously reported [9,21].
To study the impact of rapalog treatment on in vivo CHIKV pathogenesis we used mice lacking the ability to produce type I IFNs, as wild type (WT) adult animals are resistant to severe forms of infection [22]. With the knowledge that mTOR acts independently of type I IFN expression, it was possible to employ irf3-/- x irf7-/- double deficient mice as a means to evaluate the impact of mTOR inhibition. This mouse strain is highly sensitive to CHIKV infection, with adult mice succumbing by day 6 post-infection [21]. After 8 days of intra-peritoneal injection of TORISEL (10mg/kg, injected every 2 days), irf3-/- / irf7-/-mice were infected by CHIKV and tissues were analyzed at day 1 and 2 post-infection. We first validated that TORISEL inhibited mTOR activity in the skin and muscle of treated mice (S6 Fig). Consistent with our in vitro data, mice pre-treated with TORISEL had higher viral titers in both the skin (injection site) and muscle as compared to control mice, indicating that mTOR inhibition affects in vivo viral infection (Fig 2A and 2B). As one experimental caveat concerns the immunosuppressive effects of TORISEL, we nonetheless examined T and B cell inhibition as a confounding factor in our experiments. Notably, treatment with tacrolimus, a related FKBP-interacting immunosuppressive drug, did not affect viral titers in infected tissue (Fig 2A and 2B). These observations were further confirmed using mice deficient for rag2 (rag2-/-), and therefore lacking mature lymphocytes (Fig 2C and 2D). Finally, we observed a worsening of disease in TORSIEL treated animals, and a more rapid time to death of infected mice (Fig 2E and 2F). These experiments highlight the marked outcome of mTORC1 inhibition and enhanced viral replication, which lead to exacerbation of chikungnuya disease.
Having ruled out the two expected mechanisms (type I IFN and autophagy) by which mTORC1 could regulate viral infection, we addressed the mechanism of action using an unbiased approach, interrogating the effect of mTOR inhibition on the binding, entry, replication and/or spread of CHIKV. A direct analysis of viral binding (at 4°C) using a FACS-based assay showed that Rapalog treatment did not affect CHIKV binding (Fig 3A). Similarly, after a short period of infection (2h p.i. at 37°C), no difference was observed in intracellular staining of E2 (i.e., quantification of E2 present within the input virus) when comparing Rapalog-treated and untreated cells (Fig 3B). However, quantification of CHIKV genome during the first 24h of infection, showed a higher amount of both positive strand 49S genomic and subgenomic 26S viral mRNA in TORISEL-treated cells as compare to untreated cells (Fig 3C). Notably, a kinetic assessment of the timing for which Rapalog treatment influences CHIKV infection supports the conclusion that mTORC1 has its maximum impact on the CHIKV replication step (Fig 3D–3H). Indeed, binding and entry occur during the first hours of infection, and transient Rapalog exposure 1h pre-infection (Fig 3E), or two hours post-infection (Fig 3F), did not impact the eventual rate of infection. Only mTORC1 inhibition of during the first 24h of infection resulted in increased proportion of E2 positive cells (Fig 3G), yet treatment after the initial 24h of infection showed no additional impact (Fig 3H). These results suggest that mTORC1 activity specifically target the viral replication phase.
Following the evidence for Rapalog exposure acting to increase viral replication, we investigated whether enhanced translation of nonstructural proteins (nsP) accounts for the higher viral mRNA replication. To monitor nonstructural protein translation, we used a reporter CHIKV encoding luciferase under the control of the genomic promoter (CHIKV-Luc, construct illustrated in Fig 4A). Strikingly, TORISEL exposure resulted in a 3–4-fold increase in luciferase activity for all viral doses tested (Fig 4A). Notably, these results were obtained 4h post-infection, a time point prior to the completion of CHIKV replication; thereby ensuring Luciferase activity is an accurate measure of nonstructural protein translation. These data support Rapalog treatment acting in a cell-intrinsic manner to enhance the translation of viral nonstructural proteins.
As the genomic ORF (encoding for nonstructural proteins) and subgenomic ORF (encoding for structural proteins) of CHIKV utilize different promoters, we next assessed if the inhibition of mTORC1 could also improve the translation of structural proteins. To accomplish this, we utilized a recombinant CHIKV expressing GFP under the control of the subgenomic promoter (CHIKV-GFP 5’ construct illustrated in Fig 4B), and performed hourly monitoring of infection using real time microscopy. Results confirmed that Rapalog exposure increases infection (Fig 4B and S1 and S2 Movies). Similar results were obtained by examining the percentage of GFP positive cells using cytometric analysis, showing that Rapalog treatment similarly affects the recombinant CHIKV-GFP 5’ and wild type CHIKV (Fig 4C as compared to Fig 1). Importantly, evaluation of GFP expression in infected MEF (GFP+ cells) indicated that TORISEL-treatment increased both the percentage of GFP expressing cells (7 to 35%, p-value = 0.025), and the per cell expression of GFP (8500 to 15200 MFI, p-value = 0.014, Fig 4D). We confirmed this result by using a truncated form of CHIKV, lacking the subgenomic region (rCHIKV-GFP construct illustrated in Fig 4E). This defective GFP reporter virus permitted us to monitor the translation of structural proteins at later time points. At 24h post-transfection, cells were treated with TORISEL and GFP expression was analyzed after an additional 24h (Fig 4E). While the efficiency of transfection was similar (41% vs. 39%), TORISEL-treated cells exhibited a higher expression of GFP protein as compared to untreated cells (11750 to 16450 MFI, p-value = 0,026, Fig 4E and 4F). Therefore, Rapalog treatment acts to enhance translation of both nonstructural and structural CHIKV proteins, and mediates its activity in a cell autonomous manner.
Regarding the central role for mTORC1 on the activation of cap-dependent translation, our results showed a paradoxical effect of Rapalog on CHIKV proteins translation. To determine if this effect is selective to viral proteins, we investigated the global state of host protein translation (Fig 5A). Using the SUnSET method [23], we showed that TORISEL exposure decreased the global state of host protein translation, in both uninfected and infected cells. Confirming our findings, we show that in the same experiment, TORISEL treated cells expressed higher amount of CHIKV E1 and E2 (Fig 5A). These results suggest that CHIKV translation, despite it being cap-dependent, has evolved a mechanism to bypass mTORC1 inhibition.
Our results were extended to both human cell lines as well as primary human and mouse fibroblasts. Specially, human foreskin fibroblasts (HFF), human epithelial cells line (HeLa) and primary mouse fibroblast (pri. MEF) were infected by CHIKV, which all showed a similar increase in viral load when mTORC1 was inhibited (Fig 5B). Interestingly, the enhancement of virus titer was more pronounced in primary MEF as compared to cell lines (9–10-fold increase as compared to 3–4-fold increase), indicating a role for mTORC1 in physiologic conditions.
We next investigated if a similar effect of Rapalog treatment could be observed with other viral infection. We chose to study two different viruses: sindbis (SINV), a second member of alphavirus family with a replication cycle similar to CHIKV; and influenza A (Flu, strain A/Puerto Rico/8/1934 H1N1), a member of orthomyxoviridae family. Interestingly, while Rapalog exposure enhanced CHIKV and SINV infection, no effect was observed for Flu (Fig 5C). These results suggest that different viruses have established unique strategies for modulating mTORC1 activity and/or overcoming translational stop mediated by mTOR inhibition.
Several classes of proteins use a cap-independent translation mechanism which includes the expression of an internal ribosome entry site (IRES) or IRES-like structures [24]. Indeed, IRES are often used by viruses as a means to ensure that viral protein translation is active during cellular stress or other conditions leading to mTORC1 inhibition. For both the structural and nonstructural polyprotein ORF of CHIKV, no IRES or IRES-like structures have been identified identified, suggesting that CHIKV proteins are translated via a cap-dependent mechanism [5]. To test this prediction, we investigated infection efficiency in cells silenced for eIF4E, a protein essential for the initiation of capped mRNA [25]. MEF cells were pretreated with siRNA that targeted eif4e mRNA and infection efficiency was analyzed by flow cytometry and real time imaging (S7A and S7B Fig). Reduced eIF4E expression decreased the amount of CHIKV infected cells. Similarly, inhibition of the interaction between eIF4E and eIF4G, using the inhibitor 4EGI-1, markedly limited the translation of both structural and nonstructural CHIKV protein (S7C and S7D Fig). These results demonstrated a key role for eIF4E protein in CHIKV infection and supported the prior assumption that CHIKV proteins require a cap-dependent translation mechanisms to be processed.
A Rapalog resistant mechanism is present in some tumor cells to maintain translation of capped mRNA when mTORC1 is inhibited [26,27]. This bypass mechanism requires phosphorylation of eIF4E at serine 209, increasing its binding affinity for the capped mRNA, and thereby favoring formation of translation initiation complexes [28]. To investigate the role of mTORC1 activity on eIF4E phosphorylation in our model, we first analyzed the amount of p-eIF4E in TORISEL-treated MEF (Fig 6A). As shown, cells treated with TORISEL expressed higher amount of p-eIF4E as compare to untreated cells, demonstrating that Rapalog treatment leads to increased eIF4E activation (Fig 6A). Importantly, similar results were observed at 24h post-infection, demonstrating that Rapalog exposure enhances the phosphorylation of eIF4E in uninfected and CHIKV infected cells (Fig 6A). The activation of eIF4E was directly linked to Rapalog-induced CHIKV infection by showing that si-eif4e transfected cells were resistant to Rapalog treatment (Fig 6B). These results demonstrated eIF4E activity is critical for increased CHIKV replication when mTORC1 is inhibited. Of note, despite TORISEL treatment leading to an increase in p-eIF4E in uninfected cells, the global amount of host protein translation was diminished (Fig 4A), suggesting that eIF4E activity is being preferentially co-opted by CHIKV.
MnK1/2 are the major kinases mediating phosphorylation of eIF4E [29]. We therefore asked whether TORISEL-induced phosphorylation of eIF4E was dependent on MnK proteins. This was performed using as an inhibitor for both MnK1 and MnK2 (CGP57380), which prevented the increase of p-eIF4E observed after TORSIEL treatment (Fig 6C). These results are consistent with prior reports of Rapalog treatment enhancing eIF4E phosphorylation via the activation of MnKs [28,30–32]. Importantly, using the recombinants CHIKV-GFP 5’ or CHIKV-luciferase, we demonstrated that infection and/or translation of CHIKV proteins were not influenced by TORISEL exposure in cells pretreated with MnKs inhibitor (CGP57380 or MnK1/2 inhibitor II) (Fig 6D and 6E). These findings indicate that mTORC1 inhibition favors CHIKV protein translation by an increased MnK-dependent phosphorylation of eIF4E.
We next studied activators of MnK1/2 in order to define the molecular pathway by which eIF4E is engaged. Strikingly, neither of two known kinases responsible for MnK activation, mitogen-activated protein kinase kinase (MAPKK also known as MEK) and p38 MAPK, were required for the enhanced phosphorylation of eIF4E or for the increase in CHIKV proteins translation (Fig 6F and 6G). Cross talk between the phosphatidylinositol-3 kinase (PI3K) and MnKs signaling has been previously reported in human cancer cells [32]. In the context of viral infection, we show that the PI3K inhibitor LY294002 blocked the Rapalog-induced activation of eIF4E and the enhancement of CHIKV (Fig 6F and 6G), suggesting that mTOR inhibition increases eIF4E phosphorylation and subsequently the CHIKV infection through a PI3K-dependent and MnK-mediated mechanism.
To investigate potential co-regulation of mTORC1 among the different effector pathways, we analyzed the role of S6K1 and 4E-BP1. Notably, silencing of these respective genes did not impact Rapalog-induced CHIKV infection (Fig 6H and 6I). Together, these data indicate that mTORC1 has a direct impact on PI3K and the MnK/p-eIF4E pathway.
To demonstrate the physiologic relevance of our discovery, we investigated the role of CHIKV-mediated inhibition of mTORC1 on infection. Interestingly, inhibition of mTORC1 correlated with an increase of eIF4E activity, with peak phosphorylation occurring at 6 h post-infection (Fig 7A). These data advances our previous report of a rapid and transient inhibition of mTORC1 during CHIKV infection, regulated by CHIKV-mediated ROS production and AMPK activation [9]. We also demonstrate that preventing CHIKV-mediated inhibition of mTOR, using ROS inhibitor (N-acetyl-L-cysteine), abrogated the enhanced p-eIF4E (Fig 7B). Moreover, inhibition of PI3K or MnKs, using respective inhibitors, abolished the CHIKV-mediated phosphorylation of eIF4E (Fig 7C). Together, these results show that CHIKV infection increases the phosphorylation of eIF4E through a PI3K and Mnk1/2 dependent.
Finally, to define the impact of phosphorylation of eIF4E on natural CHIKV infection, we tested the direct impact of MnK or PI3Ks inhibitors on viral protein expression and viral spread (Fig 7D–7F). Interestingly, inhibition of either Mnk or PI3K significantly decreased the percentage of E2 positive cells (Fig 7D), the translation of structural (Fig 7E) and nonstructural proteins (Fig 7F). These data support our conclusion that CHIKV-induced activation of eIF4E favors infection during natural infection, and also explains the enhancement of viral infection when mTORC1 is pharmacologically inhibited by Rapalog treatment (Fig 8).
Herein, we evaluated the role of the metabolic regulator mTOR during CHIKV infection. To our surprise, inhibition of mTORC1 enhanced viral protein translation via a mechanism that is independent of autophagy and type I IFN signaling. Being an important initiator of cap-dependent protein translation, the inhibitory effect of Rapalog treatment on viral has been well documented in instances where the virus possesses capped mRNA [14,33]. The results from our current study, however, identify an unexpected effect of mTORC1. During natural CHIKV infection, mTORC1 limits viral replication despite CHIKV requiring a host mRNA cap. This was further evident when cells were exposed to Rapalog, which resulted in a massive induction of CHIKV protein expression. Indeed, our study provides the first evidence for direct enhancement of viral protein translation during mTORC1 inhibition, thus revealing a new strategy developed by CHIKV to maintain viral protein expression in the context of cell stress.
As a metabolic sensor, mTOR activity is perturbed during viral infection [12]. To ensure viral protein translation, viruses have evolved strategies to bypass cellular programs that limit the ribosomal machinery. Regulation by mTORC1 has been a major focus and prior work has illustrated the different mechanisms evolved in the context of the host / microbe détente. For example, HSV-1 enhances mTORC1 activity; whereas Poliovirus, HIV-1, Sindbis and CHIKV inhibit this complex [12]. For viruses inducing a stop in protein translation, they have means to ensure translation of their own mRNAs. Most strategies identified thus far involve the use of non-canonical translation mechanisms, including IRES, ribosome shunting or VPg [24]. These processes permit efficient viral replication in absence of the cap-dependent translation initiation protein eIF4E.
To demonstrate that CHIKV proteins use a classical cap-dependent process, we analyzed the dependence on the active form of eIF4E. Indeed, absence of eIF4E expression lost of eIF4E phosphorylation or abrogation of eIF4E/eIF4G complex formation led to a drastic inhibition in CHIKV infection (Figs 7 and S7). Thus CHIKV has evolved a newly discovered mechanism of bypassing mTORC1 inhibition. Notably, we show that a similar strategy seems to be employed by Sindbis (Fig 5C)–consistent with its replication requiring an active form of eIF4E [18]–but not by influenza A (Fig 5C), whose infection proceeds normally even when eIF4E is functionally impaired [34].
To achieve translation initiation, CHIKV engages an MnK-dependent hyper-phosphorylation of eIF4E. This mechanism has been reported in tumors, providing a means to support cap-dependent protein translation under conditions of cell stress [30,31]. In our model, however, hyper-phosphorylation of eIF4E does not restore translation of host proteins, which is shutdown after Rapalog exposure (Fig 5A). This defines a selective strategy by CHIKV to enhance its viral protein translation via p-eIF4E, overcoming mTORC1 inhibition of conventional cap-dependent pathways.
The phosphorylation of eIF4E and the formation of the eIF4E/eIF4G complex are tightly regulated by signal transduction pathways that converge on MnK1/2 and mTOR [35]. While both pathways are known to positively impact translation initiation, interconnections between the two signaling mechanisms have remained unclear. Several investigations suggested that the MnKs are activated by Rapalog treatment, resulting in the maintenance of cap-dependent translation during inhibition of mTORC1 [30,31], however the molecular scaffolds are not known.
In the present study, we demonstrated that activation of MnK1/2, induced by Rapalog or CHIKV-mediated mTOR inhibition, results in upregulation of viral protein translation via the hyper-phosphorylation of eIF4E. We show that translation of CHIKV proteins are not only resistant to mTORC1 inhibition, but can be enhanced by this process via the activation of PI3K and the subsequent engagement of the MnK/eIF4E pathway. The effect of mTOR on CHIKV infection was independent of both S6K1 and 4E-BP1 signaling. These observations align with previous work published by Wang and colleagues, who reported that Rapamycin-induced activation of MnK is independent of S6K1 activity [32]. 4E-BP1 hypo-phosphorylation increases its binding to eIF4E and prevents the formation of eIF4F complex [10]. While we clearly show that Rapalog treatment induces hypophosphorylation of 4E-BP1, eIF4E-dependent translation is increased (Fig 1). Interestingly, in our model, PI3K inhibition was able to restrict CHIKV infection and limit the effect of Rapalog on virus replication. Since the classical downstream signals of mTORC1 (S6Ks and 4E-BPs) and upstream signal of MnKs (MEK and p38) are not involved in the Rapalog effect, molecular connections between mTOR, PI3K and the MnK pathway must be examined further.
In our previous report, we described a transient inhibition of mTORC1 pathway during the first hour of infection that led to an induction of autophagy that in turn limits CHIK-induced apoptosis [9]. In infected cells, mTORC1 is inhibited through the intrinsic production of ROS leading to the activation of both AMPK and the TSC1/TSC2 complex [36]. Herein, we confirmed the transient inhibition of mTORC1 during CHIKV infection and discovered a new function of mTOR as a direct regulator for viral protein translation. Indeed, preventing mTOR-dependent hyper-phosphorylation of eIF4E using MnK or PI3K inhibitors significantly decrease infection efficiency (Fig 7). Integrating our new findings, we suggest that CHIKV-mediated inhibition of mTORC1 benefits the virus by increasing the efficiency and timing of translation, via the activation of eIF4E and autophagy respectively.
Strikingly, while ROS inhibition prevents the transient inhibition mTORC1 (Fig 7B), we did not observe a significant difference for CHIKV infection in ROS inhibitor-treated versus untreated cells (S8 Fig). These results could be explained by the fact that ROS itself has an mTOR-independent antiviral effect on CHIKV infection, as illustrated by ROS inhibition leading to a marked increase of infection in siRNA mtor or Rapalog treated cells (S8 Fig). Having this new knowledge, we postulate that inhibition of mTORC1 balances the antiviral effect of CHIKV-mediated ROS production.
mTOR has been considered in other infectious models as important for host response. Pulendran and colleagues showed that the TLR9 ligand CpG-A triggered phosphorylation of mTOR and its downstream targets 4E-BPs and S6Ks [37]. Accordingly, TLR9-mediated production of type I IFN, tumor necrosis factor (TNF) and interleukin 6 (IL-6) was suppressed in human and mouse plasmacytoid dendritic cells (pDC) treated with Rapamycin [37]. Another strategy by which mTOR has been associated to immune antiviral response relates to its cross-talk with the autophagy pathway. Indeed, several viruses have developed strategies to use autophagososmes as a membrane support for viral replication and/or release, including HCV, poliovirus and HIV-1 [38,39]. In this context, inhibition of autophagy mediated by mTORC1 could limit viral infection.
Our results highlight a new antiviral mechanism for mTOR that is independent of IFN production and autophagy. We showed that inhibition of mTORC1 increased CHIKV infection in both in vitro and in vivo models. This process was based on direct regulation of viral protein translation and is independent of new transcriptional activity, as the effect of Rapalog treatment was unaffected by ActD exposure. Importantly, we also confirmed the impact of mTOR on CHIKV infection by increasing the activity of endogenous mTORC1. These results suggest that strategies aimed at enhancing activation of mTOR (e.g., specific diets or insulin injection) may be a means of controlling CHIKV infection.
In sum, we have provided evidence for a novel and unexpected mechanism by which CHIKV adapts to mTORC1 inhibition. In the context of acute CHIKV infection, eIF4E is important for the translation of viral proteins, and CHIKV-mediated mTORC1 inhibition increases the phosphorylation of eIF4E, thus favoring viral replication. These data also suggests that targeted engagement of upstream activator of mTORC1 (e.g., Insulin receptor or Akt) or blocking inhibitors of the mTORC1 pathway (e.g., TSC2) may constitute useful strategies for limiting the pathogenesis of acute Chikungunya disease.
Wild-type mouse embryonic fibroblasts (MEF) were obtiained from the Korsmeyer laboratory (Farber, Boston, MA, USA). Atg5-/- and atg7-/- MEF were generous gifts of the Kroemer Laboratory (INSERM U848, Institut Gustave Roussy, Villejuif, France). Primary human foreskin fibroblasts (HFF) were obtained from America Type Culture Collection. All cell lines were mycoplasma free and maintained at 37°C in humidified atmosphere containing 5% CO2 in medium supplemented with 10% heat inactivated fetal calf serum, 100 μg/ml penicillin (Invitrogen); 100 U/ml streptomycin (Invitrogen), and MEM nonessential amino acid (Invitrogen). irf3-/- / irf7-/- mice were generated by Michael Diamond, with the original strains being provided by Tadatsugu Taniguchi.
The preparation of CHIKV from clinical samples has been previously described (Schuffenecker et al., 2006). CHIKV-21 strain was propagated in C6/36 cells and supernatants were harvested and frozen at -80°C before titration and further use. Recombinant CHIKV expressing GFP under the subgenomic promoter (CHIKV-GFP 5’) was generated using a full-length infection cDNA clone provided by S. Higgs (Vanlandingham et al., 2005). Recombinant CHIKV expressing Luciferase under genomic promoter (CHIKV-Luc.) was a gift from Philippe Despres and was generated as previously described [40]. The CHIKV replicon expressing EGFP (rCHIKV-GFP) was a gift from Gorben Pijlman and was generated as previously describe [41]. MEF or HFF cells (plate at ∼ 50% confluence in 24- or 96-wells plates) were exposed to the indicated viruses for 2h at 37°C, extensively washed with PBS and cultivated for various periods of time in presence of drugs before further analysis. The MOI was defined as the amount of CHIKV infectious units (calculated on BHK cells as PFU) per target cell. In indicated experiments, Rapamycin (100 nM—Calbiochem), TORISEL (0.1 mg/ml—Biovision), PP242 (0.1 μM—Euromedex), 4EGI-1 (25 μM–VWR International), CPG57380 (20 μM–R & D Systems Europe), MNK inhibitor-II (5 μM–Merck Chemicals LTD), PD 0325901 (1 μg/ml–Sigma Aldrich), SB203580 (1 μM–Sigma Aldrich) or LY294002 (25 μM—Ozyme) were used.
Smartpool siRNA targeting atg5, atg7, mTOR, raptor, rictor, eIF4E, S6K1, 4EBP-1 and control siRNA were from Dharmacon (Perbio, Berbères, France). MEF or HFF cells (0.1 × 106) were cultured in 6-well plates for 1 day in OptiMEM (Invitrogen) containing 10% FCS and transfected with 30 nM of indicated siRNA using lipofectamine RNAiMAX (Invitrogen). For all experiments, CHIKV infection was performed after 3d siRNA incubation. In all experiments, protein expression of targeted gene was confirmed to be knocked down to <90% of WT expression. Where inducated functional inhibition was evaluated.
Lysates were prepared in 1x Dulbecco’s Phosphate Buffer Saline (DPBS—Invitrogen) containing 1% Nonidet P 40 substitue (NP40 –Signa-Aldrish, MO, USA) and protease inhibitor cocktail (Roche Diagnostics, IN, USA). Total protein was determined by Lowry’s method and 25 mg was loaded on a 4–12% gradient SDS–polyacrylamide gel electrophoresis (Invitrogen). Proteins were transferred to 2 μM nitrocellulose membrane using the Trans-blot turbot kit (Bio-Rad) and blotted over-night with anti-mTOR (rabbit polyclonal, abcam), anti-Rictor (rabbit polyclonal, Cell Signaling), anti-Raptor (rabbit polyclonal, Cell Signaling), anti-Atg5 (mouse monoclonal, Cell Signaling), anti-Atg7 (mouse monoclonal, Cell Signaling), anti-pS6K1 (rabbit polyclonal, Abcam), anti-S6K1 (rabbit polyclonal, Abcam), anti-peIF4E (Ser209) (rabbit polyclonal, Cell Signaling), anti-eIF4E (rabbit polyclonal, Cell Signaling), anti-E1/E2 (rabbit polyclonal, gift from Olivier Schwartz laboratory), anti-C (monoclonal antibody, gift from Olivier Schwartz laboratory), anti-GFP (rabbit polyclonal, Cell Signaling) or anti-GAPDH (rabbit polyclonal, Cell Signaling). Secondary HRP-coupled Abs was detected using ECL Plus (Amersham Pharmacia Biotech).
MEF were infected with CHIKV-21 at indicated MOI in presence of TORISEL. After 8h and 24h of infection, cells were washed twice in PBS and RNA was TRIzol (Invitrogen) extracted, quantified and diluted to the same concentration. Samples were prepared in NorthernMax formaldehyde loading dye (Ambion) with 1μl of ethidium bromide, heated to 65°C for 10 minutes, then separated on a 1.2% agarose (Lonza) gel containing 1x morpholinepropanesulfonic acid (MOPS), running buffer (Ambion) and 6.7% formaldehyde. RNA was transferred onto nitrocellulose membrane, cross-linked by ultraviolet irradiation (UVP), and prehybridized at 68°C for 1h in ULTRAhyb ultrasensitive hybridization buffer (Ambion). A plasmid used for the expression of CHIKV RNA probes corresponding to the 3′ portion of the E2 glycoprotein was generated by first amplifying the region of the CHIKV genome from 8703 (5′-GAAGCGACAGACGGGACG-3′) to 9266 (5′-GTTACATTTGCCAGCGGAA-3′) by PCR and subsequently TOPO-TA cloning the PCR product into the pCRTOPO-II vector. RNA probes complementary to positive strand RNA were labeled with 32P using the MAXIscript SP6 In Vitro Transcription Kit (Ambion), unincorporated nucleotides were removed using illustra MicroSpin S200 HR columns (GE healthcare), and probe was hybridized to the membrane overnight at 68°C. Membranes were washed several times at 68°C with 0.1× SSC with 0.1% SDS, then imaged using Amersham Hyperfilm MP autoradiography film (GE Healthcare).
MEF or HFF cells were infected with CHIKV-21 or CHIKV-GFP 5’ at the indicated MOI for 24h and fixed with 4% PFA for 20 min. After fixation, cells infected with CHIKV-21 were permeabilized with BD Cytofix/Cytoperm (BD kit, BD Bioscniences) before labeling with anti-E2 (gift form Lecuit lab, Microorganismes et barrières de l’hôte, Institue Pasteur, France). The percentage of E2+ or GFP+ cells was measured by flow cytometery using FACSCanto (BD Biosciences, MD, USA) and FlowJo software (Tree Star, Inc.).
MEFs were infected with CHIKV-GFP 5’ in 24-well or 96-well pate and imaged using IncuCyte HD system (Essen BioScience). Frames were captured at 1-hour intervals from 4 separate 950 × 760–μm2 regions per well using a 20× objective. Cultures were maintained at 37°C in a Hera cell 240 chamber (Thermo Electron Cormpration) throughout, with all experiments run in triplicate. GFP+ cells were counted using IncuCyte ZOOM software (Essen BioScience) and results are represented as green object count per mm2. Values from all 4 regions of each well were pooled and averaged across the 3 replicates. Movies were extracted directly from IncuCyte ZOOM software.
MEF or HFF cells were infected with CHIKV-21 and supernatants were recovered at 24h after infection. Viral samples were titrated as TCID50 endpoint on Vero cells using a standard procedure. Serial 10-fold dilutions (100 μgl) of supernatants were added in six replicates in 96-well plates seeded with 104 Vero cells. The cytopathic effect was scored 5 days after infection and the titers was calculated by determining the last dilution giving 50% of wells with cells displaying a cytopathic effect. Results were expressed as TCID50/ml.
irf3-/- / irf7-/- mice were treated with intra-peritoneal injection of 100 μL solution containing TORISEL (10mg/kg) (n = 32) or PBS (n = 31) every two days for 8 days. At day 8, mice were infected with 1x106 PFU CHIKV-21 subcutaneously (s.c.) in the bottom chest. For viral titration, skin and muscle were collected after days 1 (n = 6 for PBS-treated mice; n = 7 for TORISEL-treated mice), 2 (n = 5 for PBS-treated mice; n = 6 for TORISEL-treated mice), and 3 (n = 6 for PBS-treated mice; n = 5 for TORISEL-treated mice) of infection, homogenized, and viral samples were titrated as TCID50 endpoint on Vero cells using a standard procedure. Clinical score (n = 14) was determined at day 2 of infection, based on EAE from K. Racke (0 = nothing; 1 = limp tail; 2 = mouse don’t grasp the cage with toes but with the ankle; 3 = mouse is unable to return and land on its feet when flipped over; 4 = hindlimb drag behind are not used by the mouse for movement; 5 premoribund stat). Lethality of mice (n = 14) was followed for 10 days post-infection.
SunSet experiments were performed as previously described [23]. To summarize, cells were infected with CHIKV in the presence of TORISEL for 24 hrs. Then cells were cultivated with serum free media containing puromycin (10 υg/ml) for 30 min, washed with PBS and kept in culture for an additional 1h with normal media. Western blot was performed as previously described using an antibody against puromycin (clone 12D10, 1/5000, Millipore).
Mouse studies were performed in strict accordance with the Institutional Guiding Principles for Biomedical Research Involving Animals and all experiments were performed in an A3 containment facility. The protocols were approved by the Institutional Committees on Animal Welfare of the Pasteur Institute (OLAW assurance #A5476-01). All efforts were made to minimize suffering.
|
10.1371/journal.pntd.0006968 | Safety and efficacy of the rSh28GST urinary schistosomiasis vaccine: A phase 3 randomized, controlled trial in Senegalese children | Urinary schistosomiasis, the result of infection by Schistosoma haematobium (Sh), remains a major global health concern. A schistosome vaccine could represent a breakthrough in schistosomiasis control strategies, which are presently based on treatment with praziquantel (PZQ). We report the safety and efficacy of the vaccine candidate recombinant 28-kDa glutathione S-transferase of Sh (rSh28GST) designated as Bilhvax, in a phase 3 trial conducted in Senegal.
After clearance of their ongoing schistosomiasis infection with two doses of PZQ, 250 children aged 6–9 years were randomized to receive three subcutaneous injections of either rSh28GST/Alhydrogel (Bilhvax group) or Alhydrogel alone (control group) at week 0 (W0), W4, and W8 and then a booster at W52 (one year after the first injection). PZQ treatment was given at W44, according to previous phase 2 results. The primary endpoint of the analysis was efficacy, evaluated as a delay of recurrence of urinary schistosomiasis, defined by a microhematuria associated with at least one living Sh egg in urine from baseline to W152. During the 152-week follow-up period, there was no difference between study arms in the incidence of serious adverse events. The median follow-up time for subjects without recurrence was 22.9 months for the Bilhvax group and 18.8 months for the control group (log-rank p = 0.27). At W152, 108 children had experienced at least one recurrence in the Bilhvax group versus 112 in the control group. Specific immunoglobulin (Ig)G1, IgG2, and IgG4, but not IgG3 or IgA titers, were increased in the vaccine group.
While Bilhvax was immunogenic and well tolerated by infected children, a sufficient efficacy was not reached. The lack of effect may be the result of several factors, including interference by individual PZQ treatments administered each time a child was found infected, or the chosen vaccine-injection regimen favoring blocking IgG4 rather than protective IgG3 antibodies. These observations contrasting with results obtained in experimental models will help in the design of future trials.
ClinicalTrials.gov NCT 00870649
| Vaccines represent an attractive tool in the fight against schistosomiasis. Pre-clinical immunization studies with the schistosome enzyme 28 kDa glutathione S-transferase (28GST) have shown a significant reduction of schistosome egg production and subsequent pathology. The objective of the present phase 3 trial was to assess the efficacy and safety of vaccination with the recombinant 28GST of Schistosoma haematobium (rSh28GST) named Bilhvax, in infected school children. After Praziquantel treatment before inclusion and three administrations of rSh28GST at one month interval as primo-vaccination, subjects received a boost injection one year after the first administration. The efficacy was evaluated as a delay of recurrence of urinary schistosomiasis. While immunological analysis showed that Bilhvax induced a consistent immune response characterized by antibodies able to inhibit 28GST enzymatic activity, the efficacy endpoint was not reached. This lack of significant effect may be due to the negative conjunction of a too challenging recurrence criterion associated with safety measures ensuring repeated PZQ treatment. The control of these main factors will be essential for the subsequent trials and must provide evidence of the Bilhvax efficacy as a safe vaccine against uro-genital schistosomiasis.
| Schistosomiasis is a chronic parasitic disease caused by trematodes that lay eggs in the urinary or gastrointestinal tract blood vessels [1]. It is associated with gastrointestinal or genitourinary disorders, pain, anemia, malnutrition, fatigue, and reduced exercise tolerance. These effects imply a loss of performance in parasitized individuals, especially schoolchildren, that hampers personal and community development [2]. In addition to the possible lethal outcome of the infection, the physical disability and social discomfort caused by schistosomiasis are tremendous, and meta-analyses have estimated that the current disease burden may exceed 70 million disability-adjusted life years [3].
Although 260 million people are infected by different Schistosoma species, and more than 200,000 deaths per year are registered, schistosomiasis remains a neglected disease [4]. Fewer than 40 million of those infected have received the unique drug available, praziquantel (PZQ), which has several limitations, including a lack of effect on reinfection and increased risk for emergence of drug-resistant parasites. This absence of a long-term efficient treatment emphasizes the need to develop a safe and efficacious vaccine that can be integrated into the control strategies for reducing schistosomiasis transmission and reinfection [5].
Closely associated with the parasite metabolism, the 28-kDa glutathione S-transferases (P28GSTs) have been identified in schistosomes as potent modulators of epithelial Langerhans and dendritic cell migration during infection [6], hormonal carriers for schistosomes [7], and the main enzymes involved in detoxification and antioxidant pathways [8]. Schistosome P28GSTs are potential vaccine candidates and have been extensively studied in various experimental models [1]. In addition to P28GST from Schistosoma mansoni (Sm) Sm28GST, P28GST from Schistosoma haematobium (Sh) Sh28GST has been further characterized, from molecular cloning to crystallization [9], and developed as a schistosome vaccine in non-human primates [10]. Indeed, Sh28GST can significantly reduce Sh worm fecundity in experimentally infected primates [10]. In addition, the combination of PZQ chemotherapy with 28GST DNA vaccination has been assessed in the mouse, triggering an enhanced specific immune response and decreased schistosomiasis pathology [11]. A phase 1 clinical trial conducted in healthy subjects demonstrated that the recombinant Sh28GST (rSh28GST) adsorbed to Alhydrogel did not induce significant toxicity in healthy adults and generated a Th2-type immune response characterized by cytokine and antibody profiles [12]. Phase 2 clinical testing showed that Bilhvax in combination with PZQ treatment was safe for infected adults and children (the Bilhvax program) [13] (Riveau et al, in preparation). More than 80% of the vaccinees included in these phases (adults 18–30 years, children 6–9 years) had a specific immune response following two administrations of Bilhvax at one-month intervals.
Here, we describe the results of a phase 3 trial of rSh28GST adjuvanted with Alhydrogel (Bilhvax). This phase 3 trial was designed to investigate the safety, efficacy, and immunogenicity of Bilhvax administered to infected Senegalese schoolchildren. In this clinical trial, Senegalese Sh-infected children aged 6 to 9 years were first cleared of their current schistosomiasis infection by a double PZQ treatment before receiving three injections of Bilhvax at one-month intervals, and then PZQ treatment before a booster one year after the first injection. Immediate and delayed tolerance as well as efficacy and immunogenicity of Bilhvax were studied.
The main objective of this randomized controlled trial was to show that co-administration of Bilhvax with PZQ could delay the risk of Sh clinical recurrence during the 3 years following vaccine administration to Sh-infected children living in an endemic area. The safety was monitored in these infected children according to clinical evidence, whereas immune response was followed in both Bilhvax and control groups by determining specific antibody titers as well as neutralizing antibodies.
The trial protocol was approved and tial twice audited by the Senegalese Ethical Committee (Comité National d’Ethique de la Recherche en Santé; CNERS, Dakar, Senegal; Registration number SEN 14/08). The study was conducted in accordance with the Declaration of Helsinki III and with the International Ethical Guidelines for Biomedical Research Involving Human Subjects, as laid down by the Council for International Organizations of Medical Sciences in collaboration with the World Health Organization and the Good Clinical Practice guideline CPMP/ICH/135/95 [14–16]. Written informed consent was obtained from all parents or guardians prior to enrolment.
The trial was overseen by Inserm (Institut National de la Santé et de la Recherche Médicale, France) et WHO (NTD Department). This study is registered with ClinicalTrials.gov, number NCT 00870649
This randomized, parallel-group, controlled, double-blind phase 3 trial of the schistosome vaccine candidate Sh28GST (Bilhvax) was conducted at the Biomedical Research Center EPLS (Espoir Pour La Santé) and included 250 children living in 13 villages of the Saint-Louis region located in the lower Senegal River basin. The regional prevalence of urinary schistosomiasis in schoolchildren is estimated at over 60%. The school selection took place in two areas of the river valley, the Lampsar area (zone 1, from Saint Louis to Ross-Béthio) and the Djoudj area (zone 2, from Ross-Béthio to Ronkh). Zone 1 included 20 villages representing a total of 23 public schools and 20 Koranic schools (8237 children aged 5 to 14). Zone 2 included 10 villages with 10 public schools and 5 Koranic schools (2941 children). Mass treatment with PZQ and albendazole was performed in these 60 schools by the program, representing 11,178 children aged 5 to 14 treated before the start of the clinical trial.
Only Zone 1 was used for selection. In this area, 19 schools were selected based on the following criteria: presence of a school; agreement of the village chief; number of children >40 in the youngest classes (class CI and CP); more than 5 children with hematuria; village accessible during the rainy season; teacher collaboration; and no medical research intervention planned in the village. Among these 19 villages, 13 villages were selected, including a population of 2150 children corresponding to the age criterion. This population had a Sh prevalence of 44% with 692 children with hematuria >2+ (32%), and with egg load >50 eggs/10mL (16%). A total of 298 children were assessed for eligibility.
The trial was designed to evaluate the safety, immunogenicity, and efficacy of Bilhvax for prevention of clinical and parasitological recurrences of Sh infection during a 152-week follow-up period after the first injection in a population of schoolchildren aged 6–9 years. The primary efficacy endpoint was time to first recurrence of pathology due to Sh infection, anticipating a significant delay of first recurrence between vaccine and control groups during the 3-year period (W0/V1 to W152/V11). Secondary outcome measures were safety and immune response evaluation.
Written informed consent was obtained 8 W prior to randomization (W-8) from the children’s parents or guardians. The inclusion and exclusion criteria for participation in the trial are listed in Table 1.
Each included participant was seen at 13 visits: 2 visits of pre-inclusion (Vp), at 9 weeks (W-9) and 8 weeks (W-8) prior randomization (respectively Vp1 and Vp2), the inclusion visit (W0), and 10 visits from W4 (visit V2) to W152 (visit V11) at 4, 8, 44, 52, 65, 82, 100, 117, 134, and 152 weeks after inclusion (Fig 1).
From the pediatric population living in 13 villages of the urinary schistosomiasis hyperendemic region of the Lower Senegal River Valley, 298 children were preselected according to the two criteria for Sh infection: medium or high microhematuria (≥2+) and a minimum load of 50 eggs/10 ml of urine. Microhematuria was evaluated using a strip reader (Dipstick Multistix 8 SG, Bayer, Siemens/reader Clinitek Status) as “negative,” “trace,” “1+” (corresponding to 25 red blood cells (RBCs)/mm3), “2+” (corresponding to 80 RBC/mm3), and “3+” (corresponding to 200 RBC/mm3). A positive microhematuria was defined as urine coded ≥1+. Urine samples were collected in the morning over the course of 2 hours.
To clear their ongoing schistosomiasis infection, preselected children were treated twice with PZQ (40 mg/kg) at W-9/Vp1 and W-8/Vp2 prior to randomization. Participants were assigned randomly at W0/V1 in a 1:1 ratio to control or vaccine group. The “Bilhvax group” received three subcutaneous injections of rSh28GST formulated with Alhydrogel as adjuvant at W0/V1, W4/V2, and W8/V3, followed by a vaccine boost at W52/V5. The control group received four subcutaneous injections of Alhydrogel at W0/V1, W4/V2, W8/V3, and W52/V5. Both groups received one dose of 40 mg/kg PZQ at W44/V4, i.e., 8 weeks before the booster injection (Fig 1).
Batches of rSh28GST were produced and purified from recombinant Saccharomyces cerevisiae culture (TGY73.4—pTG8889 strain) under Good Manufacturing Practice (GMP) conditions by Eurogentec S.A. (Belgium). The rSh28GST clinical batch (M-BIX-P01/189a) was lyophilized under GMP conditions by Miltenyi (Germany) at 253 μg (±10%) per vial. The lyophilized preparation was re-suspended extemporaneously using 1 ml of apyrogenic and sterile alum solution 0.2% (Al2O3 0.2%; Al(OH)3 3%; NaCl 9 g/L; ammonium carbonate buffer 10 mM, pH7.8) (Alhydrogel from Superfos, Denmark; batch #14093) and administered in a volume of 0.4 ml.
Adverse events (AEs), signs, and symptoms were defined and classified according to the DMID Safety Reporting and Pharmacovigilance for pediatric toxicity [17]. For general adverse effects, the prevalence and intensity of the following signs and symptoms were assessed: abdominal pain, vomiting, nausea, diarrhea, headache, sleepiness, fever, vertigo, and pruritus. Occurrence of local AEs at the injection site (pain, pruritus, or swelling) or any regional adverse reaction to vaccine or placebo injections was recorded 1 h after each vaccination during visits V1, V2, V3, V4 (PZQ treatment), and V5. Medical exams were also performed 4 h and 24 h after vaccinations during visits V3, V4, and V5. A questionnaire was given to parents to report whether their children felt any AE or needed any medical assistance at 48 h and 72 h after V1, V2, V3, V4, and V5. Parents of included children could be in contact (7/7; 24/24) with the medical team for any illnesses anytime over the course of the trial.
The recurrence of Sh infection was defined as a positive microhematuria (≥1+) accompanied by the presence of at least one living egg in 10 ml of urine. Hematuria was tested during the follow-up visits (V7, V8, V9, V10, and V11) or following spontaneous complaint of the subject during inter-visit periods.
Positive hematuria (≥1+) was considered as not due to schistosomiasis when three consecutive urine filtrations (UFs) performed on three different days during one week were negative. When hematuria was established using the urinary strip, UFs were systematically performed with the urine collected. When UF was positive, the hatching assay was performed with the eggs contained in the remaining urine sample to evaluate their viability (expressed in % hatching).
At any time, if an individual complaint suggested urinary disorders, clinical examinations as well as parasitological tests were immediately performed. When Sh recurrence was established in a participant, an evaluation of Sm in stool sample using the Kato-Katz assay (KK) was performed [18, 19] because concomitant Sh and Sm infection is endemic in the selected villages.
Two-hour urine samples were collected during the morning at the selection (before inclusion) and then every 4 months during the scheduled follow-up visits (from W82/V7 to W152/V11). Parasitological evaluation was carried out when the urine sample was positive for hematuria (≥1+). Two samples of 10 ml each were separately filtrated and filters were analyzed by two different readers as described elsewhere [20]. In case of high discrepancy between the two counts, both filters were examined by a third reader, and if necessary, the filtration was repeated. At the last visit (the closing visit) (W152/V11), UF and KK were performed for each individual.
The hatching assay was carried in duplicate (2 wells on two separate 12-well plates) with the number of eggs close to 100 eggs per well. If the number of eggs collected proved insufficient, a single reaction well was completed and read by two different technicians. If the number of eggs counted in the reaction wells was less than 50, the percentage of hatching was not calculated, and only the presence or absence of hatched eggs was noted [21].
When Sm eggs were observed in urine (UF) and the KK test was negative, two consecutive stool samples were analyzed.
The aim of the vaccine efficacy was to reduce the risk of S. haematobium pathology recurrences over the course of 3 years following vaccine administration in children exposed to urinary schistosomiasis. The primary endpoint of efficacy was time to first recurrence of pathology due to Sh infection, anticipating a significant delay of first recurrence between vaccine and control groups during the 3-year period from D0 (V1) to W152 (V11) (intention-to-treat (ITT) population) between the vaccine and control groups. The difference in delay to the onset of recurrence between both groups was also evaluated as the modified ITT (mITT) population considering the period post vaccination, V6 to V11.
Secondary efficacy endpoints were defined as the percentage of participants without recurrence, the number of recurrences per subject, and parasitological manifestations at the first recurrence including the number of viable eggs, the number of hatched eggs, and percent of eggs hatching.
Individuals were systematically treated with PZQ each time schistosomiasis infection/recurrence was identified.
Specific anti-rSh28GST antibodies were measured by ELISA in individual sera at W0/V1 (background), W44/V4 (after three immunizations), W52/V5 (after PZQ treatment), W65/V6 (after booster), W117/V9, and W152/V11 according to Riveau et al [12]. Results are expressed as antibody titers. Titers were defined as the highest dilution yielding an absorbance two or three times above the negative control depending on the studied isotype. (wells containing the reference control negative pool of Senegalese sera instead of participant sera in the same plate). Individuals were considered as positive responders when the antibody titer was greater than the threshold defined as 3-fold the standard deviation above the mean titer value of all individuals at W0. Specific antigen (rSh28GST 10 μg/ml) was coated on 96-well plates (Nunc-immuno plate, F96 cert., Maxisorp, Roskilde Denmark) for 2.5 h at 37°C. After blocking with phosphate-buffered saline containing 0.5% gelatin (Merck, Darmstadt, Germany), serial dilutions of individual sera were added, and plates were incubated overnight at 4°C. Specific biotinylated monoclonal antibodies to human immunoglobulin (Ig) isotypes (BD Pharmingen G(h+l), G2, G4, A1A2, and E; Sigma G3; SB G1) were added (1.5 h at 37°C) at a 1/2000 dilution for total IgG; 1/4000 for IgG1 and IgG3; 1/3000 for IgG2, IgG4, and IgA1/A2; and 1/500 for IgE detection. IgM were not measured due to the lack of adequate reagent and cross reactivity with other immunoglobulin. Peroxidase-conjugated streptavidin (1/20000; 30 min at 37°C) was then added (SPA-BIOSPA, Milano). Colorimetric development was performed with ABTS [Sigma, liquid substrate 2.2’-azino-bis (3-ethylbenz-thiazoline-6-sulfonic acid)], and absorbance (OD) was measured at 405 nm (Reader BioTek- EL808). At each step of the ELISA procedure, the plate washing was performed using the plate washer (BioTek ELx405). Titers were defined as the highest dilution yielding an absorbance two or three times above the negative control depending on the studied isotype.
In addition to the titration of specific anti-rSh28GST antibodies, functional aspects of these antibodies were investigated by following inhibition of rSh28GST enzymatic activity by sera from each individual. The glutathione S-transferase (GST) neutralizing capacity of antibodies was evaluated as previously described [12]. Briefly, 20 μl of rSh28GST solution (4 μg/ml in 50 mM potassium phosphate at pH 6.5; corresponding to 2.85 picomole/reaction well) was incubated with 20 μl of human serum for 1 h at 37°C in Immulon 3 Plates (Nunc, Roskilde, Denmark). The enzymatic reaction was carried out using 1-chloro-2, 4, dinitrobenzene (Sigma, St. Louis, MO) substrate. Enzymatic reaction intensity was measured by OD at 340 nm at 37°C (every 15” during 3’) (BioTek EL808, softwareGEN5), and appropriate controls (enzyme without serum and tested serum alone) were added. The percentage of inhibition was calculated as the ratio of GST activity after serum incubation to the GST activity in the control. An inhibition ≥10% was considered significant. Because in epidemiological surveys, individual sera inhibiting the 28GST up to 60% are associated with a low level of egg output [22], we considered this threshold of inhibition as a criterion of efficacy for the antibody response.
The pathology associated with Sh infection was assessed by image analysis of lesions in the urinary tract detected by ultrasonography using the Niamey score, defined by WHO [23]. Ultrasonography was performed at W-8/Vp2, W0/V1, W65/V6, W82/V7, W100/V8, W117/V9, W134/V10, and W152/V11.
All selected participants living in this hyperendemic region for urinary schistosomiasis had Sh infection before inclusion and had a probability for reinfection over the course of the trial estimated close to 100% in the control group. With this consideration, it was estimated that the trial would have 80% power to show the effectiveness of the vaccine as compared with control with at least 103 subjects per arm. Because a 10% dropout rate was considered probable within the 36 months after the first vaccination trial, we specified 250 (125 per arm) children as the recruitment target for this study. The ITT population included all participants who received at least one dose of vaccine or control.
Numerical factors of the study were compared using a Student’s t-test. Categorical factors such as incidences of AEs were compared using the chi-square or Fisher’s exact test. Cumulative recurrence of schistosomiasis hematuria for 36 months after inclusion (ITT) or 24 months after the booster (PP) was estimated by the Kaplan–Meier method to evaluate vaccine efficacy, and the log-rank analysis was used to assess differences between the two groups.
Univariate and multivariate Cox regression analyses of time to events were carried out to test the treatment effect. The Statistical Analysis Software version 9.1 (SAS Cary, NC) was used for all statistical analyses, and statistical significance was defined as a two-tailed p<0.05.
Of the 298 schoolchildren assessed for eligibility, 272 subjects were enrolled, and 250 were included in the study in a short period of 25 days (Fig 2). The reasons for excluding subjects are presented in Fig 2 and Table 1. A total of 250 children (125 children in the vaccine group and 125 in the control group) were randomly assigned to receive either Bilhvax containing 100±10μg rSh28GST adjuvanted with Alhydrogel or 0.4 ml Alhydrogel as placebo in the control group. Arms were similar with respect to age, height, and weight at the time of enrollment (Table 2). The median duration of follow-up was 2.9±0.1 years, with no significant difference between the two groups. All included children received all planned doses of vaccine or placebo and PZQ treatment according to the study protocol and were included in the ITT protocol analysis. The primary endpoint was also analyzed with a modified ITT analysis, which was prospectively defined to exclude all recurrences observed before the boost.
As summarized in Table 2, of 250 school children included in the study, 149 (59.6%) were boys and 101 (40.4%) were girls. The ages ranged from 6 to 9 years, with an average age of 7.3±0.9 years. The number in each age group was 51, 99, 71, and 29 for ages 6, 7, 8, and 9 years, respectively. Weight varied from 15.9 kg to 31.4 kg (average: 22.4±3.0 kg). The mean egg count among egg-excreting children was 137.7±58.1 eggs/10 ml of urine. Mean egg counts were 133.9±55.7 and 141.6±60.5 in the control and Bilhvax groups, respectively. Among 250 infected children included in the study, 97 (39%) excreted more than 200 eggs/10 ml of urine (41/125 in control group and 56/125 in the Bilhvax group). In terms of urinary tract morbidity due to Sh infection detected by ultrasonography, the score measured at Vp2 was 7.4 (range: 0–23) and 7.1 (range: 1–25) in the control and Bilhvax groups, respectively, with no significant differences between the two groups. Thus, variables such as age, gender, BMI, infection intensity at inclusion, severity of the pathology assessed by ultrasound on inclusion, were distributed in a uniform way within of both groups (control vs vaccinated). Statistical analysis found no relationship between these variables and the achievement of the primary endpoint of the study, nor with other endpoints such as immune response to vaccine or parasitological criteria such as egg viability.
Among the 250 children, 247 (99%) (control, 123; Bilhvax, 124) presented a total of 1520 adverse events (control, 733; Bilhvax, 787) with a median of six adverse events per child (range: 3–8) during the 3-year study (2 AE/year/child). AEs of grade 1 represented 47.2% (n = 718) of the total AEs with 372 for the control group versus 346 for the vaccine group. Most frequent AEs were related to infectious diseases including schistosomiasis. Only 6% of the AEs were ≥grade 3 (control, 22; Bilhvax, 64), most of them local AEs at the injection site (Table 3).
All AEs possibly related to the treatment were identified in 74% children from the Bilhvax group and 58% children from the control group (p = 0.008). A total of 152 children (50% of the placebo group vs 72% in the vaccine group) experienced 418 local or regional adverse reactions after injection. In this context, the most frequent adverse reactions included both inflammation and pain at the site of injection. These local adverse reactions were more common in the vaccine group than in the control group (p<0.0001) (S1 Table). Nine children (six in control and three in the vaccine group, respectively) developed a total of 10 serious AEs (S2 Table), but none were life threatening or related to Bilhvax, and all were completely resolved.
The primary efficacy endpoint (the time to onset of the first schistosomiasis infection recurrence) was assessed in the ITT population, comprising all randomly assigned subjects during the 3-year follow-up period (V1 to V11). The median follow-up time for participants without recurrence was 22.9 months (interquartile range: 18.8–23.0) for the Bilhvax group and 18.8 months (interquartile range: 18.8–23.0 months) for the control group (log-rank p = 0.27). At V11, 86.4% subjects had experienced at least one recurrence in the Bilhvax group and 89.6% in the control group. Kaplan–Meier curves illustrating recurrence-free survival probability in Bilhvax and control groups are shown in Fig 3. In the mITT cohort (Fig 4), from V6 to V11, recurrences of urinary schistosomiasis were documented in 84.8% of the vaccinated individuals and in 89.6% of the control group. No significant difference was observed between the two groups (log-rank p = 0.09). When results were adjusted for sex or age groups (6–7 and 8–9 years) no differences were observed, either. Secondary efficacy endpoints defined as the percentage of subjects without recurrence and the number of recurrences per subject was similar in the two groups (S3 Table).
Vaccine immunization did not affect parasitological manifestations (number of eggs, number of viable eggs, number of hatched eggs, percent of eggs hatching) at the first recurrence or at the end of the trial (V11) between groups (S4 Table) or the number of Sm co-infected participants (S5 Table). Children were also examined by ultrasonography to assess Sh-related urinary tract morbidity using Sh score during visits Vp2, V1, V6, V7, V8, V9, V10, and V11. There were no significant differences between control and Bilhvax groups (S1 Fig).
Specific antibody response to rSh28GST was measured in Bilhvax and control groups at V1, V4, V5, V6, V9, and V11. Between V1 and V4 (post-vaccination), anti-rSh28GST total IgG geometric mean was significantly increased (p<0.001) in vaccinated children, and the difference between Bilhvax and control groups persisted until the end of the follow-up period (Fig 5). When IgG isotypes were considered, antiSh28GST IgG1, IgG2, and IgG4 antibody titers significantly increased after vaccination in the Bilhvax versus control group (Fig 6). The percent of positive subjects at V11 reached 58.4% (IgG1), 78.4% (IgG2), and 91.2% for IgG4 (Fig 6). The increase in the specific IgE antibody response was more modest, with only 36.8% positive children in the vaccine group at V11. In contrast, no significant difference in anti-rSh28GST IgG3 or IgA antibodies was observed between Bilhvax and control groups from V1 to V11 (Fig 6).
An important aspect of the generated immune response is the inhibition of rSh28GST enzymatic activity by antibodies from vaccinated animals that was previously described to be related to the anti-fecundity effect of 28GST, preventing development of schistosomiasis pathology in experimental models [24]
The capacity of sera to inhibit rSh28GST enzymatic activity is presented in S6 Table. At each point after vaccination, the mean capacity of individual sera samples to inhibit rSh28GST enzymatic activity was significantly higher in Bilhvax compared with the control group (p<0.0001). The percent sera with neutralizing antibodies (≥10% inhibition) reached more than 70% (71.2 to 95.0%) in vaccinated subjects versus 8% maximum in controls. Considering the threshold of 60% inhibition, where any of the sera of the control group reached this level of inhibitory activity, more than 52% of sera samples from the vaccine group did so at one visit (V6). In this group, 82.4% of the subjects (103) reached serum inhibitory activity ≥60% at least once during the post-vaccination follow-up.
Enzymatic inhibition is expressed a percentages. The percentage of inhibition was calculated by the ratio of GST activity after serum incubation to control GST activity. This value is considered significantly positive above 10% (n = 125 for the vaccine group; n = 125 for the control group from V1 to V6 and then 124 at V9 and V11).
To our knowledge, Bilhvax (rSh28GST adjuvanted with Alhydrogel) is the first schistosomiasis candidate vaccine to reach a phase 3 clinical trial, aiming a significant delay in Sh infection recurrence in vaccinated children aged 6–9 years compared to a control group in a hyperendemic area of Senegal. The major finding of the study is that during the 3 years of the trial, no significant differences between Bilhvax and control groups were found at the level of urinary schistosomiasis, whatever criteria were considered. However, several results obtained over the course of this specific trial have to be discussed because they could be of interest in the general context of vaccine against schistosomiasis in trial designs in endemic areas.
The potential of the schistosome P28GST to provide protection in prophylactic protocols has been demonstrated in the case of Sm, Sh, and S. bovis [1, 10, 25, 26]. In these preclinical studies, the vaccine significantly inhibited female worm fecundity and egg viability, which indicated its therapeutic properties in schistosomiasis. This result led to Bilhvax phase 1 and 2 clinical trials of Sh28GST, pointing to the conclusion that Bilhvax was safe, both in adults and children, including those with infection. In infected subjects, Bilhvax was highly immunogenic, inducing an elevated Th2 response, and could be administered in combination with one dose of PZQ (Riveau et al, in preparation). The Bilhvax 3 trial was thus planned according to this combination therapy protocol, presenting the main advantage of being included in the national program of schistosomiasis control in Senegal, and as such, accepted by the authorities and the local population.
The key strengths of this study were acceptability by the population (parents and children), participation of the school teachers, and intensive health care during the trial. The trial was conducted to a high ethical and clinical standard (see Study design and participation). Moreover, the protocol intended for school-age children with a rather long period of surveillance (3 years) included a specific clause that recommended individual PZQ treatments each time a child was found to be infected. A perfect adherence to protocol in regard to administration of the treatment strategy (vaccine+PZQ) and to the schedule of this 3-year trial was obtained. One subject dropped out at V7, but no participant was lost to follow-up during the trial.
A good tolerance to the vaccine and to the induced immune response was observed without related serious AEs. The number of reactions related to the injection site was significantly greater in the vaccine group (73%) than in the placebo group (58%). These AEs were mainly grade 2 and 3 reactions at the injection site, occurring predominantly at the time of the vaccine boost injection, and AEs observed at the injection site disappeared after 48 h.
The primary endpoint of Bilhvax efficacy was a significant delay between both groups in schistosomiasis recurrence defined by a microhematuria ≥1+ (urine stick, 25 to 70 RBC/mm3) associated with the presence of at least one Sh egg. The requirement of at least one Sh egg was included as a criterion to measure the main objective to ensure that hematuria was associated with Sh recurrence and not with other etiologies (bacterial/viral infections or trauma). This primary endpoint of efficacy was not achieved, as no statistically significant delay in the recurrence of urinary Sh infection was observed in the vaccine group compared with the control group at 36 months after the first vaccination.
The lack of a significant delay in schistosomiasis recurrence between both groups despite the induced immunity in the vaccine group may indicate that this primary endpoint might not be suitable for evaluating the effectiveness of the vaccine. It might have been more appropriate to look for an effect of the vaccination limiting the intensity of the infection during post-vaccination follow-up in comparison with the control group. However, this difference in the intensity of infection after recurrence between the two groups could not be observed. Indeed, a systematic treatment at the advent of a re-infection (hematuria 1+ and 1 living egg) renders impossible to detect an effect of vaccination on the infection intensity over time. We do think that even if the 28GST vaccine efficiency did not fall within the primary endpoint of the trial, it could have been evaluated in following the intensity of infection after recurrence if PZQ treatment had not been performed systematically without endangering children, knowing that medical follow-up was continuous and intense.
It can be considered also the influence of the repeated treatment with PZQ each time a recurrence was detected, as requested in the protocol. This factor has not been considered in previous trials and is of main interest when a trial is performed in subjects having a history of infection. Previous studies demonstrated that the widespread use of PZQ may reduce levels of immunity to urogenital schistosomiasis [27]. Indeed, PZQ treatments after vaccination, i.e., during the elaboration of immune response to Sh28GST, might have interfered with the cytokine response, as recently published by Mutapi et al. [28, 29]. PZQ treatment leads to an increase in pro-inflammatory cytokine responses to antigens from whole Sh cercariae and eggs, which both express GST [28]. In addition, PZQ treatment increases the number of subjects producing Sh28GST-specific pro-inflammatory cytokines (TNFα, IL-6, and IL-8) as well as Th1-associated cytokines (IFNγ, IL-2, and IL-12p70) cytokines, as well as Th17-associated cytokine IL-23p19 [29].
The immune response induced was consistent because 99% of the vaccinated individuals were up to the threshold. This specific response was already significantly increased before the booster (V5) and was persistent 2 years after the booster. The anti-Sh28GST response showed elevated levels of specific IgG1, IgG2, and IgG4 (major) antibodies but with an unexpected absence of IgG3 and IgA antibodies. Functionally speaking, a strong capacity of sera from the vaccine arm to inhibit the enzymatic activity of rSh28GST was observed, with a high proportion of vaccinated subject sera (74%) reaching ≥60% of inhibition.
Contrary to what was expected, the administration of PZQ before the booster showed no significant effect on the intensity of the specific response in the vaccine group. When the immune response induced by the vaccine is considered, the absence of specific IgG3 antibodies in the vaccine group, together with the low IgA and IgE levels, might represent a key factor involved in the non-efficacy of the vaccine. Indeed, our previous studies have shown elevated anti-Sh28GST IgG3, IgE, and IgA antibodies compared to IgG1, in association with acquired immunity against reinfection in urinary schistosomiasis [30]. Immunoepidemiological studies in human populations indicate that the presence of IgG3-specific antibodies correlates with naturally acquired protective immunity against schistosomiasis [31]. A phase 1 clinical trial conducted in healthy subjects [12] revealed that rSh28GST induced high IgG3 antibody titers, an effect that was associated previously with reduced egg production and decreased tract urinary pathology in Sh infection (23).
In the present clinical trial, the immune response was evaluated after the third injection of the vaccine candidate. The results show the absence of induction of IgG3 antibodies, whereas in contrast, the IgG4 levels were highly increased. In previous clinical trials, IgG3 production was observed following two administrations of the vaccine (phase 1 (12) and phase 2 (in preparation)). It seems therefore likely that, when isotypic response is considered, the third administration induced a rapid isotypic maturation, favoring a production of IgG4 to the detriment of other isotype production. This observed switch may be due to a increase in IL-4 production and action of specific CD4+ T-cells [32]. When drafting the study protocol, it has been hypothesized that induction of high immune titers should provide a better demonstration of the protective effect of the vaccine candidate. As a result, we chose to proceed to a powerful vaccination protocol including three administrations in primo immunization combined with a boost one year after the first injection. Quantitatively, the immune responses obtained have been strong and lasting. However, considering the main criteria, it would be more rational to question the isotypic quality of the specific immune response rather than its amplitude. In addition, it is quite likely that the preexisting antigen response in the included subjects, or at least the existence of immune memory for 28GST, may influence the immune response to the vaccine, notably in terms of isotypic orientation. This could be a decisive argument in favor of reducing the number of vaccine injections (for example in avoiding the boost injection) in a population that has been confronted with the parasite.
In the group of the youngest children (6 and 7 year-old) that we found the most subjects with a pre-existing immune response to the antigen. In this population of 6 to 9 years in which 30% already have an anti-28GST response, youngest children have an acquired anti-28GST response superior to that of 9-year-olds. Considering that this specific immune response can negatively influence the quality of vaccination, it would have been appropriate to target a much younger age group. Nevertheless, whether we cannot say that a pre-existing anti-28GST response has an influence on the final vaccine immune response, no quantitative and qualitative difference in induced immune response was observed between the responders before vaccination and those who were "naive" to the antigen.
An alternative hypothesis is related to the nature of the adjuvant, which was selected for its unique capacity to induce a strong Th2 response and because of its relative safety in large-scale vaccination programs, including in children, all over the world. There is indeed a limited choice of adjuvants presenting similar properties and already in the clinics. Some adjuvants with immunoregulatory properties are currently developed, such as GLA (glucopyranosyl lipid adjuvant; IDRI, Seattle). The use of such an adjuvant may have resulted in a more balanced isotypic response than that observed in the present trial [33], while limiting probably the number of administrations.
Preclinical experiments reported that the vaccine-induced immune response was associated with the detection of antibodies neutralizing the GST enzymatic activity of the vaccine. This observed association in experimental models suggested that inhibition of P28GST enzymatic activity after vaccination was strongly associated with an effect on worm fertility and thus with the protective role of the vaccine [34]. In addition, it has been observed that adult infected people, with an acquired immune response showing an inhibitory activity to 28GST, presented a significantly reduced infection intensity compared to the majority of the studied population (27). This observation in a human population exposed to schistosomiasis corroborates with the numerous results observed in animals during experimental infection. Nevertheless, the inhibition of 28GST by sera from infected individuals is significant but extremely reduced in comparison with the inhibitory capabilities of sera from individuals immunized with the 28GST protein.
In the present study, sera from vaccinated subjects showed a highly significant inhibitory activity indicating the particular affinity of the induced antibodies for the active site of the rSh28GST enzyme. However, the absence of protection despite the strong and persistent inhibitory activity might indicate either that the inhibition of GST activity is unrelated to the protection previously observed in preclinical experiments, or that the criteria for the primary efficacy endpoint were not appropriate, as has been suggested above. It should be noted that the mode of action of schistosome 28GST is not limited to detoxifying oxidized radicals by its GST function as measured by the enzyme assay used in this study. Indeed, 28GST is also a molecule allowing the transport of hormone that may play a role in the reproduction of the worm (7). It is probably that despite the enzymatic activity was inhibited by specific antibodies, this transporter function is not hampered due to the absence of related antibody production. Indeed, the active site of the enzymatic activity is well known, but the one involved in the hormonal transportation is not. It is therefore conceivable that the production of inhibitory antibodies to the enzymatic activity is not related to a protective potential that would be related in fact to the inhibition of hormone binding. This hypothesis remains to be demonstrated.
In contrast with the highly significant protection obtained in vaccine trials performed in non-human primates and in cattle, the results of this first phase 3 trial in humans appear somewhat disappointing. Although the reasons for this failure are likely due to multiple factors, emphasis should be given to the nature of the antibody response and specifically to isotype selection. Immunity to schistosomes is generally regarded as Th2-mediated and antibody-dependent. In human populations, acquisition of immunity is correlated with IgG3 and IgA antibodies to P28GST. The failure to induce the desired isotypic response seems related to three factors: isotypic variation induced by repeated PQZ treatment, use of a non-optimal adjuvant, and repeated vaccine administrations, all favoring production of IgG4 antibodies known to act generally as “blocking” antibodies. IgG4 antibodies occur following chronic exposure to antigen and are generally associated with states of immune tolerance [35]. The control of these factors will be essential for the subsequent trials. It also should be pointed out that this vaccine trial aiming at reducing pathology has been undertaken in highly infected children, who also were thus undergoing chemotherapy.
Two of the positive results of this trial are certainly the safety of the vaccine in children and the duration of the specific immune response. Later trials must thus be reassessed in the absence of PZQ administration, preferably in non-infected younger children, with another pro-Th2 adjuvant.
Regardless, the strong immunogenicity of Sh28GST, its safety of administration, including in children, together with the confirmed anti-worm fecundity effect observed in several models of schistosome infections, make use of Sh28GST as an anti-schistosome vaccine still feasible and encourage further trials.
|
10.1371/journal.pgen.1003058 | RHOA Is a Modulator of the Cholesterol-Lowering Effects of Statin | Although statin drugs are generally efficacious for lowering plasma LDL-cholesterol levels, there is considerable variability in response. To identify candidate genes that may contribute to this variation, we used an unbiased genome-wide filter approach that was applied to 10,149 genes expressed in immortalized lymphoblastoid cell lines (LCLs) derived from 480 participants of the Cholesterol and Pharmacogenomics (CAP) clinical trial of simvastatin. The criteria for identification of candidates included genes whose statin-induced changes in expression were correlated with change in expression of HMGCR, a key regulator of cellular cholesterol metabolism and the target of statin inhibition. This analysis yielded 45 genes, from which RHOA was selected for follow-up because it has been found to participate in mediating the pleiotropic but not the lipid-lowering effects of statin treatment. RHOA knock-down in hepatoma cell lines reduced HMGCR, LDLR, and SREBF2 mRNA expression and increased intracellular cholesterol ester content as well as apolipoprotein B (APOB) concentrations in the conditioned media. Furthermore, inter-individual variation in statin-induced RHOA mRNA expression measured in vitro in CAP LCLs was correlated with the changes in plasma total cholesterol, LDL-cholesterol, and APOB induced by simvastatin treatment (40 mg/d for 6 wk) of the individuals from whom these cell lines were derived. Moreover, the minor allele of rs11716445, a SNP located in a novel cryptic RHOA exon, dramatically increased inclusion of the exon in RHOA transcripts during splicing and was associated with a smaller LDL-cholesterol reduction in response to statin treatment in 1,886 participants from the CAP and Pravastatin Inflamation and CRP Evaluation (PRINCE; pravastatin 40 mg/d) statin clinical trials. Thus, an unbiased filter approach based on transcriptome-wide profiling identified RHOA as a gene contributing to variation in LDL-cholesterol response to statin, illustrating the power of this approach for identifying candidate genes involved in drug response phenotypes.
| Statins, or HMG CoA reductase inhibitors, are widely used to lower plasma LDL-cholesterol levels as a means of reducing risk for cardiovascular disease. We performed an unbiased genome-wide survey to identify novel candidate genes that may be involved in statin response using genome-wide mRNA expression analysis in a sequential filtering strategy to identify those most likely to be relevant to cholesterol metabolism based on their gene expression characteristics. Among these, RHOA was selected for further functional study. A role for this gene in the maintenance of intracellular cholesterol homeostasis was confirmed by knock-down in hepatoma cell lines. In addition, statin-induced RHOA transcript levels measured in a panel of lymphoblastoid cell lines was correlated with statin-induced change in plasma LDL-cholesterol measured in individuals from whom the cell lines were derived. Lastly, a cis-acting splicing QTL associated with expression of a rare cryptic RHOA exon was also associated with statin-induced changes in plasma LDLC levels. This result exemplifies the power of applying biological information of well understood molecular pathways with genome-wide expression data for the identification of candidate genes that influence drug response.
| Genome-wide association studies (GWAS) have been used to identify genetic contributors to a number of common diseases and traits [1]. However, a major problem with this approach is that very large sample sizes are generally required to detect statistically significant associations [2]. This is especially the case for pharmacogenomics, where identification of gene variants associated with drug response may require larger sample sizes than are generally available. Consequently, GWAS has had limited success in the identification of pharmacogenetically relevant single nucleotide polymorphisms (SNPs) that survive the stringency of genome-wide multiple testing [3], [4]. In the largest single statin clinical trial GWAS published to date (the JUPITER trial of ∼7000 individuals) only three loci (ABCG2, APOE and LPA) achieved genome-wide significance for association with the magnitude of LDL cholesterol reduction, and in total accounted for only a minor fraction of the overall variation in response [5]. Moreover, GWAS studies are limited by their ability to probe only common genetic variation and thus the limited findings suggest that association studies alone are unlikely to yield the basis for all or even the majority of the genetic variance associated with drug response.
In the present report, we describe the use of transcriptome-wide profiling to identify and prioritize genes that may contribute to inter-individual variation in statin-induced plasma LDL-cholesterol lowering. Statins inhibit HMG-CoA reductase (HMGCR), the enzyme that catalyzes the rate limiting step of cholesterol biosynthesis, thus lowering intracellular cholesterol levels [6]. This in turn elicits an increase in expression of cellular LDL receptors that mediate plasma LDL clearance [7]. Since the HMGCR gene is transcriptionally regulated by intracellular sterol content [8], the magnitude of induction of this gene is a cellular marker of in vitro statin response. We used expression array data from in vitro statin-exposed immortalized human hepatoma cell lines and lymphoblastoid cell lines established from participants of the Cholesterol and Pharmacogenetics (CAP) clinical trial of simvastatin treatment [9] to establish a set of “biological rules” for identifying genes whose expression characteristics qualified them as having biologically plausible effects on cholesterol metabolism. RHOA emerged from this analysis, and subsequent functional and genetic studies, as a novel candidate gene contributing to variation in LDL response to statins.
We used a series of filters applied to genome-wide gene expression data from 480 human lymphoblastoid cell lines (LCLs) derived from participants in the Cholesterol and Pharmacogenetics study to identify genes that appeared to be biologically plausible candidates for modulating the effects of statins on cholesterol metabolism. The following filter criteria were used (Table 1): 1) expression in normal human liver; 2) change in transcript levels in HepG2 (n = 4) and Hep3B (n = 3) human hepatoma cell lines incubated with 2.0 µM activated simvastatin versus sham buffer for 24 hr, FDR<0.01, 3) change in transcript levels in CAP LCLs incubated with 2.0 µM activated simvastatin versus sham buffer for 24 hr (Q<0.05); 4) consistent directionality of statin-induced change in transcripts in hepatoma cell lines and LCLs; 5) correlation of statin-induced gene expression change in CAP LCLs with change in expression of HMGCR. After Bonferroni correction for multiple testing (p<1.17e-04) we identified 45 genes which passed all filter criteria (Table 2). When ranked in order of correlation, only two of the top thirteen genes did not encode enzymes in the cholesterol biosynthesis pathway: transmembrane protein 97 (TMEM97) and ras homolog gene family member A (RHOA). Although both had been previously implicated in lipid metabolism [10], [11], [12], neither had been shown to play a role in the cholesterol lowering effects of statin. However, RHOA was particularly intriguing since inhibition of RHOA signaling is thought to be a major mechanism by which statins exert pleiotropic (or non-lipid lowering) actions, such as anti-inflammatory effects. Figure 1A demonstrates the strong correlation between statin-induced change in RHOA and HMGCR transcript levels (p = 7.64E-16, r2 = 0.13).
To determine if RHOA has a direct effect on markers of intracellular cholesterol homeostasis, we transfected HepG2 cells (n = 10) with siRNAs specific for RHOA or a non-targeting negative control and tested for changes in expression of HMGCR, low-density lipoprotein receptor (LDLR) and sterol response element binding transcription factor (SREBF2 aka SREBP2) gene expression. Knock-down reduced RHOA transcript levels by 60-98% (Figure 1B), with no remaining detectable RHOA protein (Figure 1C), and also generated statistically significant reductions in expression of HMGCR (0.76±0.04 fold, p = 0.002), SREBF2 (0.58±0.03 fold, p = 0.0003), and LDLR (0.73±0.13 fold, p = 0.03) Figure 1D. RHOA knockdown-mediated reductions in expression of HMGCR, LDLR, and SREBF2 were confirmed in a second hepatoma cell line, Huh7 (n = 6); however, the magnitude of the effect was less dramatic than that observed in the HepG2 transfections.
To further test the functional role of RHOA, we also measured levels of secreted APOB and APOA1, the major proteins on LDL and HDL particles respectively, in the culture media 48 hours after knock-down. APOB accumulation in the cell culture media was increased in HepG2 cells after RHOA knock-down (1.28±0.08 fold, p = 0.03, n = 12), while a similar but non-statistically significant trend was observed in Huh7 cells (1.08±0.06 fold, p = 0.10, n = 8), (Figure 1D). No significant changes in secreted levels of APOA1 were observed in either hepatoma cell line. Reduced HMGCR, LDLR, and SREBF2 transcript levels together with increased APOB secretion with RHOA knock-down are all consistent with higher intracellular cholesterol levels, which was documented in the case of cholesterol esters (1.56±0.18 fold vs. controls, p = 0.004, n = 16), (Figure 1E). Although we also detected a trend for elevated free cholesterol after knock-down, this was not statistically significant. A trend of increased intracellular cholesterol ester and free cholesterol was also observed in Huh7 cells after RHOA knock-down (1.15±0.14 fold, p = 0.05 and 1.06±0.15 fold, p = 0.27, n = 8).
Lastly, since many genes involved in the maintenance of intracellular cholesterol are transcriptionally regulated in response to changes in intracellular sterol content through SREBF2, a transcription factor, we sought to test if RHOA was also subject to SREBF2 regulation. Sterol depletion activates SREBF2, thus stimulating expression of SREBF2 target genes. We confirmed that RHOA mRNA and protein levels were substantially increased by extreme sterol depletion in HepG2 cells with 2 µM simvastatin +10% lipoprotein deficient serum for 24 hr (Figure 1F and 1G). Induction of HMGCR mRNA and protein levels served as a positive control for the effects of cholesterol depletion. Finally, we found small but statistically significant reductions in RHOA transcript levels after SREBF1 knock-down in HepG2 (0.83±0.07 fold, p = 0.05) and Huh7 (0.86±0.02 fold, p = 0.001) cell lines (Figure 1H).
Although statin-induced changes of RHOA and LDLR mRNA were positively correlated in the LCL panel (Figure 2A), change of the RHOA transcript was inversely correlated with level of LDLR cell surface protein (Figure 2B). Consistent with this relationship, we also identified an inverse correlation of RHOA transcript levels in statin- treated CAP LCLs with absolute changes in plasma total cholesterol (p = 0.02, r2 = 0.01), LDL cholesterol (p = 0.04, r2 = 0.01) and APOB (p = 0.007, r2 = 0.01), measured in vivo before and after simvastatin treatment of the individuals from whom these cell lines were derived (Table S1). In contrast, levels of RHOA in sham-treated LCLs were not significantly correlated with these measures at baseline (Table S1). Moreover, RHOA transcript levels in statin-treated LCLs were not significantly associated with statin-induced changes in plasma HDL cholesterol levels (data not shown).
We next investigated the association of common genetic variation near RHOA with in vivo statin response. Analysis of HapMap3 CEU data [13] with Haploview [14], revealed that RHOA fell within a large block of linkage disequilibrium spanning almost 500 kb and that there were four major haplotypes when considering markers within 10 kb of the gene, all with frequencies greater than 10% in the CEU population (Figure S1; Table 3). Haplotypes were inferred based on directly genotyped SNPs (Table S2) or imputed genotypes (rs11716445 for H3B), and the number of copies of each haplotype were tested for association with change in LDL-cholesterol (delta log) in response to statin treatment of Caucasian participants in CAP (n = 580) and in the Pravastatin Inflammation CRP Evaluation (PRINCE: pravastatin 40 mg/day, 24 weeks, n = 1306) clinical trial, with adjustment for sex, age, BMI, smoking status, and study population. Of the four haplotypes, H3B showed the strongest association with statin response (p = 0.01), with homozygous H3B carriers having a 29% smaller reduction in the unadjusted percent change of LDL-cholesterol compared to non-carrier controls (−21.8±4.5% versus −30.7±0.4%, Figure 3). When the CAP and PRINCE cohorts were analyzed independently, the directionality of this association was consistent between the two populations (Figure S2). Haplotype H2 also demonstrated a modest association, with carriers having greater statin-induced changes in LDL-C (p = 0.04, n = 1886, Figure 3). There were no significant associations of H3B or H2 carrier status with baseline LDL-cholesterol (p = 0.3 for both).
We found no association of either H3B or H2 with RHOA transcript levels in CAP LCLs after treatment with 2 µM statin or sham buffer (n = 115) (Figure S3A). However, rs11716445, the SNP that defines the H3B haplotype, is located in a rare 45 bp cryptic exon (referred to as RHOA exon 2.5) that we identified in multiple unique sequences during RNA-Seq analysis of three human hepatoma cell lines (HepG2, Hep3B and Huh7), and CAP LCLs (n = 3), Figure 4A. Expression of the RHOA 2.5 exon was validated by Sanger sequencing. Notably, we found that the H3B haplotype showed a very strong association with RHOA exon 2.5 levels under both sham (p = 2.7×10−7, n = 119) and statin (p = 9.1×10−13, n = 115) conditions, with carriers exhibiting the highest levels of exon 2.5 inclusion (Figure 4B and Figure S3B). The H2 haplotype also exhibited a more modest association with RHOA exon 2.5 levels in the opposite direction from H3B (p<0.01), consistent with their in vivo relationships.
Using Sanger sequencing, we found evidence of allele-specific expression (ASE) at rs11716445 with over 90% of the exon 2.5-containing transcripts originating from the H3B chromosome (Figure 4C and 4D). We also observed evidence of ASE at rs2878298, another SNP found within exon 2.5 (Figure 4D and Figure S3C). There was no significant difference in the relative amount of ASE between the statin- and sham-treated states. Finally, to determine if rs11716445 was a general expression quantitative trait locus (eQTL) or a splicing QTL, we tested for ASE at rs3448, a SNP in the 3′ UTR of RHOA, in eight heterozygous carriers (or H2/H3B) and found no evidence that the ASE extended beyond exon 2.5 (Figure 4D and Figure S3C). These results strongly suggest that rs11716445 is a cis-acting splicing QTL.
We here present results of applying a set of biologically meaningful filters to identify and rank candidate genes associated with inter-individual variation in statin effects on cholesterol metabolism based on their gene expression characteristics. Using unbiased genome-wide screens, we identified genes that were normally expressed in the human liver and changed in response to statin treatment in a manner that was correlated with statin-induced change in HMGCR quantified as an in vitro marker of statin response. From these analyses we identified a number of genes not previously implicated in the lipid-lowering response to statin as potential candidates for future study. We selected RHOA since many of the non-lipid lowering benefits, or pleiotropic effects, of statin treatment have been attributed to its ability to inhibit RHOA activity. Our validation of RHOA as a modulator of cellular cholesterol metabolism, as well as the discovery that genetic variation within RHOA is associated with the magnitude of LDL-cholesterol response to statin treatment, support the continued studies of other novel candidate genes identified through this integrative genomics strategy.
RHOA has been previously implicated in cholesterol metabolism through the modulation of ABCA1-mediated cholesterol efflux via two distinct and opposing mechanisms. RHOA inhibition stimulates ABCA1 gene expression via PPARγ and LXR activation [15], while RHOA activation increases ABCA1 protein stability [11]. Although excess intracellular levels of free cholesterol have been shown to increase RHOA activity [16], here we demonstrate that RHOA knock-down results in increased levels of secreted APOB, suggesting that RHOA may influence the pool of intracellular cholesterol available for lipoprotein production. Consistent with this hypothesis, we found that knock-down of RHOA in hepatoma cell lines resulted in increased intracellular content of cholesterol esters, the storage form of cellular cholesterol that can be mobilized for lipoprotein secretion. This occurred in conjunction with reduced expression of HMGCR and LDLR, presumably due to cholesterol-induced down-regulation of SREBF2. Very recently a novel protein, LAMTOR1 (also known as Pdro/p27RF-Rho), was found to both activate RHOA [17], and to regulate LDL-C uptake and intracellular cholesterol egress from the late endosome/lysosome [10], further supporting a link between RHOA and cholesterol metabolism.
Additional evidence for such a link is provided by the strong correlations that we observed between statin-induced changes in RHOA mRNA levels and both HMGCR and LDLR transcripts. On the other hand, there was an inverse correlation between change in RHOA mRNA and cell surface LDLR protein. While this may appear to be at odds with the change in LDLR transcript level, it is consistent with our finding that greater statin-induced RHOA gene expression was associated with reduced in vivo response lipid response to statin treatment. It is possible that increased RHOA expression directly or indirectly reduces functional LDLR at the cell surface by altering post-translational processing or cellular trafficking, hypotheses that will be tested in future studies. Increased magnitude of this effect may contribute to attenuation of statin-induced plasma LDL lowering.
Based on its role in mediating the pleiotropic effects of statin response, RHOA has been proposed as a candidate gene for the study of statin pharmacogenetics; however, genetic variation within RHOA associated with statin response has not been previously identified [18]. Here, we report that a common RHOA haplotype, H3B, is associated with reduced LDL-cholesterol lowering in response to statin treatment in data derived from two independent clinical trials. Within RHOA, this haplotype was defined by a single SNP, rs11716445; however, since rs11716445 is in strong linkage disequilibrium with many SNPs in other genes up to 500 kb away from RHOA, it is possible that its association with statin response may also be due to genetic variation affecting other genes. rs11716445 explained less than 1% of the overall variation in LDL cholesterol response to statin, so neither the H3B haplotype or rs11716445 genotype alone would be a clinically useful diagnostic, but it could be included with other known markers of statin response to improve prediction algorithms.
Here we demonstrate that rs11716445 is a cis-acting splicing QTL also associated with allele-specific expression of RHOA exon 2.5, a rare exon found within RHOA intron 2. The presence of this exon does not disrupt the open reading frame and is predicted to cause a 14 amino acid inclusion in the B3 domain of the RHOA protein, a region with no known interactions [19], [20]. Although the functional impact of exon 2.5 inclusion is unknown, the fact that the two RHOA haplotypes associated with its expression levels, H3B and H2, are also the only two RHOA haplotypes found to be associated with in vivo variation in statin-induced change in LDL-cholesterol, strongly supports the likelihood that RHOA alternative splicing is functionally relevant.
In silico analysis with ESEfinder 3.0 identified SRSF2 (aka SC35), SRSF5 (aka SRp40), and SRSF1 (aka SF2/ASF) binding sites within 20 bp of the exon 2.5 splice donor [21]. Notably, the rs11716445 “T” allele is predicted to disrupt an SRSF5 binding motif (TAGA[T/C]C) (Figure S4). This finding is consistent with previous reports demonstrating that SRSF5 and SRSF2 antagonize SRSF1 to promote exon exclusion, as the loss of the SRSF5 binding with the “T” allele would be predicted to result in exon 2.5 inclusion [22]. Thus, these results strongly suggest that the rs11716445 “T” (minor) allele enhances expression of the RHOA 2.5 exon. We also found that the proportion of the expressed RHOA 2.5 exon containing the “T” allele was reduced in H1/H3B compared to H2/H3B and H3A/H3B heterozygotes (Figure S3C). Since the H1 haplotype contains the minor allele of the second common SNP within the 2.5 exon, rs2878298, which is predicted to generate a SRSF1 binding site, these findings suggest that there are multiple gene variants that regulate expression of this novel exon; however the functional effects of these SNPs (rs11716445 and rs2878298) as well as the expression of the cryptic RHOA exon remain to be tested.
In summary, we here report using a combination of expression array data, functional studies, and genetic analyses that RHOA is a novel candidate gene associated with variation in both in vitro and in vivo response to statin. Although additional studies of statin effects will be required to corroborate these findings, they demonstrate the value of using data from a variety of molecular techniques, including the combination of in vivo and in vitro genetically-regulated phenotypes, as a novel approach for identifying genes involved in drug response.
Lymphoblastoid cell lines (LCLs) from 480 Caucasian participants from the Cholesterol and Pharmacogeneomics (CAP) clinical trial [9] and HepG2 and Hep3B cell lines were grown under standard conditions and exposed to 2 µM simvastatin or sham buffer for 24 hours as previously described [23]. Although much higher than normal circulating levels of plasma simvastatin, 2–40 nM [24], this concentration of simvastatin was selected based on previous dose response experiments that were used to determine the amount that elicits a consistent and significant induction of both HMGCR and LDLR mRNA (Figure S5) [23]. Simvastatin was provided by Merck Inc. (Whitehouse Station, NJ) and activated to the β-hydroxyacid prior to use [25]. Cell surface LDLR protein was measured in statin and sham treated CAP LCLs as previously described [26]. To confirm statin regulation, HepG2 cells were grown in 6-well plates and incubated with 2 µM activated simvastatin +10% lipoprotein deficient serum (Hyclone) for 24 hours.
Genome-wide gene expression was measured in RNA from CAP samples and statin and sham treated HepG2 and Hep3B cells. RNA was converted to biotin-labeled cRNA using the Illumina TotalPrep-96 RNA amplification kit (Applied Biosystems, Foster City, CA). cRNA was hybridized to Illumina HumanRef8v3 expression beadchips (Illumina, San Diego, CA). Data were analyzed using GenomeStudio (Illumina). All beadchips had a signal P95/P05>10. Significance analysis of microarrays (SAM) [27] was performed on the 10,291 of 18,630 probed genes that were expressed in LCLs (FDR<0.05). Expression traits were adjusted for known covariates (age, gender, exposure batch, cell growth rate as determined by cell count on exposure day, and RNA labeling batch) and also for unknown sources of variation through adjustment for those principal components that described greater than 5% variance across the dataset [28]. Adjusted data were quantile normalized across each gene to ensure normality.
Gene expression in human liver was determined using mean detection p-value as determined by GenomeStudio (Illumina, San Diego, CA) from expression profiles measured by Illumina Ref8v2 beadarray on 120 human liver samples (2 technical replicates each of 60 samples, GEO accession number: GSE28893 [29]). Mean detection p-values across all 120 samples was assessed, and genes with a p<0.05 were called expressed.
HMGCR, LDLR, SREBF1, SREBF2, RHOA (total), and RHOAexon2.5 transcript levels were quantified by qPCR with gene expression normalized to CLPTM (TaqMan Assay number: Hs00171300_m1, Life Technologies) as previously described [30]. Primers used for qPCR of total RHOA were F: CGGAATGATGAGCACACAAG and R: TGCCTTCTTCAGGTTTCACC and those used for qPCR of RHOA exon 2.5 were F: TATCGAGGTGGATGGAAAGC and R: GCCAACTCTACCATAGTACATTGAAA.
RHOA, SREBF1 and SREBF2 knock-down was achieved by 48 hour transfection of 80,000 HepG2 or Huh7 cells/well in 12-well plates using either the Ambion Silence Select siRNA (s759) or non-targeting control according to the manufacturer's protocol. Cell culture media was collected from all samples at time of harvest, and APOB and APOAI were quantified in triplicate by sandwich-style ELISA. Samples with a coefficient of variation greater than 15% were subject to repeat measurement. Cholesterol was extracted from the cell pellets with hexane-isopropanol (3∶2, v/v) and dried under nitrogen. The extracted cholesterol was reconstituted with reaction buffer (0.5 M potassium phosphate, pH 7.4, 0.25 M NaCl, 25 mM cholic acid, 0.5% Triton X-100). Total cholesterol content was determined with the Amplex Red Cholesterol Assay Kit (Invitrogen) and normalized to total cellular protein quantified by the Pierce BCA Protein Assay Kit (Thermo Scientific). To quantify RHOA protein levels, cells were lysed in Cell Lytic lysis buffer (Sigma), loaded on a 4–12% Tris-Glycine Gel (Invitrogen), and proteins were transferred onto a PDVF membrane using the iBLOT gel transfer system (Invitrogen). The blot was then probed with antibodies diluted 1∶200 to RHOA (SC26C4), HMGCR (SCH300) and β-actin (SC ACTBD11B7), all purchased from Santa Cruz Biotechnology. Band densities were analyzed using the Mulitplex Band Analysis tool in Alphaview SA version 3.4.0.
Haplotypes H1, H2, and H3A were assigned using genotype data from tag SNPs (Table S2), while haplotype H3B was inferred using imputed rs11716445 genotypes. Imputation was performed in BIMBAM using 317K or 610K genotypes in a similar manner as previously described [31] except for use of the HapMap3 and 1KGP CEU pilot data as a reference population.
LDL-cholesterol was quantified in self-reported Caucasian American participants of the Cholesterol and Pharmacogenetics (CAP) clinical trial twice at baseline and after both 4 weeks and 6 weeks of simvastatin 40 mg/day and in the participants of the Pravastatin Inflammation and CRP Evaluation (PRINCE) clinical trial after 12 and 24 weeks of pravastatin 20 mg/day as previously described [9], [32]. Delta log LDL-cholesterol was calculated as the log average value of LDL-cholesterol on treatment minus the log average of the two baseline measurements, and percent change was the average on-statin value minus the average baseline value over the average baseline value. The CAP trial is registered at ClinicalTrials.gov (NCT00451828). Informed consent was obtained and approved by the institutional review boards of the sites of recruitment, University of California Los Angeles and San Francisco General Hospital. In addition, all research involving human participants was approved by the Children's Hospital Oakland Research Institute IRB. All haplotypes with a minor allele frequency greater than 5% were identified using Haploview [14] with HapMap3 CEU data. Using an additive genetic model, haplotypes were tested for association with change in delta log LDL-cholesterol using combined results of both clinical trials with adjustment for age, sex, BMI, smoking status, and study population as well as for each trial separately with adjustment for age, sex, BMI, and smoking status.
Hep3B, HepG2, and Huh7 cells were incubated in duplicate under either standard growth conditions (MEM supplemented with 10% FBS, 1% nonessential amino acids and 1% sodium pyruvate) or sterol depleted conditions (MEM supplemented with 1% nonessential amino acids, 1% sodium pyruvate, 2.0 µM simvastatin and 10% lipoprotein deficient serum) for 24 hours. RNA was extracted as previously described and samples from the duplicate experiments were pooled. Sequencing libraries were prepared by isolating mRNA from 7–10 µg total RNA using two rounds of the MicroPoly(A)Purist kit (Ambion), fragmenting the mRNA for 20 seconds, synthesizing cDNA using random primers, repairing ends, dA-tailing, ligating adapters, gel purifying fragments, amplifying libraries using indexed primers for 15 PCR cycles, and performing another round of gel purification. Libraries were sequenced to an average depth of 60 million 100 bp reads (30 million paired-end fragments). Expression of the novel RHOA exon was verified in independent samples through RT-PCR and Sanger sequencing.
DNA and RNA was isolated from CAP LCLs after 24 hours of exposure to sham buffer or 2 µM simvastatin. The DNA sequences of exon 2.5 and exon 5 were amplified using F: CAAGGCAGGAGAATGGTGTG and R: CCACTGACGATGATTGCTTC and F: GGCCATATTACCCCTTTTCG and R: CCAGAGGGATCTAGGCTTCC, respectively. RT-PCR was performed to amplify the transcript sequences of exon 2.5 and exon 5 (3′UTR) using F: TCGTTAGTCCACGGTCTGGT and R: GCCAACTCTACCATAGTACATTGAAA and F: CGGAATGATGAGCACACAAG and R: TTGGAAAAATTAACTGGTACAGAAA, respectively. PCR products were then subject to Sanger sequencing.
|
10.1371/journal.ppat.1005668 | CLCuMuB βC1 Subverts Ubiquitination by Interacting with NbSKP1s to Enhance Geminivirus Infection in Nicotiana benthamiana | Viruses interfere with and usurp host machinery and circumvent defense responses to create a suitable cellular environment for successful infection. This is usually achieved through interactions between viral proteins and host factors. Geminiviruses are a group of plant-infecting DNA viruses, of which some contain a betasatellite, known as DNAβ. Here, we report that Cotton leaf curl Multan virus (CLCuMuV) uses its sole satellite-encoded protein βC1 to regulate the plant ubiquitination pathway for effective infection. We found that CLCuMu betasatellite (CLCuMuB) βC1 interacts with NbSKP1, and interrupts the interaction of NbSKP1s with NbCUL1. Silencing of either NbSKP1s or NbCUL1 enhances the accumulation of CLCuMuV genomic DNA and results in severe disease symptoms in plants. βC1 impairs the integrity of SCFCOI1 and the stabilization of GAI, a substrate of the SCFSYL1 to hinder responses to jasmonates (JA) and gibberellins (GA). Moreover, JA treatment reduces viral accumulation and symptoms. These results suggest that CLCuMuB βC1 inhibits the ubiquitination function of SCF E3 ligases through interacting with NbSKP1s to enhance CLCuMuV infection and symptom induction in plants.
| Viruses pose a serious threat to field crops worldwide; therefore, understanding the mechanisms of viral disease can help crop improvements. Here, we investigate how Cotton leaf curl Multan virus (CLCuMuV) interacts with plant to cause viral disease. We found that CLCuMuV uses its sole satellite-encoded protein βC1 to regulate the plant ubiquitination pathway for effective infection. By interrupting the interaction of NbSKP1 with NbCUL1 through its interaction of SKP1, βC1 interferes with the plant ubiquitination pathway and impairs plant hormone signallings to enhance viral accumulation and symptoms. These new insight into the mechanisms of viral disease may help crop improvements in the future.
| Monopartite begomoviruses often possess an essential disease-specific betasatellite and are responsible for devastating diseases in many crops [1]. For example, at least six distinct begomoviruses that are associated with a single betasatellite, Cotton leaf curl Multan betasatellite (CLCuMuB), cause Cotton leaf curl disease (CLCuD), which is a major constraint to cotton production in Asia [2]. Cotton leaf curl Multan virus (CLCuMuV) is one of these begomoviruses and can infect cotton and many other plants including Nicotiana benthamiana. CLCuMuV consists of a circular single-stranded DNA genome that encodes only 6 proteins (V1 and V2 in virion-sense strand whilst C1, C2, C3 and C4 in virion complementary-sense strand). CLCuMuB is a small circular single-stranded DNA molecule that is essential for CLCuMuV to induce disease symptoms in plants [3].
Betasatellites, such as CLCuMuB, are approximately half the size of the begomovirus DNA genomes. They require the helper begomoviruses for replication and movement in plants and only encode a single multifunctional pathogenicity protein βC1 [1]. βC1 can up-regulate the proliferation of its cognate helper virus [4], and complement the movement function encoded by the DNA B component of some bipartite begomoviruses [5]. βC1 is essential for producing viral disease symptoms [4, 6–12] and plays important roles in suppression of transcriptional (TGS) [13] and posttranscriptional gene silencing (PTGS) [14–18]. Furthermore, βC1 can also promote the performance of the whitefly and impair plant development [19–22]. More details about the multiple functions of βC1 can be found in recently published reviews [1, 23]. However, how geminiviruses exploit βC1 to perform these diverse functions needs further investigations.
Ubiquitination is a highly dynamic posttranslational modification process that is a major protein degradation and rapid regulatory mechanism in plants [24]. Through the action of a sequential cascade of three enzymes consisting of E1, E2, and E3, ubiquitin is covalently attached to substrate proteins, and then, in most cases, the polyubiquitinated proteins will be degraded by the 26S proteosome. As the most abundant member of the E3 family, the SKP1/CUL1/F-box (SCF) complex is the best characterized multi-subunit ubiquitin ligase. In the SCF complex, SKP1/ASK1 (S-phase kinase-associated protein) acts as a bridge between CUL1 (Cullin1) and F-box proteins. CUL1 is the major structural scaffold and F-box proteins are responsible for recognizing target substrates. RBX1 is the fourth subunit that is heterodimerized with CUL1, and binds E2 through its RING Finger domain. More than 700 predicted F-box proteins are encoded by the Arabidopsis thaliana genome, suggesting these F-box proteins have highly targeting potentials for extensive regulatory functions [25, 26].
The SCF complex-based E3 ubiquitin ligases have been known to regulate plant hormone signaling. Several phytohormone receptors are F-box proteins in SCF complexes, such as SCFTIR1 for auxin, SCFCOI1 for jasmonates, SCFSLY1/GID2 for gibberellins and SCFMAX2 for strigolactones [27–30]. In addition, SCF complexes regulate ethylene (ET) signal transduction at multiple points (SCFETP1 and SCFETP2 for EIN2, SCFEBF1 and SCFEBF2 for EIN3) [31, 32]. Since phytohormones have pivot functions in vegetative growth, compromising of these pathways is usually accompanied by abnormal developmental phenotype. Among them, JA plays a crucial role in defense against pathogens and insects. Recently, JA pathway was reported to be involved in plant defense against geminivirus infection [33].
In this study, we report that a geminivirus uses its satellite-encoded βC1 to interfere with the ubiquitination function of SCF E3 ligases to enhance viral infection and symptom development in plants.
CLCuMuB was reported to enhance DNA accumulation of the helper virus and be necessary for producing viral disease symptoms [4]. To see whether βC1 is responsible for these functions, we constructed a null mutant betasatellite for the βC1 gene [34] with a ATG-TGA transition in the start codon, hereafter called βM1 (S1 Fig). Different from N. benthamiana plants infected with CLCuMuV and β (CA+β) causing severe downward leaf curling and darkening as well as swollen veins, plants infected with CLCuMuV and βM1 (CA+βM1) grew taller, developed much milder symptoms and accumulated much less CLCuMuV genomic DNA (S2A and S2B Fig).
Further, we generated transgenic N. benthamiana plants expressing non-tagged or tagged βC1. However, most transgenic plants have very severe symptoms and are infertile or dead finally. Nevertheless, we were able to obtain five lines expressing non-tagged βC1 under control of its native promoter (βC1pro:βC1), 2 lines expressing GFP-tagged βC1 driven by CaMV 35S promoter (35Spro:GFP-βC1) and 4 lines expressing HA-tagged βC1 driven by CaMV 35S promoter (35Spro:HA-βC1). All these transgenic plants showed aberrant development phenotype (S3 Fig).
Taken together, these results suggest that CLCuMuB βC1 is required for development of typical disease symptoms and enhancement of CLCuMuV DNA accumulation.
To understand how CLCuMuB βC1 facilitates virus infection, we used CLCuMuB βC1 as bait in a yeast two-hybrid (Y2H) system [35] to identify host CLCuMuB βC1-interacting proteins. From screening the Solanum lycopersicum cDNA library, we characterized a full-length SKP1-like protein (designated as SlSKP1) that interacted with βC1. Furthermore, 12 putative NbSKP1 homologues identified in the N. benthamiana genome through bioinformatics analysis (http://solgenomics.net), encode proteins with more than 44% amino-acid identity to SlSKP1. However, we obtained only 4 predicted cDNAs by RT-PCR. Indeed, RNA-seq results (ftp://ftp.solgenomics.net/transcript_sequences/by_species/Nicotiana_benthamiana/) indicates that other 8 putative homologues are not or rarely expressed in leaf tissues. Three of the 4 NbSKP1 homologues NbSKP1.1, NbSKP1.2 and NbSKP1.3, collectively called NbSKP1s, interact with CLCuMuB βC1, whilst the other do not or interact very weakly with βC1 in yeasts and it is named as NbSKP1L1 (NbSKP1-like 1) (Fig 1A). NbSKP1.1 shares 95.5%, 91.7% and 44.9% amino-acid identity to NbSKP1.2, NbSKP1.3 and NbSKP1L1, respectively (S4 Fig).
To examine whether CLCuMuB βC1 directly interacts with NbSKP1.1, in vitro GST pull-down assay was performed. His-HA double-tagged NbSKP1.1 (His-HA-NbSKP1.1) was expressed in E. coli BL21 (DE3) and then purified by Ni-NTA Agarose (Qiagen, Netherlands) column. After elution, His-HA-NbSKP1.1 was incubated with Glutathione Sepharose 4B (GE, American) bonded with E. coli-expressed GST, GST-tagged CLCuMuB βC1 (GST-βC1) or its mutant with the deletion of C-terminal 43 amino acids (GST-βC1ΔC43). His-HA-NbSKP1.1 was pulled down by GST-βC1 but not GST and GST-βC1ΔC43 (Fig 1B), indicating that NbSKP1.1 can directly interact with βC1. To our surprise, His-HA double-tagged NbSKP1L1 (His-HA-NbSKP1L1) was also pulled down by GST-βC1 but not GST and GST-βC1ΔC43. (Fig 1B).
We also demonstrated in planta interaction of CLCuMuB βC1 with NbSKP1.1 using co-immunoprecipitation (Co-IP) assay. In this assay, HA-tagged NbSKP1.1 (HA-NbSKP1.1) was co-expressed transiently with GFP or GFP-tagged CLCuMuB βC1 (GFP-βC1) in N. benthamiana by agroinfiltration. GFP-βC1 transgenic N.benthamiana exhibits leaf curl symptoms, which indicates GFP-βC1 is a functional protein (S3C Fig). Total protein extracts were immunoprecipitated by GFP-Trap beads (ChromoTek, German). The resulting precipitates were analyzed by western blot assays using an anti-HA antibody (CST, USA). We found that HA-NbSKP1.1 was co-immunoprecipitated by GFP-βC1 but not GFP (Fig 1C). Similarly, we also found that HA-tagged NbSKP1L1 (HA-NbSKP1L1) was co-immunoprecipitated by GFP-βC1 but not GFP (Fig 1C). To confirm these Co-IP results, we performed the reverse IP. GFP-βC1 was co-expressed transiently with HA-tagged GUS (HA-GUS), HA-NbSKP1.1 or HA-NbSKP1L1 in N. benthamiana by agroinfiltration. Total protein extracts were immunoprecipitated by HA-beads (Abmart, China). The resulting precipitates were analyzed by western blot assays using an anti-GFP antibody (ChromoTek, German). GFP-βC1 was pulled down by HA-NbSKP1.1 and HA-NbSKP1L1 but not HA-GUS (S5A Fig).
To find where CLCuMuB βC1 interacts with NbSKP1.1 and NbSKP1L1 in plant cells, citrine yellow fluorescent protein (YFP)-based bimolecular fluorescence complementation (BiFC) assays [36] were performed. HA-tagged βC1 or βC1ΔC43 was fused to the N-terminal domain of YFP (nYFP) to generate HA-βC1-nYFP or HA-βC1ΔC43-nYFP. NbSKP1.1, NbSKP1L1 and the N-terminal fragment of firefly luciferase (nLUC) as a negative control were fused to HA-tagged C-terminal domain of YFP (HA-cYFP) to generate HA-cYFP-NbSKP1.1, HA-cYFP-NbSKP1L1 and HA-cYFP-nLUC. Western blot assays using an anti-HA antibody showed that all chimeric proteins can be expressed correctly (S5B Fig). HA-βC1-nYFP or HA-βC1ΔC43-nYFP was transiently co-expressed with HA-cYFP-NbSKP1.1, HA-cYFP-NbSKP1L1 or HA-cYFP-nLUC respectively in N. benthamiana. No such interaction between HA-βC1-nYFP and HA-cYFP-nLUC was found. However, positive interactions between HA-βC1-nYFP and HA-cYFP-NbSKP1.1 or HA-cYFP-NbSKP1L1 were observed in both nucleus and cell periphery, as indicated by occurrence of yellow fluorescence (Fig 1D). As a control, HA-βC1ΔC43-nYFP didn’t interact with HA-cYFP-NbSKP1.1 or HA-cYFP-NbSKP1L1 (Fig 1D).
Taken together, these results demonstrate that NbSKP1s and NbSKP1L1 interact with CLCuMuB βC1 both in vitro and in vivo, and the interaction of the two proteins occurs in nucleus and cell periphery of plant cells.
The crystal structures of human SKP1 [37] and Arabidopsis ASK1 [38] suggest that SKP1 can be divided into N-terminal and C-terminal domains. The N-terminal BTB-POZ domain of SKP1 is responsible for its binding to CUL1 whilst its C-terminal domain is thought to be essential for SKP1 to interact with F-box proteins. The Y2H assays showed that CLCuMuB βC1 interacted with the first 98 amino-acid N-terminal region of NbSKP1.1 (N98aa), but not with the C-terminal region (aa 99–155) of NbSKP1.1(C57aa), as indicated by growth of yeast on Leu− plates containing galactose (Gal) and raffinose (Raf) (Fig 2).
In human and Arabidopsis, SKP1/ASK1 interacts with CUL1 to assemble into SCF complexes through its N-terminal domain [37, 38]. We found that CLCuMuB βC1 interacts with N-terminal domain of NbSKP1.1 (Fig 2). This prompted us to investigate whether CLCuMuB βC1 interferes with the assembly of NbSKP1.1 into the SCF complex. To test this hypothesis, GFP competitive pull-down assay was performed. Because E. coli-expressed NbCUL1 was insoluble, GFP and GFP-tagged NbCUL1 (GFP-NbCUL1) were expressed in N. benthamiana, then precipitated by GFP-Trap beads. To eliminate the influence from endogenous NbSKP1s and NbSKP1L1, an excessive amount of E. coli-expressed His-HA-NbSKP1.1 was used to saturate the beads and endogenous NbSKP1s and NbSKP1L1 were crowded out from GFP-NbCUL1, then the supernatant was removed. After an increasing amount of E. coli-expressed His-tagged βC1 (His-βC1) was added, more and more His-HA-NbSKP1.1 was pulled off from GFP-NbCUL1. and levels of His-HA-NbSKP1.1 released into the supernatant were increased (Fig 3A).
Further, we confirmed CLCuMuB βC1 interfering with the interaction between NbSKP1.1 and NbCUL1 by BiFC assays. We generated nYFP-NbSKP1.1 and cYFP-NbCUL1 fusion constructs and co-expressed them with HA-tagged nLUC (HA-nLUC) or HA-tagged CLCuMuB βC1 (HA-βC1) in N. benthamiana. HA-βC1 is a functional protein (S3D–S3F Fig). Stronger signals were detected for the combination of nYFP-NbSKP1.1 and cYFP-NbCUL1 in the presence of HA-nLUC than in the presence of HA-βC1 (Fig 3B and 3C). Meanwhile, the protein level of nYFP-NbSKP1.1 and cYFP-NbCUL1 seem similar between the two groups (Fig 3D).
These data suggest that CLCuMuB βC1 interferes with the interaction between NbSKP1.1 and NbCUL1 via binding to NbSKP1.1.
βC1 but not βC1ΔC43 interacts with NbSKP1s and NbSKP1L1. Meanwhile βC1 but not βC1ΔC43 induces viral symptoms (S6 Fig). These results promote us to check whether silencing NbSKP1s can produce some viral symptoms. We constructed a deletion mutant betasatellite by replacing the entire βC1 gene from CLCuMuB with sites of two restriction enzymes AscI and XbaI to generate CLCuMuB (ΔβC1), hereafter called βM2 (S1 Fig). We guessed that our CLCuMuB-based vector βM2 may be used as a VIGS vector. To confirm this, we cloned a N. benthamiana phytoene desaturase (NbPDS) gene fragment into βM2 to generate βM2-PDS. Photo-bleach phenotype was observed around the leaf veins of N. benthamiana plants agroinoculated with βM2-PDS in the presence of helper virus CLCuMuV (S7 Fig). This result demonstrates that βM2 can be used as a CLCuMuB-based VIGS vector to effectively silence genes, and CLCuMuV may exhibit a phloem limitation.
To investigate the role of NbSKP1s in CLCuMuV infection, we silenced NbSKP1s using our CLCuMuB-based VIGS vector, βM2. To exclude the effect from size, three cDNA fragments corresponding to the 176-bp, 184-bp and 345-bp NbSKP1.1 sequences were fused with 169-bp, 161-bp and 0-bp βC1 sequences respectively and then were cloned into βM2 to generate βM2-SKP1F1, βM2-SKP1F2 and βM2-SKP1F3 (Fig 4A1–4A3). A 345-bp fragment of βC1 was inserted into βM2 to generate βM2-βC1F as the control. The position relationship among 176-bp, 184-bp and 345-bp NbSKP1.1 fragments was shown in S8 Fig. N. benthamiana plants were agroinfiltrated with CLCuMuV (CA) and βM2-βC1F, βM2-SKP1F1, βM2-SKP1F2 or βM2-SKP1F3. Silencing of NbSKP1s resulted in an increasing accumulation of CLCuMuV DNA at 14 dpi (Fig 4B1–4B3). Since the mRNA level of NbSKP1L1 was very low in normal plants (S9 Fig), and similar results can be found in the RNA-seq data of N. benthamiana in Sol Genomics Network (ftp://ftp.solgenomics.net/transcript_sequences/by_species/Nicotiana_benthamiana/), we gave up to check the mRNA level of NbSKP1L1. Silencing of NbSKP1s (NbSKP1.1, NbSKP1.2 and NbSKP1.3) was triggered by all three constructs, and the levels of NbSKP1s mRNA were significantly reduced when compared to the βM2-βC1F control (Fig 4C1–4C3). βM2-SKP1F3 was more effective than βM2-SKP1F1 and βM2-SKP1F2 to cause silencing of NbSKP1s (Fig 4C1–4C3). At 21 dpi, 50% plants infected with CA+βM2-SKP1F1, 50% plants infected with CA+βM2-SKP1F2 and 100% plants infected with CA+βM2-SKP1F3 exhibited severe downward leaf curling and darkening as well as swollen veins, typical symptoms in plants infected by CA+β (Fig 4D1–4D3). If we continue to observe the symptom development, growth retardation will also be found (S10 Fig).
We also confirmed the effect of silencing NbSKP1s on CLCuMuV accumulation and symptoms using another control βM2-GFPF, which 345-bp GFP fragment was cloned into βM2. N. benthamiana plants were agroinfiltrated with CLCuMuV (CA) and βM2-GFPF or βM2-SKP1F3. We found again that silencing of NbSKP1s enhances CLCuMuV DNA accumulation and results in viral symptoms (S11 Fig).
TYLCCNB-based VIGS works mainly in vascular tissues [39], the tissues which CLCuMV tends to be limited to [40]. We further confirmed the effect of silencing NbSKP1s on CLCuMuV infection by TYLCCNB-based VIGS system [39]. We inserted the 345-bp GFP fragment and the 345-bp SKP1F3 fragment into pBinPLUS-2mβ of TYLCCNB-based VIGS system [39], then agroinoculated them respectively with TYLCCNV for silencing. Similarly, silencing of NbSKP1s enhanced CLCuMuV DNA accumulation and 100% NbSKP1s silenced plants exhibited viral symptoms (S12 Fig).
We have demonstrated that βC1 is able to interfere with the interaction between NbSKP1s and NbCUL1 (Fig 3). Moreover, silencing of NbSKP1s has a dramatic influence on viral DNA accumulation and symptom development (Fig 4). We therefore investigated whether silencing of NbCUL1 could also enhance CLCuMuV DNA accumulation and cause severe viral symptoms. Two cDNA fragments corresponding to the 268-bp and 345-bp sequences of NbCUL1 were fused with 77-bp and 0-bp βC1 sequences respectively and then were cloned into βM2 to generate βM2-CUL1F1 and βM2-CUL1F2 respectively (Fig 5A1 and 5A2). The position relationship among 268-bp, and 345-bp NbCUL1 fragments were shown in S8 Fig. These two VIGS vectors along with CLCuMuV were then agroinfiltrated respectively into N. benthamiana plants. Silencing of NbCUL1 by either CA+βM2-CUL1F1 or CA+βM2-CUL1F2 resulted in an higher accumulation of CLCuMuV DNA (Fig 5B1 and 5B2) and severer viral symptoms (Fig 5D1 and 5D2).
Taken together, these results suggest that βC1 may enhance its helper geminivirus’ accumulation and viral symptom induction by interfering with the interaction between SKP1 and CUL1 through its binding to SKP1.
Because βC1 interferes with the interaction between SKP1 and CUL1, and cul1 mutants are altered in JA responses [41, 42], we tested whether βC1 can interfere with JA pathways. First, we evaluated root growth rate in HA-βC1 transgenic plants, the root length of 6-day-old seedlings was measured every 24 h for 5 days. Data showed that HA-βC1 transgenic roots grow more slowly than wild-type roots (Fig 6A). Meanwhile, we measured inhibition of primary root elongation caused by treatment with methyl-jasmonate (MeJA), and HA-βC1 transgenic plants showed less sensitivity than wild-type plants to 50 μM MeJA (Fig 6B). Further, quantitative real-time PCR was used to quantify the mRNA level of marker genes for JA responses. Three genes: Defensin-like protein 1, Defensin-like protein 2 and Pathogen like protein were chosen for JA responses. Compared to wild-type plants, all three markers genes showed lower mRNA expression level in two independent HA-βC1 transgenic lines (#2 HA-βC1 and #3 HA-βC1) (Fig 6C). Auxin and gibberellins signalings are also regulated by CUL1-based SCF ubiquitin E3 ligases [27, 29]. Real-time PCR assays showed lower mRNA expression level of their marker genes (Gibberellin-regulated protein 14 and Gibberellin-regulated protein 6 for gibberellins, SAUR14 and PID for auxin) in HA-βC1 transgenic lines than in wild-type controls (S13A and S13B Fig).
Taken together, CLCuMuB βC1 can really cause deficient function in SCF complexes and interfere with hormone signaling pathways.
SCFCOI1 is the receptor for JA, and some geminiviruses interfere with JA pathway [20, 21, 33, 43, 44]. Meanwhile CLCuMuB βC1 seems to have no inhibition on jasmonates biosynthesis according to JA level data measured by mass spectrum and HPLC. Regardless of being wounded or not, plants infected with CA+β showed higher JA level compared to plants infected with CA+βM1 or healthy plants (S14 Fig). These results imply that CLCuMuB βC1 doesn’t impair JA biosynthesis. Higher JA level in plants infected with CA+β may be derived from the feedback due to the impaired JA signaling.
The stability of JA receptor COI1, a F-box protein, is dependent on an intact SCFCOI1 complex [45]. Because βC1 can interfere with the interaction between SKP1 and CUL1, we assumed that it may reduce the stability of COI1 in vitro. Co-IP analysis indicated that GFP-CUL1 associated with both Myc-COI1 and HA-NbSKP1.1 (S15 Fig), suggesting that Myc-COI1 can be integrated within SCF complexes. After Myc-COI1 was transiently expressed in N.benthamiana and purified with anti-Myc affinity beads. Myc-COI1 protein was then mixed with total protein extracts prepared from N.benthamiana which was transiently expressed HA-βC1 or HA-nLUC. The stability of Myc-COI1 was assessed by western blot assays after the treatment at 25°C for various periods of time up to 8 h. The Myc-COI1 protein degraded more rapidly in HA-βC1 extracts compared to in HA-nLUC extracts (Fig 6D and 6E). Moreover, the accumulation of Myc-COI1 in HA-βC1 transgenic lines was reduced 84–92% compared to that in wild-type plants (WT) (S16 Fig), whilst the accumulation of GFP (as an expression control) in HA-βC1 transgenic lines was reduced by 26–41% in WT plant (S16 Fig).
Taken together, these data implied that CLCuMuB βC1 damages the integrity of SCFCOI1 complex to hinder JA responses.
GA releases the brakes of plant growth. During this process, DELLA protein GAI is ubiquitinated by the SCFSLY1 and eventually degradated by the 26S proteasome [46]. Mutant plants that are deficient in GA pathways exhibit a dwarf phenotype [46]. Further, plants infected with CA+β is dwarf compared to plants infected with CA+βM1 (S2 Fig). To check whether the function of SCFSLY1 is hindered by CLCuMuB βC1, we co-expressed YFP-GAI with either HA-βC1 or HA-nLUC to investigate its degradation as described [33]. At 48 hpi, YFP-GAI fluorescence was observed in the nuclei 48 hpi (Fig 7A), indicating YFP-GAI can be co-expressed with HA-βC1 or HA-nLUC normally in N. benthamiana leaves. However, whether plants were treated with 100 μM GA3 or not, YFP-GAI fluorescence was enhanced when co-expressed with HA-βC1 (Fig 7A). Western blot assays using an anti-GFP antibody indicated that YFP-GAI accumulation was less in plants co-expressed with HA-nLUC than those co-expressed with HA-βC1 (Fig 7A). Meanwhile, co-expression with HA-βC1 or HA-nLUC did not significantly affect mRNA level of YFP-GAI at this time point (Fig 7B). Moreover, co-expression of HA-βC1ΔC43 did not enhance YFP-GAI accumulation (S17 Fig). As an internal control, a GFP expression construct was coinfiltrated with HA-βC1 or HA-nLUC expression construct. No significant differences in GFP fluorescence or GFP protein accumulation were detected between them (Fig 7C).
Taken together, these results indicate that CLCuMuB βC1 can increase the accumulation of GAI by hindering its degradation to hinder GA responses.
βC1 interferes with SCF function to enhance geminivirus DNA accumulation and damages the integrity of SCFCOI1 complex to hinder JA responses. This would suggest that JA is likely to be involved in plant defense against CLCuMuV. To test this hypothesis, we inoculated CLCuMuV along with CLCuMuB into MeJA or mock-treated N. benthamiana plants. Symptoms were daily monitored from 9 to 14 dpi. We found that application of exogenous MeJA resulted in milder symptoms (Fig 8A–8E) and lower viral DNA accumulation (Fig 8F). These results demonstrate that MeJA could compromise viral pathogenicity. We also inoculated CLCuMuV along with βM1 into MeJA or mock-treated N. benthamiana plants. Real-time results show no difference on viral DNA accumulation between the two kinds of treatment (Fig 8G). Thus, βC1 may enhance geminivirus infection, at least partially by inhibiting JA pathway through interfering with the function of SCFCOI1.
In this study, we found that CLCuMuB βC1 inhibits the function of SCF ligase to enhance geminivirus DNA accumulation and symptom development by disrupting SKP-CUL1 interaction through its binding to SKP1. In addition, we found that JA treatment improves plant defense against geminivirus infection.
Betasatellites are indispensable for some monopartite geminiviruses to induce viral symptoms in host plants. The sole protein βC1 encoded by several betasatellites, has been reported to be responsible for this phenomenon [1]. However, how βC1 induces viral symptoms remain obscure. CLCuMuB βC1 was previously reported to interact with a tomato ubiquitin conjugating enzyme (UBC), SlUBC3, by its C-terminal myristoylation-like motif [22]. The myristoylation-like motif only exists in CLCuMuB βC1 and its close relative okra leaf curl betasatellite (OLCB βC1). However, OLCB βC1 does not interact with SlUBC3 [22]. Further, silencing of UBC3 in N. benthamiana did not cause any obvious phenotype and enhanced viral DNA accumulation in this study (S18 Fig). Thus, it is possible that symptoms induced by CLCuMuB might not be mediated by interaction between βC1 proteins and host UBC3 enzyme. Here, we demonstrate that CLCuMuB βC1 is also indispensable for symptom production (S2 Fig). Through a series of interaction assays, we found that CLCuMuB βC1 interacts with NbSKP1s, important components of SCF complexes (Fig 1). Further, CLCuMuB βC1 interferes with the interaction between SKP1 and CUL1 (Fig 3) to impair the function of SCF complexes, such as SCFCOI1 and SCFSYL1 (Figs 6 and 7), which is consistent with the previous observation that overexpression of CLCuMuB βC1 in tobacco causes a global reduction of polyubiquitinated proteins [22]. We found that disrupting the function of SCF complexes by silencing of either SKP1 or CUL1 leads to some typical virus symptoms, such as severe leaf curling, crimping, leaf darkening and growth retardation (Figs 4 and 5). Indeed, perturbation of the ubiquitin system can cause leaf curling and vascular tissue abnormalities [47]. Further, overexpression of CLCuMuB βC1 blocked the degradation of GAI (Fig 8), the target of the SCFSLY1, repressed plant responses to GA, which may explained why the presence of CLCuMuB make plant dwarf phenotype. These results suggest that some geminiviral βC1 proteins can elicit viral symptoms by disrupting the plant ubiquitination pathway by interfering with SKP1-CUL1 interaction through its interaction with SKP1.
Although NbSKP1s silencing is in fact causing higher accumulation of viral DNA (Fig 4B1–4B3), the symptoms seem simply due to NbSKP1s silencing but not higher accumulation of virus, because we found higher accumulation of CLCuMuV DNA, but no symptom in plants infected with CLCuMuV and βM2-SKP1-176 which is generated though inserting the 176-bp NbSKP1.1 fragment directly into βM2, without fused with the 169-bp βC1 fragment (S19 Fig).
We noticed that silencing of either SKP1 or CUL1 did not produce all symptoms caused by CLCuMuB βC1. Besides leaf curling, crimping, darkening and growth retardation caused by silencing of either SKP1 or CUL1, the viral symptoms elicited by CLCuMuB βC1 also include bending shoot and enations from abaxial side of leaves. Tomato yellow leaf curl China virus (TYLCCNV) βC1 was reported previously to elicit leaf morphological changes in Arabidopsis by mimicking the functions of ASYMMETRIC LEAVES 2 through its interaction with ASYMMETRIC LEAVES 1 and by repressing the accumulation of miR165/166 to subvert leaf polarity [20]. Meanwhile, suppression of miR165/166 can cause enations from abaxial side of leaves [48]. It is possible that CLCuMuB βC1 induces enations by suppression of miR165/166. Further, TYLCCNV βC1 may also induce viral symptoms by up-regulating the expression of a calmodulin-like protein (rgsCaM) [16]. Considering that geminivirus βC1 is a multiple functional protein, CLCuMuB βC1 may contribute to the viral symptoms by multiple mechanisms including disrupting the plant ubiquitination pathway.
In this study, we demonstrate that CLCuMuB βC1 impairs the interaction between NbSKP1s and NbCUL1 by interacting with NbSKP1s and silencing of either NbSKP1s or NbCUL1 enhances CLCuMuV DNA accumulation. Deletion of CLCuMuB βC1 reduced CLCuMuV titer (S2 Fig). Silencing of either NbSKP1s or NbCUL1 caused enhanced virus accumulation (Figs 4 and 5). Geminiviruses may interfere with plant ubiquitination to suppress plant defense against geminivirus infection [49]. It has been reported that V2 protein of Tomato yellow leaf curl Sardinia virus (TYLCSV) interacts with UBA1, a ubiquitin-activating enzyme, which is a positive regulator of plant defense [50, 51], and silencing of either UBA1 or RHF2a (RING-type E3 ubiquitin ligase) in N. benthamiana enhances TYLCSV infection [50, 52]. Geminiviral C4 activates expression of host RING E3 ligase RKP to ubiquitinate cell cycle inhibitors ICK/KRPs to help the replication of Beet severe curly top virus (BSCTV) via promoting cell division [53, 54]. However, how geminivirus βC1 proteins interfere with plant ubiquitination pathway to enhance viral accumulation is still obscure.
In this study, we found that CLCuMuB βC1 disrupted the integrity of SCFCOI1 (Fig 6D and 6E). Meanwhile CLCuMuB βC1 does not inhibit JA biosynthesis (S14 Fig). More importantly, JA treatment reduces the plant susceptibility to CLCuMuV (Fig 8), which is consistent with the previous observation that JA treatment attenuates the infection of plant with Beet curly top virus (BCTV) [33]. TYLCCNB βC1 was reported to suppress JA-related host defenses for increasing population densities of their whitefly vectors [19, 21]. Further, Cabbage leaf curl virus (CaLCuV) infection can also repress JA response [21, 44]. The C2 proteins of TYLCSV, Tomato yellow leaf curl virus (TYLCV) and BCTV were reported to impair derubylation of SCF E3 ligase complexes and inhibit jasmonate signaling by interacting with CSN5 [20, 33]. Thus, CLCuMuB βC1 could enhance CLCuMuV accumulation, at least partially by repressing JA responses through interfering with plant ubiquitination.
We observed that the levels of CLCuMuV DNA in SKP1- or CUL1-silenced plants were lower than that in the presence of CLCuMuB with functional βC1 although silencing of either SKP1 or CUL1 resulted in a higher accumulation of CLCuMuV DNA (Figs 4 and 5 and S2). It has been reported that knock-down of either CSN5A or CSN3, two components of protein degradation-related CSN complexes, hinders BCTV infection although knockout of Arabidopsis csn5a mutant can partially complement BCTV C2 mutant [50, 52, 55]. Further, overexpression of a given F-box protein can circumvent the general SCF malfunction [56, 57]. These observations suggest that begomoviruses might not only hamper, but also redirect the activity of SCF complexes for begomoviruses propagation [33]. Very recently, ubiquitination is reported to regulate the stability of TYLCCNV βC1 [58]. Thus, host plants, geminiviruses and their satellites may have evolved to exploit the dual roles of the ubiquitination pathway in plant defense and viral pathogenesis to co-survive in their long-term arm races.
The full-length infectious CLCuMuV clone contains 1.7-mer CLCuMuV DNA genome. Two separate DNA fragments were PCR amplified using primer pairs HindIII-A-F/XbaI-A-R, or XbaI-A-F/KpnI-A-R respectively and total DNA extracted from cotton leaf tissues with CLCuD [34] as the template, double-digested with HindIII and XbaI or XbaI and KpnI, and then inserted into pBinplus ARS digested with HindIII and KpnI.
The βDNA infectious clone contains 2-mer CLCuMuB genomes. Two DNA fragments were PCR amplified using primer pairs KpnI-β-F/HindIII-β-R or HindIII-β-F/SacI-β-R respectively and total DNA from cotton samples with CLCuD [34] as the template, digested with KpnI and HindIII or HindIII and SacI, and then inserted into pCAMBIA-2300 digested with KpnI and SacI to generate βDNA.
The null mutant betasatellite vector βM1 was constructed by introducing a ATG-TGA transition in the start codon. βDNA was used as the template. Two DNA fragments were PCR amplified using primer pairs βM1-R/SacI-β-R or HindIII-β-F/βM1-F respectively, then were fused to obtain SacI-βM1-HindIII with ATG-TGA mutation. the other two DNA fragments were PCR amplified using primer pairs HindIII-β-F/βM1-R and βM1-F/KpnI-β-F, then were fused to obtain HindIII-βM1-KpnI with ATG-TGA mutation. digested with SacI and HindIII or HindIII and KpnI, SacI-βM1-HindIII and HindIII-βM1-KpnI were inserted into pCAMBIA-2300 digested with KpnI and SacI to generate βM1.
The T-DNA silencing vector βM2 was constructed by introducing a multiple cloning site to replace the βC1 ORF in CLCuMuB. Two DNA fragments were PCR amplified using primer pairs KpnI-βMF/XbaI-βM2-R or XbaI-βM2-F/SacI-βM2-R respectively using βDNA as the template, digested by KpnI and XbaI or XbaI and SacI, and then inserted into pCAMBIA-2300 digested by KpnI and SacI to generate vector βM2.
DNA fragments of HA-βC1-nYFP, HA-βC1ΔC43-nYFP, HA-cYFP-NbSKP1.1, HA-cYFP-NbSKP1L1, HA-cYFP-nLUC, GFP-βC1, HA-βC1, HA-βC1ΔC43, HA-NbSKP1.1, GFP-NbCUL1, nYFP-SKP1, cYFP-NbCUL1, Myc-COI1 and YFP-GAI were obtained by overlapping PCR. The resulting PCR products were cloned between the duplicated Cauliflower mosaic virus 35S promoter and Nos terminator of pJG045, a pCAMBIA1300-based T-DNA vector [59]. βC1pro:βC1, a βC1expression vector with its native promoter, was generated by inserting 1–1346 nt of CLCuMuB genome (GQ906588) into pCAMBIA-2300. Among these vectors, βC1pro:βC1, 35Spro:GFP-βC1 and 35Spro:HA-βC1 were used to generate transgenic plants respectively. PVX-cLUC, PVX-βC1 and PVX-βC1ΔC43 were constructed by introducing DNA fragments of cLUC, βC1 and βC1ΔC43 into a PVX vector [60]. pBinPLUS-TA and pBinPLUS-2mβ were kindly provided by Professor Xueping Zhou [61]. All constructs were confirmed by DNA sequencing. Primers used in this study were listed in S1 Table.
Total DNA was extracted from apical developing leaves using the DNAsecure Plant Kit (TIANGEN, China). DNA concentration of each sample was calculated through OD260 via Epoch Multi-Volume Spectrophotometer System (Bio-Tek, USA) and then diluted to around 60ng/ul for PCR amplication. A single copy of CLCuMuV genome was amplified by PCR and then was ligased into pMD19-T (TaKaRa, Japan) to generate a CLCuMuV-positive plasmid. A 10-fold serial dilution of the plasmid DNA from 2×108 to 200 copy was prepared and used as the standard. A CLCuMuV-specific primer set (qCLCuMuV V1-F and qCLCuMuV V1-R) was used to amplify a 198-bp amplicon. For SYBR Green-based real-time PCR performed in a 10 μL reaction mixture containing 5 μl Power SYBR Green PCR Master Mix (2×) (Life, USA), primer concentration was optimized by running the assay using the plasmid DNA dilution series with two different primer concentration (10 and 20 μM). 0.1 μL of each 20 μM primer and 0.3 μL 60 ng/μL templet were finally chosen to amplify viral DNA in samples for following assays. Because the standard curves generated were linear in the whole range tested with a coefficient of regression R2:0.99 and calculated slope around -3.5 for SYBR Green assay. The copy number of viral DNA can be calculated via Ct value of each sample and the standard curve.
To obtain the ratio of viral DNA: plant genome DNA, Plant genome DNA can also be calculated via internal reference method. The genome DNA of healthy N.benthamiana was extracted and a 2-fold serial dilution of the genome DNA from 94.5ng to 1.48ng was prepared and used as the standard. An eIF4a-specific primer set (qeIF4a-F and qeIF4a-R) was used to amplify a 60-bp amplicon. Primer concentration was optimized by using the plant genome DNA dilution series with three different primer concentrations (10, 15 and 20 μM). 0.1 μL of each 15 μM primer was finally chosen because the standard curves generated were linear in the whole range tested with a coefficient of regression R2:0.99 and calculated slope around -3.3 for SYBR Green assay. The plant genome DNA can be calculated via Ct value of each sample and the standard curve.
The full-length CLCuMuB βC1 was PCR amplified and cloned into yeast vector pYL302 to generate the LexA DNA binding domain (BD) containing bait vectors BD-CLCuMuB βC1. The full-length NbSKP1.1, NbSKP1.2, NbSKP1.3, NbSKP1L1 and NbSKP1.1 deletion derivatives were PCR amplified and cloned into the B42 activation domain (AD)-containing vector pJG4-5. The yeast two-hybrid prey library containing tomato cDNAs was used to screen CLCuMuB βC1-binding proteins. The yeast two-hybrid screen and interaction assays were performed as described [35].
N. benthamiana plants were grown in pots at 25°C in growth rooms under 16 h light/8 h dark cycle with 60% humidity. Light intensity is 4000 lx. Solt mixed with vermiculite at a 1:1 ratio was used as the substrate for plants to grow. the plants were watered with a nutrient solution.
For CLCuMuB-based VIGS assays, CLCuMuV or βM2 and its derivatives were introduced into Agrobacterium strain GV2260. Agrobacterium cultures containing CLCuMuV or βM2 derivative plasmids were grown overnight at 28°C until OD600 = 2.0, then CLCuMuV with corresponding βM2 derivative vector were mixed at 1: 1 ratio, pelleted, resuspended in infiltration buffer (10 mM MgCl2, 10 mM MES, and 200 μM acetosyringone, pH 5.6) to OD600 = 1.0, kept at room temperature for 4 h and infiltrated into the lower leaf of 6-leaf stage plants using a 1-ml needleless syringe.
For Agrobacterium tumefaciens-mediated transient expression studies, GV2260 strains containing the relevant expression vectors were cultured and prepared as described above, then were infiltrated into N. benthamiana leaves. The infiltrated leaves were detached at 48 to 60 hpi for the corresponding assays. For coexpression, equal amounts of A. tumefaciens cultures were mixed and used for infiltration.
MeJA treatments: a 50 μM MeJA solution or mock solution (ethanol) were applied to 6-week-old N. benthamiana plants by spray every other day from 1 day before the inoculation to 14 dpi.
Citrine YFP-based BiFC was performed as described [36]. The experimental group and corresponding control group should be inoculated in a same leaf to reduce the difference of expression condition. Live plant imaging was performed on a Zeiss LSM710 confocal microscope. Enhanced citrine YFP-derived fluorescence was acquired using 514-nm laser and emission 519- to 587-nm filters. 8-bit confocal images were acquired with an EC Plan-Neofluar 103/0.30 M27 objective for 103 magnification and a Plan-Apochromat 403/0.95 Korr M27 objective for 403 magnification. Images were analyzed with ZEN 2012 Light Edition.
The experimental group and corresponding control group were inoculated in a same leaf. At 48 dpi, images of live plant samples from experimental and corresponding control groups were taken under the same parameters via a Zeiss LSM710 confocal microscope. Software ZEN 2012 was used to measure the fluorescence intensity mean value of an image. 4 independent images for each group were measured and values were analyzed via t-test. Three biological repeats were needed.
Because βC1 protein was reported not stable in vivo and may be degraded through ubiquitin 26S proteasome system (UPS) [20], so in this assay we added MG132, an inhibitor against the 26S proteasome, to improve the accumulation of GFP-βC1. For Co-IP assays, 50 μM MG132 (Sigma, USA) was inoculated into N. benthamiana leaves 12 h before being detached. total proteins from leaves were extracted with a ratio of 1:2 of native extraction buffer 1 [NB1; 50 mM TRIS-MES pH 8.0, 0.5 M sucrose, 1 mM MgCl2, 10 mM EDTA, 5 mM DTT, 50 μM MG132, protease inhibitor cocktail CompleteMini tablets (Roche, http://www.roche.com/)] [62]. Protein extracts were incubated with the GFP-Trap beads (ChromoTek, German) for 2 hours at 4°C, The beads were washed three times with ice-cold NB1 at 4°C. IP samples were analyzed by SDS-PAGE, immunoblotted using anti-HA (CST, USA) and anti-GFP antibodies (Abmart, China) and detected using Pierce ECL western blotting substrate (Thermo, USA).
GST-CLCuMuB βC1 and HA-His-NbSKP1.1 fusion proteins were produced in BL21(DE3) codon plus RIL cells. HA-His-NbSKP1.1 was purified using Ni-NTA Agarose (Qiagen, Netherlands) column. GST-CLCuMuB βC1 was purified using Glutathione Sepharose 4B (GE, USA) and then used to pull down HA-His-NbSKP1.1 in vitro for 2 hours at 4°C. The beads were washed three times with ice-cold elution buffer (300 mM NaCl, 50 mM Tric-HCl, pH 8.0, 0.1% Triton-X 100) at 4°C. The washed beads were boiled in SDS sample buffer, and proteins were separated by SDS-PAGE and detected by western blot using an anti-HA antibody(CST, USA).
His-CLCuMuB βC1 and HA-His-NbSKP1.1 fusion proteins were produced in BL21(DE3) codon plus RIL cells. E. coli cells harboring the corresponding clones were cultured in LB medium (5 mL) containing kanamycin (50 μg/mL) at 37°C, till the O.D. at 600 nm reached 0.6. Then the cells were inoculated for large scale expression. The expression of corresponding genes were induced by the addition of isopropyl-β-D-thiogalactopyranoside (IPTG, Sigma) to the final concentration of 0.2 mM and cells were further allowed to grow for 20 hours at 16°C. The cells were spun down at 4,000 rpm, resuspended in the ice-cold lysis buffer (50 mM Tris-HCl, 300 mM NaCl, 1 mM PMSF, 50 mM DTT, pH 8.5). Resuspended cells were sonicated till suspension became optically clear. HA-His-NbSKP1.1 was soluble and purified using Ni-NTA Agarose (Qiagen, Netherlands) column. His-CLCuMuB βC1 was in inclusion bodies and was dissolved by 8 M Urea (50 mM DTT, 8 M Urea) with a ratio of 0.1g: 1ml. Insoluble substance were removed by centrifugation at 14,000 rpm, 30 min, 4°C. Supernatant was dripped slowly using a 1-ml syringe with needle into 200 mL ice-cold refolding buffer (50 mM Tris-HCl, 300 mM NaCl, 500 mM Arginine, 2 M Urea, 1 mM PMSF, pH 8.5) agitated by a magnetic stirring apparatus. Then this His-CLCuMuB βC1 solution was dialyzed against the dialysis buffer (50 mM Tris-HCl, 300 mM NaCl, pH 8.5). The protein obtained by this method was enriched by Ni-NTA Agarose (Qiagen, Netherlands) column and eluted for further experiments.
1 mL GFP-CUL1 or GFP extracts were prepared and immunoprecipitated by 20 μL GFP-Trap beads (ChromoTek, German) for each sample as described in the Co-Immunoprecipitation (Co-IP) part. After two washes with wash buffer (50 mM Tris-HCl, 300 mM NaCl, 1 mM PMSF, 50 mM DTT, pH 8.5), 1 mL 100 μg/mL E. coli-expressed His-HA-NbSKP1.1 was added and incubated at 4°C for 1 hour. After two washes with wash buffer, 80 μg, 40 μg, 20 μg His-βC1 or 80 μg BSA was added in 1 mL corresponding samples and incubated at 4°C for 1 hour. After three washes with wash buffer, samples were separated by SDS-PAGE, transferred to PVDF membrane, and detected with corresponding antibodies.
The experiments were performed as described by Lozano-Duran [33]. Seeds of wild-type or HA-βC1 transgenic N. benthamiana used in this study were surface sterilized and sown on Murashige and Skoog (MS) agar plates with 30 g/L sucrose and 0.6% Agar. Seedlings were grown at 25°C under 4000 lx white light with a 16-h-light/8-h-dark photoperiod. MS plates were placed in a vertical orientation for 6 d, and seedlings were then transferred to MS plates containing no or 50 μM MeJA (Sigma, USA). Root length was scanned every day until 5 days later.
14–15 days Nicotiana benthamiana plants were inoculated with CA+β or CA+βM1. Leaves in three replicate plants for each treatment were sampled. The leaf materials from each plant were flash-frozen in liquid nitrogen, weighed and stored at -80°C until JA analysis. Sample preparation was performed as described by Glauser and Wolfender, [63]. Except methanol–water, 40:60 (v/v) was used to resolubilize the final residue and do subsequent UHPLC-Q-TOFMS Analysis. Drug Discovery Facility, Center of Biomedical Analysis, Tsinghua University provided the service for sample determination.
GAI was cloned from cDNA of N. benthamiana and the experiments were performed as described by Lozano-Duran [33]. At 48 h past inoculation, the agroinfiltrated leaves were sprayed with a 100 μM GA3 solution or with mock solution (ethanol). Fluorescence was visualized 1 to 2 hours later using a Zeiss LSM710 confocal microscope. Leaf samples were grind by liquid nitrogen, Then total proteins were extracted with a ratio of 1:4 of extraction buffer (50 mM Tris-HCl, 100 mM NaCl, 25 mM imidazole, 10% glycerol, 0.1% Tween-20, 20 mM β-mercaptoethanol) [45]. Samples were separated by SDS-PAGE, transferred to PVDF membrane, and detected with the anti-GFP (ChromoTek, German).
Myc-COI1 was expressed in N. benthamiana and purified as described [45]. 60 μL of purified protein was added to 540 μL of total crude protein extracts (1 mg/mL) from N.benthamiana which was transiently expressed HA-βC1 or HA-nLUC, and then were incubated at 25°C for indicated time periods, separated by SDS-PAGE, transferred to PVDF membrane, and detected with the anti-Myc (Abmart, China).
Total DNA was extracted from apical developing leaves using the DNAsecure Plant Kit (TIANGEN, China). Total RNA was extracted from apical developing leaves using the Trizol reagent (TIANGEN, China) and treated with RNase-free DNase I (Sigma-Aldrich). First strand cDNA was synthesized using 2–5 μg of total RNA with oligo-d(T) primer and M-MLV reverse transcriptase (TIANGEN, China). Real time RT-PCR was performed using Power SYBR Green PCR master mix (Life, USA). EIF4a and Actin were used as internal control for N. benthamiana for normalization. Primers were designed with Primer3web (http://primer3.ut.ee/) and listed in Supplemental Table S1. The values were calculated using the comparative normalized Ct method and all the experiments were repeated at least two times. Data were analyzed and plotted with Origin 8.1.
Sequence data from this article can be found in the GenBank data libraries under accession numbers: CLCuMuV (GQ924756); CLCuMuB (GQ906588); SlSKP1 (XM_004250675); NbSKP1.1 (KP017273); NbSKP1.2 (KP017274); NbSKP1.3 (KP017275); NbSKP1L1 (KP017276); NbCUL1 (KP017277); UBC3 (KR296788); eIF4a (KX247369); Actin (JQ256516); PID (KR082145); COI1 (AF036340); GAI (KR082148); GFP (U87973); Defensin-like protein 1 (KX139060); Defensin-like protein 2 (KX139061); Pathogen like protein (KX139062); Gibberellin- regulated protein 14 (KX139063); Gibberellin-regulated protein 6 (KX139064); SAUR14 (KX139065).
|
10.1371/journal.ppat.1002638 | Anthrax Lethal Factor Cleavage of Nlrp1 Is Required for Activation of the Inflammasome | NOD-like receptor (NLR) proteins (Nlrps) are cytosolic sensors responsible for detection of pathogen and danger-associated molecular patterns through unknown mechanisms. Their activation in response to a wide range of intracellular danger signals leads to formation of the inflammasome, caspase-1 activation, rapid programmed cell death (pyroptosis) and maturation of IL-1β and IL-18. Anthrax lethal toxin (LT) induces the caspase-1-dependent pyroptosis of mouse and rat macrophages isolated from certain inbred rodent strains through activation of the NOD-like receptor (NLR) Nlrp1 inflammasome. Here we show that LT cleaves rat Nlrp1 and this cleavage is required for toxin-induced inflammasome activation, IL-1 β release, and macrophage pyroptosis. These results identify both a previously unrecognized mechanism of activation of an NLR and a new, physiologically relevant protein substrate of LT.
| Anthrax lethal toxin (LT) is a protease which can induce rapid death of macrophages accompanied by activation and release of pro-inflammatory cytokines. The previously identified cellular substrates for this toxin have not been shown to play a role in this rapid cell death. This report identifies a new substrate for LT, and demonstrates that its cleavage by the toxin is required for macrophage death. The substrate, Nlrp1, is a member of a large family of intracellular sensors of danger. These sensors, once activated, form a multiprotein complex called the inflammasome and are essential to the host innate immune response. The mechanism of activation for these sensors is not known. The demonstration of cleavage-mediated activation of Nlrp1 in this study represents the first report on a direct biochemical mechanism for inflammasome activation.
| Anthrax lethal toxin (LT) is a key virulence determinant of Bacillus anthracis. This bipartite toxin consists of the receptor-binding protein protective antigen (PA) and the metalloprotease lethal factor (LF) [1]. LT injection into experimental animals induces a vascular collapse similar to that occurring during anthrax infections. LT also induces rapid lysis of macrophages from certain inbred rodent strains, but macrophage lysis in vivo is not essential for toxin-induced death of mice. LF's only known substrates are the mitogen-activated protein kinase kinases (MEKs 1, 2, 3, 4, 6 and 7). To date, no link between their cleavage and macrophage lysis or animal death has been found [1].
LT induces death in certain inbred and outbred rat strains in as little as 37 minutes [2], [3]. We recently used recombinant inbred rats to map the gene controlling LT sensitivity of both the rats and their macrophages to a region of rat chromosome 10 encoding the NOD-like receptor (NLR) Nlrp1 [3]. The Nlrp proteins are pattern recognition receptors (PRRs) that act as cytoplasmic sensors of danger signals ranging from dividing bacteria to crystalline materials [4]. The activated sensors oligomerize and recruit caspase-1 to a multiprotein complex (the inflammasome). Inflammasome-mediated activation of caspase-1 allows it to cleave the precursors of the inflammatory cytokines IL-1β and IL-18, as well as currently unidentified substrates that cause the rapid lysis of macrophages and dendritic cells (pyroptosis). Unlike other Nlrp sensors that have been shown to be activated by a diverse set of stimuli [4], Nlrp1's only identified activator is LF. The mechanism for activation has not yet been identified for any inflammasome sensor. In this study we report that LF cleaves Nlrp1 and that susceptibility of Nlrp1 to this cleavage dictates sensitivity of macrophages to the pyroptosis induced by this toxin.
Our earlier studies found that polymorphisms within the N-terminus of the rat Nlrp1 (rNlrp1) protein correlate perfectly with LT sensitivity in twelve inbred rat strains [3]. The top two sequences in Fig. 1 are rNlrp1 sequences which represent the previously identified alleles associated with resistance or sensitivity. The CDF Fischer rat (CDF) sequence represents Nlrp1 from LT-sensitive rats (Nlrp1S, abbreviated as S in Fig. 1). The Lewis rat (LEW) sequence represents Nlrp1 from LT-resistant rats (Nlrp1R, abbreviated as R in Fig. 1). Residues 41–48 of the Nlrp1 from LT-sensitive rats include 3 Arg and 2 Pro residues and are entirely different from the largely hydrophobic residues in the corresponding positions of the Nlrp1 of LT-resistant rats. The other 8 aa differences between the resistant and sensitive Nlrp1 proteins from 12 inbred rat strains are distributed throughout the N-terminal 100 aa [3]. Notably, residues 41–48 of the Nlrp1 from LT-sensitive rats fit consensus cleavage motifs of its MEK substrates [5]–[7] (Fig. 1, top row shows one previously identified consensus motif where the locations of basic (positively-charged) residues (B) and hydrophobic residues (h) relative to the cleavage site are shown). We hypothesized that LF cleaves Nlrp1 proteins at this site. Nlrp1 proteins are expressed at low levels and are difficult to detect. Therefore we expressed full-length (1218 aa) N-terminally hemagglutinin (HA) epitope-tagged rNlrp1S (CDF), rNlrp1R (LEW), as well as chimeric protein constructs consisting of aa 1–53 of rNlrp1S and aa 54–1218 of rNlrp1R (CDF53-LEW) and the reciprocal construct (LEW53-CDF) (Fig. 1, lines 3,4) in HT1080 human fibroblasts. The proteins were expressed both transiently and in stable lines selected for maximal expression. Western blots and immunoprecipitation (IP) with anti-HA antibodies showed expression of a full length HA-tagged rNlrp1 (140 kDa) in addition to a shorter protein of about 110 kDa (Fig. 2a, 2b, 2c). This smaller protein, which is often present in equal (data not shown) or larger quantities (Fig. 2a, 2b, 2c) than the full-length rNlrp1, probably results from autoproteolysis within the ZU5-UPA/FIND domain that lies upstream of the CARD (caspase-1 recruitment) domain [8]. LT treatment of fibroblasts expressing the HA-tagged rNlrp1 proteins showed a small quantity of a 6-kDa HA-antibody reactive fragment present in LT-treated CDF (Fig. 2a) and CDF-LEW53 cells (data not shown), but not in LEW lines (Fig. 2a), suggesting a possible cleavage event in proteins with rNlrp1S sequences at their N-termini. HT1080 cells expressing any of the three Nlrp1 variants were all equally viable after the LF treatment (data not shown). Cleavage of MEK3 confirmed that LF was active within the cells under the conditions used (Fig. 2a). Direct LF treatment of sucrose lysates of these cells showed loss of the HA-tag from the full length and 110-kDa CDF and CDF53-LEW proteins, but not from the LEW and LEW53-CDF proteins having N-terminal rNlrp1R sequences (Fig. 2b). Loss of HA western blot reactivity from the 110-kDa proteins was consistent with cleavage occurring within aa 1–53, and pointed to the 8-aa polymorphic sequence as being the cleavage site. Anti-HA IP of the toxin-treated lysates showed a single HA-tagged 6-kDa fragment appearing coincidentally with loss of HA-epitope reactivity from both the 140-kDa and 110-kDa rNlrp1 proteins, further supporting the 8-aa sequence as the site of LF action (Fig. 2c). HA-tagged rNlrp1R (LEW) was not cleaved in response to LT (Fig. 2c). An inactive LF variant (LF E687C) was unable to cleave any of the proteins (data not shown).
To precisely identify the cleavage site, we expressed and purified aa 3–100 of the rNlrp1S (designated CDF100) as a fusion containing an N-terminal 6His-GST tag (Fig. 1, line 7). This CDF100 protein was cleaved efficiently even when using an enzyme (LF): substrate (Nlrp1) molar ratio of 1∶1000 (Fig. 3a). Mass spectrometry analyses of the intact protein and the two cleavage fragments yielded protein sizes of 39720, 33260, and 6478 Da (Fig. S1), showing that LF cleaved the Pro44-Leu45 bond in the sequence RPRP∧LPRV of rNlrp1S. Variants of the tagged CDF100 protein in which the 2 or 4 aa immediately preceding the LF cleavage site were changed to the corresponding rNlrp1R sequence (CDF100(EQ) and CDF100(QVEQ)) (Fig. 1, lines 8,9) were also tested as substrates for LF or the enzymatically inactive LF variant (LF E687C). LF E687C did not cleave any proteins (Fig. 3b). CDF100(EQ) was cleaved by LF at a lower efficiency than was CDF100 and CDF100(QVEQ) was not cleaved (Fig. 3b). Thus, LF cleaves rNlrp1S at a single site and the specificity of LF for rNlrp1 depends at least in part on the aa sequence immediately preceding the bond cleaved.
We next asked whether rNlrp1S cleavage is required for LT-mediated macrophage death. Functional inflammasomes containing various NLR proteins have previously been reconstituted in non-macrophage cell lines such as HEK293 cells by expressing the NLR proteins in conjunction with caspase-1 and using IL-1β as a reporter for activation in response to stimuli. Expression of the mouse paralog Nlrp1b and caspase-1 in HT1080 fibroblasts was also shown to be sufficient for induction of approximately 30% cell death following LT treatment [9]. However, we found that robust transient or stable expression of rNlrp1S in conjunction with rat pro-caspase-1 in HT1080 (or CHO WTP4) cells did not sensitize these cells to LT-mediated cell death (data not shown). Because macrophages may express additional components needed for induction of Nlrp1 inflammasome-mediated cell death, we retrovirally expressed various rNlrp1 proteins in the LT-resistant BMAJ mouse macrophage cell line. This cell line expresses the mouse Nlrp1bR protein. Expression of CDF53-LEW in the BMAJ macrophages sensitized them to LT-induced cell death (Fig. 4a). Expression of the CDF53-LEW protein containing the 2 aa substitutions preceding the cleavage site (CDF53(EQ)-LEW) resulted in partial sensitization, while substitution of 4 aa (CDF53(QVEQ)-LEW abrogated sensitization (Fig. 4a). These data strongly implicate the proteolytic cleavage of rNlrp1 by LF as a key step in inflammasome activation and induction of pyroptosis in macrophages. Coincident with cell death, there were equally robust releases of IL-1β from the cells expressing either CDF53-LEW or CDF53(EQ)-LEW (Fig. 4b), in spite of the weaker sensitization of the latter to LT-induced cell death (Fig. 4a). This indicates that there may be different efficiency and timing requirements for the LT-induced caspase-1 activation events that lead to cell death and to IL-1β processing. Supporting this hypothesis, a recent report showed that a single stimulus can be sensed by a single NLR, but activate distinctly different inflammasomes, which activate caspase-1 through different mechanisms [10].
Both mouse and rat Nlrp1 inflammasome activation and the ensuing macrophage death in response to LT can be prevented by a variety of treatments, including thermal shock [11], or inhibition of caspase-1 [12]–[14], cathepsin B [14], [15], and proteasome activity [12], [14], [16]. These treatments all act downstream of LF endocytosis and delivery to the cytosol, as evidenced by their failure to prevent cleavage of the MEK substrates. We tested a panel of these inhibitors for their abilities to protect BMAJ cells expressing CDF53-LEW against LT-induced cell death and to prevent IL-1β release. Protection was provided by nearly all the inhibitors (Fig. 4c,d), the exceptions being the proteasome inhibitors, which are usually highly protective against LT toxicity toward mouse macrophage cell lines [12], [14], [16]. This poor protection may be linked to the longer times (9–10 h) required to achieve complete death of CDF53-LEW-expressing BMAJ cells compared to CDF rat bone marrow-derived macrophages (which succumb in 3 h). The presence of endogenous mouse Nlrp1bR in BMAJ cells likely contributes to the slower death of the transformed cells.
We present here data indicating that cleavage of rNlrp1 by anthrax LT, in an N-terminal domain having no known function, leads to inflammasome activation and macrophage death. This represents the first report on biochemical modification of an inflammasome sensor by a bacterial protein as a mechanism of inflammasome activation. Recent reports demonstrate inflammasome activation mediated by binding of flagellin to NAIP sensor proteins as a simple receptor-ligand activation mechanism that allows for interaction of NAIP proteins with Nlrc4 and caspase-1 in a large oligomer [17], [18]. It remains to be determined how cleavage activates the Nlrp1 inflammasome. Cleavage may induce conformational changes, further proteolytic/autoproteolytic events, release from inhibitory proteins, or alteration in cytosolic localization. Furthermore, it remains to be seen whether cleavage of Nlrp1 directly activates the Nlrp1 inflammasome, or simply allows for subsequent biochemical or cellular events that allow for this sensor or another NLR to recruit and activate caspase-1.
Rodent Nlrp1 proteins differ from most NLRs in that they do not contain a pyrin domain, a domain that is found at the N-terminus of human Nlrp1. This pyrin domain is required for association of NLR proteins with the scaffold protein, ASC, which is required for caspase-1 autoproteolysis and IL-1β processing, but not for caspase-1 mediated cell death [10]. The data presented here raises the question whether mouse or human Nlrp1, or other NLRs, can also be activated by LF through similar cleavage-based mechanisms. Although mouse and human Nlrp1 proteins do not contain the same sequence found in Nlrp1 of sensitive rats, it is possible that LF can also cleave these proteins at a different site in the N-terminal domain or elsewhere. Furthermore, one can speculate that these and other NLR may be activated through cleavage-based mechanisms by other cellular proteases responding to the stimuli associated with inflammasome activation, or even by a conformational change-mediated autoproteolytic event, in a manner similar to caspase-1. An autoproteolytic capability of Nlrp1 was recently reported [8], although a link between this cleavage and inflammasome activation was not demonstrated. It would not be surprising, however, if cleavage of Nlrp1 allowing release of the CARD domain leads to constitutive activation of caspase-1. Expression of the CARD domain of murine Nlrp1 has been shown to result in constitutive caspase-1 activation [9], and deletion of the leucine-rich repeat of Nlrc4 also leads to constitutive caspase-1 activation [17].
This report also identifies a new LF substrate, distinct from the only previously-identified physiologically relevant LF substrates, the MEKs. Although the identification of a substrate with a role in a cell death (i.e., pyroptotic) pathway now explains how the toxin induces rapid macrophage lysis, a role for this pyroptosis in the rapid rat death induced by LT remains to be elucidated.
An increasing number of bacterial virulence factors have been identified which act catalytically to perturb normal cellular processes. In some cases, the catalytic activity mimics activities already present in the cell, as in the case of anthrax edema factor adenylate cyclase, which produces an endogenous signaling molecule, cAMP [19]. In other cases, the bacterial enzyme catalyzes a reaction that inactivates a host process. In the case of the LF-induced cleavage of NLR protein demonstrated here, it is not yet clear whether this cleavage mimics a natural process of inflammasome activation or whether it is an event with no parallel in a normal process. Resolution of this question will come only after there is a better understanding of the mechanisms of inflammasome activation. As has occurred through the study of other bacterial toxins, the further analysis of LF's effects may aid in deciphering the biochemical basis of a key cellular process - in this case, inflammasome activation.
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. All bone marrow harvests were from euthanized rats and were performed in accordance to protocols approved by the NIAID Animal Care and Use Committee (approved protocols OSD1, 3).
PA, LF, and LF E687C were purified from B. anthracis as described previously [20]. Concentrations of LT correspond to the concentration of each toxin component (i.e., 1 µg/ml LT has 1 µg/ml PA and 1 µg/ml LF). 6His-GST-fused rNlrp1 proteins were expressed in plasmid pEX-N-His-GST (cat# PS100028, Origene, Rockville, MD) and purified on glutathione-Sepharose columns by standard methods. High affinity anti-HA (cat# 11867423001, Roche Diagnostics, Indianapolis, IN), anti-Mek3 (cat# sc-959, Santa Cruz BT, Santa Cruz, CA), anti-IL-1β (cat# AF-401-NA, R&D Systems, Minneapolis, MN) and various IR-dye conjugated secondary antibodies (Licor Biosciences, Lincoln, NE and Rockland Immunochemicals, Gilbertsville, PA) were purchased.
BMAJ mouse macrophages (gift of D. Radzioch), Phoenix 293T (gift from Iain Fraser, NIAID, NIH), HT1080, and L929 cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) with 10% fetal bovine serum, 10 mM HEPES and 100 µg/ml gentamicin (all obtained from Invitrogen, Carlsbad, CA) at 37°C in 5% CO2. CHO WTP4 cells were grown in AMEM with the same supplements. For certain experiments, cells were heat shocked (42°C) for 1 h prior to toxin treatments. For all transfections, endotoxin-free plasmid DNA was prepared (Endo-Free Maxiprep, Qiagen, Valencia, CA) and cells were transfected with TurboFect reagent (Fermentas, Glen Burnie, MD) according to the manufacturer's instructions. For transient transfections, cells were treated with LT 24 h after transfection [3]. Transfected cells were also lysed at the same time point in RIPA buffer and subjected to Western blot (WB) analyses to verify expression of rNlrp1, caspase-1 or IL-1β proteins, as previously described [12]. For stable transfections, HA-rNlrp1 overexpressing cell lines were derived by selection with hygromycin B (500 µg/ml; Invitrogen) for 12 days. High expressing clones were identified by Western blot with anti-HA antibody. For retroviral transduction of macrophages, Phoenix 293T packaging cells were transfected first, as above and at 24 h, culture media were replaced with DMEM. At 48 h, virus-containing media was removed, filtered (0.45 µm membrane, Millipore, Billerica, MA) and supplemented with polybrene at 10 µg/ml (Millipore) prior to incubation with BMAJ macrophages. Virus-containing media were removed 24 h later and replaced with DMEM. Stably-transfected cells were selected over 2–3 weeks in the continuous presence of G418 sulfate (500 µg/ml, Invitrogen). Bone marrow-derived macrophages (BMDMs) used as a source of cDNA were obtained from rats purchased from Charles River Laboratories (Wilmington, MA) and cultured as previously described [12].
cDNA was obtained from different rat strains by isolating RNA from BMDMs [11], [12], [14]. To construct N-terminally hemagglutinin (HA) tagged rNlrp1, an N-terminal HA tag was incorporated into forward primers along with two restriction sites (5′ NheI and 3′ EcoRV- for primer table, see Table S1). For rNlrp1 from LT-sensitive Fischer (CDF) and LT-resistant Lewis (LEW) rats, full length rNlrp1 was amplified from CDF cDNA with Ex Taq polymerase (TAKARA Bio Inc., Otsu, Japan) and TOPO TA cloned into the pCR2.1-TOPO vector (Invitrogen). Full length rNlrp1 was amplified from LEW cDNA with Phusion polymerase (Finnzymes, Woburn, MA) and blunt-end TOPO cloned into the pCR-Blunt II-TOPO vector (Invitrogen). NheI/EcoRV digested fragments were ligated into the pIREShyg3 mammalian expression vector (Clontech, Mountain View, CA). CDF53-LEW- and LEW53-CDF chimeric constructs were made by digestion of CDF-pIREShygV3 and LEW-pIREShygV3 with XcmI/XbaI (New England Biolabs, Ipswich, MA), gel purification of fragments, followed by ligation to yield the appropriate fused constructs. For retroviral expression, 5′ BamHI and 3′ NotI sites were introduced into full length CDF53-LEW and LEW rNlrp1 amplified from CDF53-LEW-pIREShygV3 and LEW-pIREShygV3 using Phusion polymerase (New England Biolabs). PCR products were cloned into the CloneJET vector (Fermentas) prior to BamHI/NotI digestion and ligation into the pFB-NEO retroviral vector (Agilent Technologies, Santa Clara, CA). QuikChange (Agilent Technologies) was used to introduce 4 and 7 nucleotide changes into CDF53-LEW-pFB-NEO to create CDF53(EQ)-LEW-pFB-NEO and CDF53(QVEQ)-LEW-pFB-NEO. CDF100 was synthesized by Blue Heron Biotechnologies (Bothell, WA) and cloned into the pEX-N-6HIS-GST vector using AsiSI and AscI sites. The sequence of this construct differs from the originally reported CDF rNlrp1sequence at one residue, K61N, resulting in the rNlrp1S allele 1 sequence [3]. The QuikChange system (Agilent Technologies) was used to introduce 4 and 8 nucleotide changes. All clones were identified by restriction digests and verified by sequencing (Macrogen, Rockville, MD).
WB were performed using either anti-HA (1∶1000), anti-MEK3NT (1∶500) or anti-IL-1β (1∶2,500) and proteins were detected using the Odyssey Infrared Imaging System (Licor Biosciences) [11], [12], [14]. For immunoprecipitation (IP) experiments, anti-HA antibody (Roche Diagnostics) was added to each lysate sample (5–15 µg/ml) and samples were continuously mixed by rotation at 4°C for 1 h. Protein A/G agarose (Santa Cruz Biotechnology) was added, and samples were incubated at 4°C overnight with rotation. Beads were centrifuged at 4,000 rpm for 2 min and washed with 10 mM HEPES three times prior to elution of proteins using SDS loading buffer (10% SDS, 0.6 M DTT, 30% glycerol, 0.012% bromophenol blue, at 90°C, 5 min).
To assess Nlrp1 cleavage in cell lysates, cells were grown to confluence in 10-cm tissue culture dishes. For canonical cleavage, cells in two confluent plates were treated with LT, detached by trypsinization, washed 2 times (PBS, 1000 rpm for 5 min) and lysed in 130 µl of sucrose buffer (250 mM sucrose, 10 mM HEPES, 0.05 M EDTA, 0.2% Nonidet-P40) containing 5 ng/ml LF inhibitor PT-168541-1 (gift of Alan Johnson, Panthera Biopharma) followed by addition of 60 µl SDS loading buffer (90°C, 5 min). For cleavage by LF treatment of pre-lysed cells, 5 confluent dishes were detached by trypsinization and washed 3 times (PBS, 1000 rpm for 5 min) and resuspended in 2.5 ml of sucrose buffer (250 mM sucrose, 10 mM HEPES). Cells were centrifuged (2500 rpm, 10 min), resuspended in 350 µl sucrose buffer containing 0.2% Nonidet-P40 and incubated on ice 30 min. All samples were extensively syringed by passage through a 29 gauge hub-less syringe (Terumo, Somerset, NJ) until >99% of all cells were lysed. ZnCl2 and NaCl were added to final concentrations of 1 µM and 5 mM, respectively. Lysates were treated with LF for 3.5 h at 37°C. Cleavage was analyzed by WB or IP/WB. For in vitro cleavage assays with purified proteins, CDF100, CDF100(EQ) and CDF100(QVEQ) (at final concentrations of 1 mg/ml were incubated in cleavage buffer (1 µM ZnCl2, 5 mM NaCl, 10 mM HEPES for 3 h at 37°C) with purified LF at varying concentrations. Samples were separated on an 8–25% SDS-PAGE gel using the PhastSystem (GE Life Sciences, Piscataway, NJ) and stained with Coomassie blue.
All toxicity and protection assays were modified from methods previously described [11], [12], [14] with minor modifications. Drugs were tested over concentrations ranging from 0.1–100 µM and were added 1 h prior to toxin (cathepsin B and caspase-1 inhibitors) or 5 h post-toxin (proteasome inhibitors) treatment. LT treatment was performed with a concentration of 5 µg/ml (9 h). Cell viability was assessed by staining with MTT [3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl tetrazolium bromide] as previously described.
Cells were pretreated with 0.5–1 µg/ml LPS (Calbiochem, San Diego, CA) for 2 h prior to toxin treatment. LT was added for 7 h at varying concentrations. Cytokine in culture supernatants was measured using an IL-1β ELISA kit (R&D Systems) according to the manufacturer's instructions.
The molecular masses of the H6-GST-LEW and CDF proteins and their cleavage products were determined by liquid chromatography-electrospray mass spectrometry using an HP/Agilent 1100 MSD instrument (Hewlett Packard, Palo Alto, CA) at the NIDDK core facility, Bethesda, MD.
|
10.1371/journal.ppat.1004581 | Systemic Expression of Kaposi Sarcoma Herpesvirus (KSHV) Vflip in Endothelial Cells Leads to a Profound Proinflammatory Phenotype and Myeloid Lineage Remodeling In Vivo | KSHV is the causative agent of Kaposi sarcoma (KS), a spindle-shaped endothelial cell neoplasm accompanied by an inflammatory infiltrate. To evaluate the role of KSHV vFLIP in the pathogenesis of KS, we constructed mice with inducible expression of vFLIP in endothelial cells. Abnormal cells with endothelial marker expression and fusiform appearance were observed in several tissues reminiscent of the spindle cells found in KS. Serum cytokines displayed a profound perturbation similar to that described in KSHV inflammatory cytokine syndrome (KICS), a recently described clinical condition characterized by elevated IL6 and IL10. An increased myeloid component with suppressive immune phenotype was found, which may contribute to functional changes in the microenvironment and cellular heterogeneity as observed in KS. These mice represent the first in vivo demonstration that vFLIP is capable of inducing vascular abnormalities and changes in host microenvironment with important implications for understanding the pathogenesis and treating KSHV-associated diseases.
| Kaposi’s sarcoma (KS) is the most common cancer in men infected with HIV, and also among the most frequent malignancies in Sub-Equatorial Africa. KS is a tumor of endothelial cell origin that is caused by infection with a gamma-herpesvirus, called KS herpesvirus (KSHV) or human herpesvirus 8 (HHV-8). KSHV vFLIP is a viral oncoprotein expressed during latent infection. We report here the generation and characterization of mice expressing KSHV vFLIP in an inducible manner in endothelial cells. Transgenic mice showed: 1) systemic endothelial abnormalities, with the presence of fusiform cells reminiscent of the spindle cells found in KS, 2) development of a profound perturbation in serum cytokines, reminiscent of the cytokine storm characteristic of KSHV-associated cytokine syndrome (KICS), and 3) remodeling of myeloid differentiation with expansion of myeloid cells displaying a suppressive immunophenotype that potentially favors host immune evasion, angiogenesis and tumor progression. This is the first example of significant changes in myeloid differentiation, vascular abnormalities and cytokine perturbation entirely initiated by ectopic expression of a single viral gene, making this mouse model a useful system to dissect the mechanisms viruses use to manipulate the host microenvironment culminating in sabotage of immunity and development of vascular lesions.
| Kaposi sarcoma herpesvirus (KSHV), also called human herpersvirus 8 (HHV-8), one of the most recently discovered human oncoviruses [1], displays tropism for different cell types and a dual oncogenic role, both in lymphomagenesis and vascular oncogenesis. KSHV is specifically associated with Kaposi sarcoma (KS) and two B-cell lymphoproliferative diseases, namely primary effusion lymphoma (PEL) and a large subset of cases of multicentric Castleman’s disease (MCD) [1–3]. KSHV is also associated with KSHV inflammatory cytokine syndrome (KICS), a newly described clinical condition characterized by systemic illness, poor prognosis, elevated KSHV titers, increased levels of viral IL6 and IL10 comparable to those seen in KSHV–MCD but lacking the characteristic lymphadenopathy of KSHV–MCD [4,5], and KSHV-associated hemophagocytic syndrome (VAHS), an extremely rare syndrome reported in immunocompromised patients with MCD and markedly elevated levels of serum human IL6 [6]. KSHV has been found associated also with POEMS syndrome, a rare multisystemic nosological entity characterized by polyneuropathy, organomegaly (particularly cardiomyopathy), endocrinopathy, monoclonal gammopathy and skin lesions [7]; however, a role for KSHV in this disease is controversial, and POEMS may be part of the spectrum of the inflammatory abnormalities seen in MCD, whether KSHV-associated or not.
Similarly to other related herpesviruses, there is dependency on latency for transformation, although this dogma encountered exceptions and has been subjected to debate [8–11]. KSHV genes regulating viral genomic persistence and capable of inducing cellular transformation are transcribed during latency (i.e., LANA, v-cyclin, vFLIP), and the KSHV mode of infection is predominantly latent in KSHV-induced tumors [12]. Experimental data indicate a role for the viral FLICE-inhibitory protein (vFLIP) in KSHV pathogenesis, as it is a latent gene capable of activating NF-κB [13,14], a hallmark cellular pathway constitutively active in PEL and indispensable for the maintenance of lymphoma cell survival [15–17]. FLIP proteins are a group of cellular and viral proteins identified as inhibitors of death-receptor (DR)-induced apoptosis [18,19]. They contain two death effector domains (DED) capable of inhibiting DED-DED interactions between FAS-associated protein with death domain (FADD) and pro-caspase 8 and 10 within the death-inducing signaling complex (DISC) responsible for DR-induced apoptosis [20]. Based on the homology of KSHV vFLIP with cFLIP proteins, it has been thought that vFLIP becomes part of the DISC, preventing the recruitment and processing of procaspase 8 and, thereby, FAS-induced apoptosis [19], although there is little experimental evidence supporting this direct role in apoptosis inhibition. Nonetheless, it is clear is that vFLIP directly binds to IκB kinase (IKK) γ, inducing IKKα/β phosphorylation, IκBα degradation and p100 cleavage, resulting in the activation of both the classical and alternative NF-κB pathways [13,14,21]. Another established function of vFLIP is inhibition of cell death by blocking autophagy [22].
Several groups have developed mice expressing vFLIP in B-cells [23–25]. Among these, our group used a Cre-Lox recombination approach to express vFLIP in all B-cells and specifically in germinal center B-cells, confirming its role in lymphomagenesis and defining the in vivo immunological functions of vFLIP as an abrogator of germinal center formation and immunoglobulin (Ig) maturation [23]. Tumors occurring in mice expressing vFLIP in B cells retain major features of PEL, namely B-cell origin, as formally demonstrated by the presence of monoclonal Ig gene rearrangements, and remodeling of BCR with downregulation of B-cell markers, including CD19 and lambda. However, these tumors were also characterized by expression of histiocytic/dendritic cell (DC) antigens, consistent with transdifferentiation from B-cells into the myeloid lineage, without excluding a coexisting paracrine effect on the surrounding myeloid cells [23]. Notably, KS lesions are characterized by the presence of inflammatory cells, including numerous histiocytes [26]. Thus, induction of myeloid cell proliferation by vFLIP could be part of the cellular events and microenvironment alterations that occur during KS pathogenesis.
The role of vFLIP in vascular oncogenesis is suggested by the in vitro observations that vFLIP induces spindle cell morphology and expression of inflammatory cytokines in endothelial cells and phosphorylation of STAT1 and STAT2|[27–29]. Both spindling and a proinflammatory microenvironment are key features of KS, defined as a chronic inflammation-associated malignancy due to the presence of spindle-shaped endothelial cells, slit-like neovascular structures, and abnormal vascular spaces with extravasation of red blood cells, as well as variable quantities of infiltrating inflammatory cells and secretion of angiogenic and inflammatory cytokines such as VEGF, PDGF, bFGF, TGFβ, IL1β, IL6 and INFγ [30].
However, the role of vFLIP in the initiation of KSHV-related vascular pathogenesis, if any, is largely unknown. A substantial number of studies have indicated that the cell of origin of KS spindle cells is of endothelial origin as these cells express both blood (e.g., CD34) and lymphatic (e.g., VEGFR3, podoplasmin, LYVE-1, Prox1) endothelial cell markers (BEC, LEC) [31–34] and display a gene signature that falls in between the two cell types, albeit closer to LEC [35]. KSHV can infect both BECs and LECs and is capable of reprogramming their transcriptomes to make BECs more alike LECs and viceversa [35–37]. Therefore, to address the role of vFLIP in vascular oncogenesis, we generated mice that express vFLIP under the control of VE-Cadherin promoter, which has been reported to be active in both BECs and LECs [38]. These transgenic (TG) mice showed systemic endothelial alterations with increased spindle-like cells and changes in serum cytokines, reminiscent of certain features of KS and KICS. We also observed remodeling of myeloid differentiation toward cell types known to have implications in host microenvironment, tumor immune evasion, angiogenesis and vascular lesion development.
We generated mice expressing vFLIP in endothelial cells by using a recombinant inducible system. Previously generated conditional mice for vFLIP (ROSA26.vFLIP knock-in mice) [23] were bred with mice expressing cre recombinase in the form of a fusion protein with the estrogen receptor under the transcriptional control of VE-Cadherin promoter (Cdh5(PAC).creERT2 mice) [39], thus resulting in vFLIP expression in endothelial cells upon tamoxifen treatment (Fig. 1A). Before generating ROSA26.vFLIP;Cdh5(PAC).creERT2 TG mice, we tried to constitutively express vFLIP in endothelial cells by crossing ROSA26.vFLIP mice with mice expressing cre recombinase under the control of Tie2 promoter. However, embryonic lethality was observed, suggesting that constitutive expression of vFLIP is detrimental for embryogenesis and incompatible with life. Instead, the inducible ROSA26.vFLIP;Cdh5(PAC).creERT2 TG mice (carrying both cre and vFLIP) were born at the expected Mendelian frequency and were indistinguishable from their wild-type (WT) littermate controls (carrying only vFLIP) in terms of fertility and developmental features. Expression of vFLIP was evaluated in 2–3 month-old mice, approximately one month after intra-peritoneal (i.p.) injection of tamoxifen in both TG and littermate control mice. vFLIP expression was detected at the RNA and protein level in lung, spleen, liver and heart (Fig. 1B).
The level of vFLIP expression was assessed by quantitative real-time RT-PCR in lung, spleen, liver and heart derived from both TG and controls mice, as well as in BC3 PEL cell line and primary KS with lymph node involvement (Fig. 1C). As expected, the highest level of expression was observed in BC3, where all the cells harbor KSHV multiple copies of the viral genome. The splenic fraction derived from B-cell-specific TG mice show also high level of vFLIP expression, comparable with BC3, and this reflects with the high percentage of B-cells in the spleen and the fact that vFLIP expression is controlled by a strong promoter (i.e., CD19). Instead, the endothelial-specific TG mice express lower levels of vFLIP, although comparable with vFLIP expression seen in primary KS. This is consistent with the percentage of endothelial cells in the organs analyzed, which is lower than the percentage of splenic B-cells.
Since antibodies to vFLIP are not adequate for immunohistochemistry or flow cytometry, we monitored transgene expression using antibodies to EGFP, which is expressed in a common transcript with vFLIP due to the insertion of an IRES between the two gene sequences (Fig. 1A). EGFP was detected by immunohistochemistry in cells lining vascular spaces and with the morphologic appearance of endothelial cells in different organs, including intestine (S1A Fig.); these cells were also positive for the endothelial marker CD34. While B-cell-specific TG mice expressed vFLIP in the splenic B-zone as expected, the endothelial-specific TG mice expressed vFLIP in the vascularized interfollicular area (S1B Fig.).
The endothelial identity of transgenic cells and the endothelial specificity of vFLIP expression was confirmed by flow cytometry performed in the heart, where EGFP was expressed in the vast majority (70.8% ± 1.4%), of endothelial cells defined as CD45−CD11b−CD31+ (Fig. 1D, middle panel), but not in splenic B-cells (Fig. 1E, middle panel). Conversely, B-cell specific vFLIP TG mice expressed EGFP in splenic B-cells (Fig. 1E, right panel), but not in endothelial cells (Fig. 1D, right panel). Taken together, these data showed that vFLIP had the expected pattern of expression restricted to endothelial cells.
Virtually all organs and tissues were affected by pathological changes ultimately related to endothelial dysregulation. Numerous elongated cells frequently lining poorly formed vascular spaces was diffusely found throughout several organs, but most notably in the myocardial parenchyma of TG mice. In the heart, these endothelial cells lined the capillaries surrounding individual myofibers, but also proliferated into the parenchyma, expressed vFLIP and many retained endothelial markers (CD34 and/or CD31) and expressed Ki67 (Fig. 2A). Similar findings with the presence of spindle cells, and plump endothelial cells lining vascular spaces, were found in several organs including skeletal muscles (Fig. 2B), brown fat (Fig. 2C) and brain (S2 Fig.). These proliferating endothelial cells do not express the lymphatic marker PROX1, in spite of successful staining of lymphatic endothelial cells in sites where these normally occur including skin, intestines and splenic red pulp (S3A-S3C Fig.). An abnormal perineurial proliferation of endothelial-like cells was found in several tissues in the TG mice, including perirenal capsule, diaphragm muscle, salivary gland, pancreas, but not in the controls (Fig. 3). The abnormalities observed in the pancreas prompted us to check for signs of endocrinopathy (e.g., diabetes); serum glucose levels were slightly increased, but the difference was not significant (Fig. 3). On the abdominal side of diaphragm and in the peripancreatic region, few nerve bundles were also surrounded by hyperplastic perineurial cells, mixed inflammatory cells, lymphocytes and plasma cells and inflammation extended to the adjacent connective tissue. Chronic inflammation, documented with the presence of mixed cell infiltrate of neutrophils, lymphocytes, plasma cells and histiocytes, was found in several tissues, including the peritoneum (Fig. 4A), meninges (Fig. 4A), kidney and skeletal muscle. Both kidneys showed subcapsular areas with numerous spindle cells (Fig. 3), and the perirenal fat was infiltrated by neutrophils, lymphocytes, plasma cells and histiocytes. Extra-medullary hematopoiesis, with both erythroid and myeloid hyperplasia, was present in the spleen and liver. Peripheral blood analysis showed that TG mice have left-shift (i.e., high metamyelocytes and bands with normal neutrophil count), suggesting a demand for neutrophils that exceeded their production and release, a scenario usually seen in case of chronic inflammation at different anatomic sites as observed in our TG mice.
The mice were viable after tamoxifen administration, but starting as early as few weeks after induction they developed the pathological abnormalities here described and by the age of 3–4 months more than 60% of mice had died (Fig. 4B) as result of a systemic illness that comprised myocardial, meningeal, skeletal muscular, peritoneal and perineurial pathological changes. Although i) the pattern of cytokine perturbation indicates the existence of M2-type polarization, which eventually favors immune suppression and tumor immune evasion rather than autoimmunity, [ii) vFLIP does not appear to be a particularly immunogenic protein and iii) KSHV, in general, has developed a wide array of strategies to evade the host immune responses, the mice were not exposed to the transgene during their embryonic development and, thus, they could have theoretically developed immune response toward vFLIP, resulting in a pseudo-autoimmunity that could partially account for the pathological findings and the poor mouse overall survival. Thus, we assessed the presence of a humoral immune response against vFLIP by immunoblotting, but no cross-reactivity was found between a pool of mouse sera isolated from seven TG mice and whole cell lysates derived from lung, spleen, liver and heart of both TG and control mice (Fig. 4C).
In vitro ectopic expression of vFLIP in either endothelial or B-cells has been shown to confer a myeloid-prone gene expression profile with production of cytokines that have potential tropism for myeloid cells [28,40]. To assess whether in vivo expression of vFLIP is capable of exerting similar effects, a panel of fourteen cytokines and growth factors (IL10, IL6, INFγ, IL1β, IL12p70, TNF, IL4, IL2, IL13, GM-CSF, Phospho Stat1, RANTES, IL12/IL23p40, MCP1) was tested in serum samples collected from mice one month after vFLIP induction by tamoxifen (Fig. 5). We used a flow cytometry bead-based assay, which provides quantitative data and is linear within a large range of concentration (from 30 fg/ml to 200000 fg/ml) (S4 Fig.). Compared to control mice, vFLIP TG mice showed increase of IL10, IL6, IL2, IL13, INFγ, TNF, MCP1 and RANTES. These findings are in line with in vitro data on gene expression profiling obtained in PEL and endothelial cells that ascribed to vFLIP the ability to activate the expression of several cytokines and growth factors potentially implicated in remodeling of the tumor microenvironment by myeloid cells [28,40]. Noteworthy, the systemic illness with poor prognosis and the profound changes in cytokines profile, particularly with increased IL6 and IL10, are aspects similar to those described for MCD and KICS.
To gain insights into the mechanism and consequences of the cytokine storm and further assess the effect of transgene expression in vivo, myeloid differentiation was analyzed by flow cytometry with particular emphasis at the cell subsets that could be influenced by or responsible for the observed cytokine perturbation. A large increase in number of CD45+CD11b+Gr1+/− cells was found in lung, spleen, liver and heart, both in endothelial and B-cell specific vFLIP TG mice (Fig. 6A). A more detailed analysis revealed that the myeloid subpopulation preferentially expanded was Ly6G+Ly6Cint (Fig. 6B). These cells were large (FSChigh), have high granularity (SSChigh) and expressed high levels of Gr1, therefore they likely represent granulocytic myeloid derived suppressor cells (MDSCs) (also called polymorphonuclear-MDSCs, PMN-MDSCs), as opposed to monocytic-MDSCs (Ly6GintLy6C+) that lack granularity and express lower level of Gr1 [41–43]. Moreover, Ly6G−Ly6C− cell population, which was expanded only in lung and heart, represented cells that did not express Gr1, were smaller, had no granularity and potentially represent tumor associated macrophages (TAM) or DCs based on immunophenotype, although a functional characterization would be necessary to confirm this. Considering that the endothelial cells are a component of the hematopoietic niches, we compared bone marrow from control and TG mice, but no abnormalities in myeloid or lymphoid hematopoiesis were found that could be ascribed to the expression of vFLIP in endothelial cells.
The tumor microenvironment has been shown to be deeply affected by myeloid cells, including CD11b+Gr1+ cells, which are able to produce soluble factors, such as Bv8, that influence angiogenesis, extracellular matrix remodeling, anti-VEGF resistance and mobilization of additional myeloid cells toward premetastatic sites [44]. Therefore, we checked whether the expanded myeloid cell subpopulations were differentially expressing any these factors. No differences were observed in TG versus WT mice in the levels of expression of Bv8, VEGF and MMP9, indicating that these cell subsets exert their function in vFLIP-mediated pathogenesis through different mechanisms.
In this study, we have investigated the effect of inducible recombinant vFLIP expression in endothelial cells to model KSHV-associated vascular pathogenesis as observed in KS. Mice developed pathological abnormalities with systemic changes and appearance of elongated spindle-like endothelial cells, mimicking aspects of KS and other KSHV-associated diseases. Mice developed a profound proinflammatory phenotype with perturbation of serum cytokines, similarly to KICS, as well as expansion of myeloid cells, which unveiled a key role of vFLIP in initiating a cascade of events that lead to changes in host microenvironment, ultimately favoring tumor immune evasion, angiogenesis and tumor progression during KSHV pathogenesis.
Given the evidence that KSHV can infect both BECs and LECs [35–37], and vFLIP induces spindling of endothelial cells in vitro [27], we tested the hypothesis that in vivo expression of vFLIP in endothelial cells would lead to the development of KS-like disease. Mice developed vascular abnormalities with the presence of spindle cells expressing endothelial antigens in virtually all organs, but, unexpectedly, not in the skin, which is the most common location for KS in humans. While the reasons for this finding are unclear, other viral genes are likely to contribute to the many aspects of KSHV pathogenesis in humans in the context of natural infection, including specific organ involvement [45]. Nevertheless, the endothelial-specific vFLIP TG mice generated showed a proliferation of spindle cells, and a proinflammatory phenotype, indicating that this characteristic of KS can be induced by vFLIP alone. In this setting, vFLIP induces expression of cytokines including those that can result in formation of autocrine loops. For example, there is increased production of IL2, and the IL2 receptor alpha chain is upregulated by NF-κB [46], which is turn is activated by vFLIP. Similarly, there is an increase of TNF production in the vFLIP TG mice, and the TNF receptor (CD120B) molecule is induced by NF-κB [47], which in turn can further activate the NF-κB pathway creating a positive regulatory loop. However, we did not obtain complete KS phenotype, so it is likely that cooperation with other KSHV proteins (e.g., vGPCR, LANA, vCyclin, vIL-6, K1) and/or noncoding transcripts (e.g., miR 17–92, miR K12-7), which are co-expressed in KS and relevant for vascular tumorigenesis, are required for full pathogenesis [48–52]. In this regard, previously reported TG mice for vGPCR and vCyclin also failed to fully recapitulate KSHV-associated vascular diseases although tumorigenic properties of the viral products were otherwise demonstrated [48,49,53–58]. Expression of multiple viral products has been achieved in B-cells using the latency locus under the control of the native viral promoter, but specific expression in endothelial cells has not been assessed [24]. Our mouse model contrasts with previous TG models of KSHV-encoded genes in the extent of a proinflammatory phenotype.
The severity and systemic nature of the endothelial changes were reminiscent of certain features of the POEMS syndrome [7]. Although the etiopathogenesis of this syndrome is still largely unknown, a role for KSHV has been suggested by few studies. First, there is frequent association with KSHV-associated MCD and angioma formation. Second, in POEMS syndrome there is overproduction of proinflammatory cytokines, including TNFα, IL1β, IL10, IL6, VEGF [59], similarly to what is observed in MCD and KICS, suggesting that these three clinical entities partially overlap. Moreover, POEMS is characterized by the presence of monoclonal Ig, usually IgG or IgA with lambda light chain, and KSHV encodes for viral IL6 that is functionally active on human myeloma cells [60]. KSHV was found in the lymphoid cells of MCD, as well as in the microvenular hemangioma, the pathognomonic endothelial lesion, positive for CD34, CD31, LYVE-7 and Prox-1, that characterizes this syndrome [61–64]. However, other studies failed KSHV detection in this syndrome [65,66]. Similarities of POEMS with ROSA26.vFLIP;Cdh5(PAC).creERT2 TG mice included: i) neuropathic symptoms, which in mice are likely related to hyperplasia of the perineurium around nerve bundles in spinal nerve roots, ganglion and skeletal muscle, ii) systemic presence of elongated endothelial cells, particularly in the heart, which is also increased in size, reminiscent of organomegaly seen in POEMS iii) proneness to develop endocrinopathy (e.g., diabetes), as suggested by increased glycemic levels observed in TG mice, and iv) overproduction of proinflammatory cytokines, including TNFα, IL10, IL6. While the association between POEMS and KSHV remains controversial, a role for KSHV in KICS is well-established and the cytokine storm observed in the vFLIP TG mice is very reminiscent of that seen in this syndrome [4,5].
We also observed remodeling of myeloid differentiation with expansion of CD11b+Gr1+Ly6G+Ly6C+/− cells, phenotypically corresponding to granulocytic myeloid derived suppressor cells (MDSCs). Under physiological conditions, immature myeloid cells from the bone marrow differentiate into granulocytes, macrophages or dendritic cells (Fig. 7). Tumors are capable of secreting several factors in the tumor microenvironment responsible for changes in myeloid differentiation that ultimately can favor tumor immune evasion, angiogenesis and tumor progression. M1 toward M2 polarization is favored by increase in IL10 and reduction in IL12, which lead to reduced Th1 activity and tumor immune evasion, along with angiogenesis and tumor promotion. The main myeloid subpopulations responsible for these effects in tumors are TAM, MDSC, and suppressive DC (Fig. 7). Aberrant CD11b+Gr1+ myeloid cells have also been found in the mouse placenta, where most likely exert immune suppressive and angiogenetic functions to promote immune tolerance and growth of the developing embryo [67]. Mouse MDSCs consist of two major subsets: granulocytic CD11+Ly6G+Ly6Clow cells and monocytic CD11b+Ly6G+/−Ly6Chigh cells (M-MDSCs), which differ in their immunosuppressive mechanisms [43,68]. MDSCs derive from the bone marrow hematopoietic precursors due to the altering of myelopoiesis by chronic inflammatory mediators [69], such as STAT1 and NF-κB, signaling pathways known to be vFLIP targets (56, 59). MDSCs exert their immunosuppressive functions primarily by inhibiting antitumor T-cell function. Moreover, MDSCs are able to secrete angiogenic factors, matrix metalloproteinases and cytokines promoting neoangiogenesis and tumor growth and skewing immune responses towards protumoral Th2-type with activation of Tregs. Thus, MDSCs play a central role in the development of immunosuppressive tumor microenvironment [43], as also emphasized by the fact that functionally active tumor-specific CD8+ T-cells can develop anergy or undergo apoptosis when adoptively transferred into a microenvironment containing MDSCs; moreover, depletion of MDSCs restore CD8+ T cell function, thus confirming their role in induction and maintenance of host immunosuppression [41].
The cooperation between chronic inflammation and myeloid cell expansion is particularly relevant. In our vFLIP TG mice there is evidence of chronic inflammation at different anatomic sites, sustained also by left-shift in myeloid differentiation. Moreover, vFLIP transcriptome, as defined by in vitro gene expression profiling of both vFLIP-expressing endothelial cells and PEL [28,40], highlights the fact that vFLIP activates several proinflammatory cytokines directly implicated in tumor microenvironment and remodeling of myeloid cells, particularly IL4, IL10, IL6, IL13, TGFβ, CCL5/RANTES,IL2, IL1β, G-CSF, similar to those seen in our in vivo data. The myeloid phenotype observed in our vFLIP TG mice, with expansion of phenotypically bona fide granulocytic-MDSCs, is the first demonstration that vFLIP exerts in vivo induction and remodeling of myeloid differentiation with changes in critical components of the microenvironment toward a proinflammatory, angiogenic and immunosuppressive effect. The aberrant myeloid differentiation seems to be a consequence of vFLIP-mediated perturbation of cytokine profiles; once the microenvironment is polarized toward M2, development of MDSCs rather than Th1 activity is favored. In turn, MDSCs, through the upregulation of molecules such as VEGF, Bv8 and MMP9, can favor angiogenesis, tumor progression and tumor immune evasion (Fig. 7). Additional studies are necessary to dissect whether this cytokine storm is produced by myeloid cells or, alternatively, by the endothelial cells with the myeloid cells being a target of this cytokine overproduction. However, the myeloid phenotype with expansion of CD11b+Gr1+cells was observed in both endothelial and B-cell specific vFLIP TG mice, therefore it likely represents myeloid cells chemotactically recruited by the ectopic expression of vFLIP in either cell type, which, in turn, precedes the expression of cytokines known to have tropism for myeloid cells.
In addition to vFLIP’s ability to impair GC formation and Ig maturation, this change in cytokine profile with remodeling of myeloid differentiation might represent a novel mechanism developed by KSHV to achieve immune evasion by altering the microenvironment to prevent immune recognition of KSHV-infected cells. Considering that Th1-type responses promote cellular immunity against intracellular pathogens and tumors, particularly meaningful is the evidence that KSHV as oncovirus has developed mechanisms to induce Th2 polarization and sabotage host immunity through manipulation of the microenvironment. Interestingly, also KSHV miR-K12-7 induces the expression of IL6 and IL10 [70], which by inhibiting DC maturation protect PEL from host immune recognition [71] and simultaneously act as independent growth factors for these cells [72,73]. It is likely that myeloid differentiation is also perturbed in KSHV-infected individuals, with M2 polarization and impairment of Th1 activity. Although there is need for prospective studies on myeloid cells in KSHV-infected patients, quantitative and functional defects of peripheral blood DC and monocytes with reduced IL12 and increased IL10 were reported as becoming even more pronounced in advanced stages of KS [74]. Moreover, KSHV-specific CTLs are very rare in patients who progress to KS, supporting the role of Th1 immune responses in controlling KSHV replication and transformation [75]. Finally, the cytokine profile from patients with KSHV-associated disease further sustains the hypothesis based on our in vivo finding that vFLIP-induced M2 polarization of the microenvironment (with increased IL10, IL13, IL4, INFγ and reduction in IL12) is critical for KSHV pathogenesis.
KSHV is associated with KS in which tumor identity has been made extremely puzzling by the presence of a rich myeloid component, as well as KICS and MCD, both associated with inflammatory cytokines. Our findings suggest this phenomenon is a result of vFLIP-driven remodeling of the microenvironment through a paracrine effect due to the secretion of myeloid-stimulating factors from vFLIP-expressing endothelial or B-cells (Fig. 7). Most macrophages in KS lesions do not contain KSHV, largely favoring a paracrine effect, although rare cells co-express LANA and histiocytic antigens [76].
In conclusion, we have revealed a previously unknown function for vFLIP in inducing in vivo expansion of the myeloid compartment with the emergence of a cellular component of immunosuppressive phenotype. This has important implications for the pathogenesis of KSHV-associated malignancies that invariably display a rich myeloid inflammatory infiltrate, which remains poorly characterized. The profound myeloid phenotype induced by vFLIP supports the key role vFLIP has in contributing to host immune dysfunction with development of tumor immune evasion during KSHV pathogenesis. The high-level coordination between cellular and soluble components seen in these mice provide a model to test inhibitors of vFLIP or other immunotherapeutic approaches targeting the microenvironment as potential anticancer agents for KSHV-associated diseases.
To generate mice expressing the transgene in an endothelial-cell specific manner, homozygous ROSA26.vFLIP TG mice [23] were crossed with heterozygous Cdh5(PAC).creERT2 knock-in mice [39] of C57BL/6 genetic background; therefore, all experimental mice were on 129/Sv-C57BL/6 genetic background and age-matched littermates were used as controls. Genotyping was performed by PCR analysis on mouse tail DNA. All mice were housed, bred and studied according to the guidelines of Institutional Animal Care and Use Committee at Cornell University. Mice were monitored for pathological changes weekly and sacrificed when visibly ill, according to approved protocols. Statistical analysis of event-free survival was performed by GraphPad Prism v.5 (San Diego, CA, USA) using Kaplan-Meier cumulative survival curve and the log-rank test to evaluate statistical significance.
Lung, spleen, liver and heart were isolated during autopsy and promptly processed to obtain a single cell suspension using collagenase A and DNaseI treatment. RNA extraction, RT-PCR and quantitative RT-PCR were performed using standard protocols as detailed in Supporting Methods (S1 Methods). Total protein extracts were prepared from lung, spleen, liver and heart using RIPA buffer, gel electrophoresed on 12% SDS-PAGE gel, transferred to a polyvinylindene difluoride membrane (Millipore) and immunostained according to standard methods using anti-FLAG (M2; Sigma) and anti-β-actin (Sigma) antibodies.
Single-cell suspensions prepared from lung, spleen, liver and heart were stained using standard procedures with a panel of fluorescent-labeled antibodies (see S1 Methods). 7AAD was used for the exclusion of dead cells. Data were acquired on LSRII or Aria flow cytometer (Becton Dickinson) and analyzed using FlowJo software (Tree Star).
Four μm thick formalin-fixed, paraffin-embedded sections were stained for H&E or immunostained with the following antibodies: anti-EGFP (Abcam) and anti-CD34 (MEC14.7; Abcam).
Mice 8–12 weeks of age were subjected to i.p. injection with 0.2 ml of tamoxifen (150 mg) (Sigma), dissolved in a mixture of 90% corn oil (Sigma) and 10% ethanol (Sigma), and analyzed after 30–45 days. Transgene expression was assessed as early as 1 week after induction and remained constitutive over time.
To determine the concentration of a panel of fourteen serum cytokines, a flow cytometry bead-based assay was used, which exploits particle with discrete fluorescence intensities to detect soluble analytes at very low concentrations. GM-CSF, Phospho Stat1, RANTES, IL12/IL23p40 and MCP1 were quantified using BD Cytometric Bead Array (CBA) Mouse/Rat Soluble Protein Master Buffer Kit, while for the detection of IL10, IL6, INFγ, IL1b, IL12p70, TNF, IL4, IL2, IL13 BD CBA Mouse Enhanced Sensitivity Master Buffer Kit was used. Each capture bead has a distinct fluorescence and is coated with a capture antibody specific for a soluble protein. The bead populations are resolved in two fluorescence channels of a flow cytometry, and each bead population is given an alphanumeric position indicating its position relative to other beads. Beads with different position can be combined to create multiplex assay and analyze multiple proteins from a single sample. After incubation of the capture beads with analytes and detection reagent, the PE mean fluorescence intensity (MFI) of the complex was measured and readings within the assay linear range were used to calculate the serum cytokines concentrations against cytokines standard curve for each analyte (Becton Dickinson).
Statistical significance, defined as P<0.05, was assessed by two-tailed unpaired Student’s t-test.
|
10.1371/journal.pcbi.1002133 | Effective Stimuli for Constructing Reliable Neuron Models | The rich dynamical nature of neurons poses major conceptual and technical challenges for unraveling their nonlinear membrane properties. Traditionally, various current waveforms have been injected at the soma to probe neuron dynamics, but the rationale for selecting specific stimuli has never been rigorously justified. The present experimental and theoretical study proposes a novel framework, inspired by learning theory, for objectively selecting the stimuli that best unravel the neuron's dynamics. The efficacy of stimuli is assessed in terms of their ability to constrain the parameter space of biophysically detailed conductance-based models that faithfully replicate the neuron's dynamics as attested by their ability to generalize well to the neuron's response to novel experimental stimuli. We used this framework to evaluate a variety of stimuli in different types of cortical neurons, ages and animals. Despite their simplicity, a set of stimuli consisting of step and ramp current pulses outperforms synaptic-like noisy stimuli in revealing the dynamics of these neurons. The general framework that we propose paves a new way for defining, evaluating and standardizing effective electrical probing of neurons and will thus lay the foundation for a much deeper understanding of the electrical nature of these highly sophisticated and non-linear devices and of the neuronal networks that they compose.
| Neurons perform complicated non-linear transformations on their input before producing their output - a train of action potentials. This input-output transformation is shaped by the specific composition of ion channels, out of the many possible types, that are embedded in the neuron's membrane. Experimentally, characterizing this transformation relies on injecting different stimuli to the neuron while recording its output; but which of the many possible stimuli should one apply? This combined experimental and theoretical study provides a general theoretical framework for answering this question, examining how different stimuli constrain the space of faithful conductance-based models of the studied neuron. We show that combinations of intracellular step and ramp currents enable the construction of models that both replicate the cell's response and generalize very well to novel stimuli e.g., to “noisy” stimuli mimicking synaptic activity. We experimentally verified our theoretical predictions on several cortical neuron types. This work presents a novel method for reliably linking the microscopic membrane ion channels to the macroscopic electrical behavior of neurons. It provides a much-needed rationale for selecting a particular stimulus set for studying the input-output properties of neurons and paves the way for standardization of experimental protocols along with construction of reliable neuron models.
| Ever since the seminal study of Hodgkin and Huxley [1] on the biophysical basis of the squid giant axon action potential, conductance-based models (CBMs) have provided a critical connection between the microscopic level of membrane ion channels and the macroscopic level of signal flow in neuronal circuits. Indeed, as we have sought to further our understanding of single neuron and network computation [2], [3], CBMs have become one of the powerful computational approaches in Neuroscience [4], [5], [6], [7]. They have been of great assistance in incorporating diverse experimental data under a coherent, quantitative framework and for interpreting experimental results in a functionally meaningful way [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19]. Considering the dramatic advancements in our knowledge of single neurons and neural circuits along with the equally impressive increase in computing power during the last decade, CBMs can be expected to become of even greater utility than they already are today [20], [21], [22].
The most fundamental difficulty in accurately modeling neurons stems from the fact that their electrical behavior arises from the complex interaction of a large number of non-linear elements – the membrane ion channels [10], [23]. Furthermore, the identity and density of different ion channels vary from neuron to neuron and cannot presently be directly determined experimentally. Instead, these are treated as free parameters that are typically constrained by an iterative process of comparison between a set of experimental recordings (e.g. voltage response to current-clamp steps) and the model's responses until a close resemblance is found. Yet successful matching of model response to a given target experimental data set is not, in and of itself, sufficient to establish the validity of a model, as complex models with numerous parameters run the risk of systematic biases (or errors) in the estimation of their parameters, i.e. over-fitting. Specifically, such a bias may not be apparent in the accuracy of matching the response to stimuli used to construct the model, but may be revealed by further testing of the model's generalization to different conditions. One can imagine many different such tests: predicting the response to pharmacological manipulation, examining the stability of the model to small perturbations of model parameter values, etc.
Here we describe the application of a particularly intuitive yet powerful measure of generalization: the model's ability to generate an accurate response to a set of current stimuli to which it has not been previously exposed during the model's construction. We favor this form of testing generalization since it gets to the heart of the purpose of conductance-based models (CBMs) – to examine whether the model can indeed be considered a valid approximation of the neuron's underlying dynamics. If that were indeed the case, one would expect the response of the model to match the experimental response not only to the stimuli used to constrain it, but also to different, novel inputs. Moreover, measuring the response to different stimuli is experimentally straightforward. Such measurement of generalization has only been sporadically applied in CBM research papers [24] most likely due to the fact that the vast majority of CBMs studies involved hand tuning of parameters [25], [26] in which clean separation between test sets and generalization sets is difficult due to human involvement. Thus, despite its importance, the quantification of generalization has been lacking from conductance-based neuron modeling.
Since there are many different choices for stimuli that can be used to train and test a model, it is crucial to have a clear way of selecting an optimal (and minimal) set of stimuli (and corresponding targets to be preserved) that will ensure accurate generalization to a wide range of inputs. The present work is the first to have addressed these fundamental principles in order to assess the validity of CBMs in a thorough manner. We experimentally recorded the responses of a variety of cortical neurons to a wide set of different current stimuli (step, ramp and noise currents) each with multiple intensities and many repetitions. We then selected a subset of the experimental data (a training set) to be employed in generating models of the respective cells, using automated multiple objective optimization algorithms, and reserved another set of stimuli to test the accuracy of these models in generalizing to novel stimuli.
We show that, for all neurons tested, CBMs were able to accurately predict stimuli not encountered during the parameter constraining process. Furthermore, by systematically changing the number and type of stimuli used to constrain the models, we determined how each stimulus contributes to the models' predictive power. Notably, models trained solely on responses to step currents were able to accurately predict both the responses to simple stimuli such as ramp current as well as to the responses to physiologically inspired noisy current injections. In contrast, models trained on either ramp or noisy inputs were not as successful in predicting the response to other types of stimuli, i.e. do not generalize well. We discuss the reasons why some stimuli are more successful for estimating the properties of the underlying ion channels than others as well as the implication of this work on the way we understand the process of constraining biophysical neuron models and on the data collection approach required to allow the generation of accurate, predictive CBMs. We believe that our method will become a standard tool for generating in-silico models for a variety of neuron types and that these models could then be used in realistic models of large scale neuronal networks.
The process of constraining (training) the CBMs is portrayed in Figure 1 and explained in detail in the Materials and Methods section. Briefly, from the experimental data, voltage responses to suprathreshold current inputs, (Figure 1a) we first extract a set of features (Figure 1b). We then obtain the reconstructed morphology of the neuron and assume a set of membrane conductances (ion currents) to be present in the neuron's soma (Figure 1c). In the present study we assumed that the modeled cortical cells contain: Transient sodium channel-Nat, Delayed potassium channel-Kd, Slow inactivating persistent potassium channel-Kp, fast non-inactivating potassium channel Kv3.1 channel, high-voltage-activated calcium channel Ca, calcium dependent K channel - SK, Hyperpolarization-activated cation current – Ih, M-type potassium channel Im (for full details see Materials and Methods). In the interests of simplicity, and since recordings were performed in the soma, we assumed the neuron's dendrites to be passive. We then run an optimization algorithm (a Multiple Objective Optimization (MOO) algorithm [27]) to constrain the values of the maximal conductances of these ion channels and of the passive properties of the neuron. The optimization generates a set of multiple models from which we select for further analysis only the models that pass a selection criterion (Figure 1e). These constitute the final set of acceptable models (Figure 1f).
The procedure of assessing the models' generalization power is depicted in Figure 2. In Figure 2a three suprathreshold step currents (together with the corresponding experimental voltage responses) were injected to a rat layer V pyramidal cell and used as the training set. Model parameters, maximal conductance values for the eight excitable ion channels modeled and the neuron's passive properties, were automatically constrained until the response of the resulting set of models closely matched the experimental data (Figure 2b).
Then, while keeping the model parameters fixed, we applied to the models a new set of stimuli (ramp currents in this example) that were not encountered during the parameter constraining procedure, and recorded the models' voltage response to these new stimuli (Figure 2c, red trace, generalization). Finally, we quantified the degree of resemblance of the model response to that of the corresponding experimental response (Figure 2d). This is quantitatively expressed as the model mismatch, or error, as measured by the feature-based distance between model and experiment in units of experimental standard deviation (SD) [27].
Figure 3a depicts the ability of models constrained by responses to step currents to predict (generalize to) the response to suprathreshold ramp current injections. We find that as the size (the number of different step currents) of the training set increases, the average training error between the model and the experimental responses slightly increases (Figure 3a, blue circles). This is expected from learning theory as a model of a given complexity is challenged to fit a growing number of targets [28]. However, the more interesting measure of model accuracy is the error in matching responses to stimuli not encountered during the parameter constraining procedure, the generalization error. This error steeply decreases with the size of the training set (Figure 3a, red circles, difference between one and four stimuli, P<0.0001) indicating more accurate, reliable (better constrained) models.
We next turn to constraining models by ramp current injection (Figure 3b). Surprisingly, when attempting to generalize to step currents using models that were trained on ramp currents, an increase on the size of the training set did not yield better generalization for the response to step currents and the distribution of the generalization errors was very broad (Figure 3b).
In order to determine the impact of the nature and number of stimuli used to constrain models on the conductance values of successful solutions, we portray in Figure 3c the spread of parameter values found at the end of the parameter optimization process, as well as simulations of all points on a grid [29]. The spread of solutions consistent with one step stimulus was considerably larger than that of solutions consistent with four (ratio of areas 0.24; Figure 3c, light and dark blue areas for one and four stimuli, respectively, shown in two dimensional space). Note that though visualization is difficult beyond three dimensions, the calculation of the consistency of points on a grid with each stimulus can be readily performed on the high dimensional grid. This reduction in area with increasing number of training set stimuli can be seen for most individual conductance dimensions when considered separately as well (Figure S1a).
When considering models constrained on ramp currents we again find that for most conductances an increase in the number of stimuli leads to reduction in the spread of successful solutions (Figure S1b). However, the relative size of the area of solutions consistent with four ramp and step current stimuli (Figure 3d blue areas marked with stimulus icon) in relation to the area of intersection (Figure 3d dark blue) is much larger for ramp currents than step currents. Thus, a solution chosen at random from those consistent with step currents is far more likely to be in the area of intersection, i.e. to be consistent with responses to both ramp and step currents. This is directly in line with the more successful generalization from step current responses to responses to ramp currents than vice versa. We note that the different ways in which stimuli “carve out” zones in parameter space is highly relevant to the problem of solution non-uniqueness and return to this subject in the Discussion.
To ensure that the asymmetric generalization is not due to an inherent difficulty with constraining models to match responses to ramp stimuli we quantified the ability of models trained on ramp currents to generalize within stimulus, e.g., to other ramp current stimuli not encountered during the parameter constraining procedure. We find that, in contrast to the between stimulus generalization, addition of stimuli results in decreased generalization error (Figure 4a, difference between one and four ramp stimuli, P<0.0001). We compared this to the within stimulus generalization in step currents and found it to be qualitatively similar (Figure 4b, difference between one and four step stimuli, P<0.0001).
Step or ramp currents are clearly not likely currents for a neuron to encounter in its natural setting. Thus, we consider in Figure 5 the ability of models constrained with these simple stimuli to predict more physiological noise current injections. We employed the gamma coincidence factor (GCF) [30], [31] in order to measure how well-locked is the timing of model APs generated in response to noisy current injection to the APs recorded experimentally in response to the same current (across multiple experimental repetitions of the current injection). Two different noisy currents were used, one with high mean and low standard deviation (noise type 1) and one with low mean and high standard deviation (noise type 2). We find that models trained on two steps currents and two ramp currents were the best predictors of the experimental AP times that were generated in response to both noisy currents. When comparing the number of APs coincident between the voltage responses derived from two repetitions of the current input, the number of model APs coincident with those of any given experimental repetition was over 90% of the number of APs consistent between two different experimental repetitions (Figure 5a black traces experimental voltage, black dots experimental AP times, red trace model voltage response, red dots model AP times; GCF 0.91±0.03). Very similar accuracy was obtained for the second type of noise current (GCF 0.92±0.04, Figure S2a).
Responses to noise currents can themselves be used to constrain the model by attempting to maximize the temporal fidelity of the model to the experimental AP times. Indeed, models trained on responses to noisy currents achieve a perfect within model accuracy of GCF = 1. Generalization within stimulus type (to the other noisy current type) was also highly successful (Figure 5b GCF 0.95±0.09). However, models trained on noisy currents poorly matched responses to step and ramp current inputs (Figure 5c, 5d average feature error 2.58±0.85 and 2.92±0.97 respectively in experimental SD units). The discrepancies in feature values fell beyond 2.5 experimental SD units, more than twice as much as the between stimulus generalization error of step currents. In addition, the spread of parameter values of successful solutions was very broad (not shown). Thus, the generalization from responses to noise currents to that of simple currents was asymmetric, with the combined step and ramp currents generalizing well to noise currents but not vice versa.
We determined the accuracy of generalization from all different training sets to all generalization test sets (Table 1). We find that the combined set of ramp and step stimuli was the most effective in generalizing to responses of both the simple and noise current injections. Among the single stimuli, the step stimulus was the most successful. Additionally, we find that though adding stimulus intensities improves the generalization error, the added benefit of including additional stimulus intensities of the same type in the training set drops after more than three stimulus intensities.
We note that there is no theoretical guarantee that models that generalize well to a certain type of stimulus will also generalize well to different ones. An important class of stimuli are stimuli that continuously sweep through a range of frequencies, sometimes referred to as “chirp” or “zap” stimuli [32]. Though we did not explore the space of such stimuli extensively in our experiments, for the data we have we find that models trained on the combined step and ramp stimuli generalize well to subthreshold frequency sweeps (Figure S3).
Results presented so far pertained to models of a rat layer V pyramidal cell. In order to assess the generality of the results we applied the analyses described above to four additional cells. These cells provided examples of different cell types (pyramidal, interneuron), different ages (juvenile, adult) and different animals (rat, mouse). We were able to generate successful CBMs for each of the cells selected (shown in Figure 6a). We found that, in general, the major results highlighted above are consistent across all cells. Namely, the combined set of step and ramp stimuli was the most effective and achieved very high temporal precision values (Figure 6b). The generalization error was reduced as the number of stimuli increased (Figure 6c) and the generalization between stimuli was asymmetrical, with this set capable of matching responses to noisy currents, but not vice versa.
To the best of our knowledge, this is the first study to rigorously quantify and successfully incorporate the concept of generalization into the construction of experimentally-constrained conductance-based neuron models (CBMs). Several previous studies have fit models to surrogate data [33], [34], [35], [36] or to experimental data [29], [36], [37] but none have systematically compared the generalization of models derived from different experimental stimuli to novel stimuli. Furthermore, it is the first study showing a systematic successful application of automated parameter constraining of CBMs for a wide set of different stimuli types, different neuron-types and different animals. For the five cells studied, we obtained general results regarding the utility of different stimuli types in constraining CBMs. We believe that the paradigm we propose should hold also for other neuron types (e.g., hippocampal CA1 pyramidal cells) but this requires further exploration.
Importantly, by considering the ability of CBMs trained on one stimulus type to predict the responses to a set of different stimuli, we provide a simple and valuable way of measuring the utility of a certain stimulus in generating faithful CBMs. Clearly, evaluating the utility of a given stimulus is of great practical importance not only to those directly involved in biophysical modeling but also to experimentalists as it will provide an objective method of selecting which stimuli to be applied experimentally to a neuron in the limited time of stable recording. Notably, despite its centrality to the modeling effort, this subject has evoked little systematic study, perhaps due to the technical difficulty of generating CBMs that generalize well to experimental data (for surrogate data see ref. [38]). Evaluation of the utility of different stimuli has been performed for simpler biophysical models, such as integrate and fire type models [39]. However, the stimuli found are typically closely tied to the specific phenomenological nature of the model assumed (e.g., a stimulus tailored to accurately measured the AP threshold) and are thus not always applicable to models of a different nature (e.g. models that do not have an explicit parameter for the threshold, such as CBMs).
For the step and ramp currents studied here, we find that multiple suprathreshold intensities of two second long step and ramp currents are required to generate faithful models. For the number of stimulus intensities studied here additional intensities reduce the generalization error (Figures 3,4). Yet, the added benefit of stimuli beyond three intensities diminishes. For the noise currents, we find that ten second long stimuli were sufficient to generate models that generalize well for different noise currents. For each of the stimuli, we use ten repetitions to estimate the intrinsic variability. A combined set of step and ramp stimuli was able to achieve even better generalization (Table 1). Thus, training sets combining different stimuli are expected to be more effective than single stimulus sets in their generalization as we indeed find (see below).
To what do we attribute the success of step stimuli in generalizing to other stimuli? More generally, what could make one stimulus more useful than another in generating models that generalize to a wide variety of stimuli? The intuition behind the success of the step stimulus relies on a combination of the nature of the stimulus itself and single-cell biophysics. The ion-channels expressed by a neuron exhibit a wide range of time constants, from the very brief (less than a millisecond) to the very long (hundreds of milliseconds and more). Different stimuli activate these membrane ion channels to different degrees. If a certain channel is only partially activated by a given stimulus, the contribution of this channel to shaping the model dynamics (and hence the sensitivity of its parameter values) will not be well estimated.
The slow transition through voltage prior to firing an AP elicited by ramp currents strongly inactivates transient currents (e.g., fast inactivating Na+ channels). Thus, if only ramp currents are present in the training set, the parameter constraining procedure has no opportunity to “learn” of the possibility of transient activation, leading to an underestimate of the sensitivity of parameter values of transient channels. When this model is challenged with the need to generalize to depolarizing step stimuli, in which the degree of inactivation prior to the first AP is much smaller, the full sensitivity of transient channels comes into play and some of the models fail to generate accurate responses. In contrast, white noise (or noise smoothed by a short correlation time) is essentially a continuous series of transients. This rapid transition between voltage values is ineffective at activating channels with longer time constants. Hence, the sensitivity of channels with long time constants (e.g. slow inactivating potassium channels such as Kp) is underestimated by models trained solely on noise currents. In other words, noise currents are composed only of transient responses and ramp currents lack a strong transient. Step currents, on the other hand, contain both an initial strong transient followed by a sustained level of depolarization. Thus, they are able to activate both transient channels and channels with long time constants, yielding more accurate estimates of their contribution to the overall response of the cell. Note, that had we been dealing with a linear system, white noise would be sufficient to determine its transfer properties and no other stimuli would be required [40]. However, neurons are of course highly nonlinear systems.
The intuitive description above is in line with the quantitative results regarding the effectiveness of generalization from different stimuli i.e., the failure of models trained on ramps to generalize to step currents, (Figure 3), the failure of models trained on noise currents to generalize to steps and ramps (Table 1) and the spread of acceptable parameter values (Figure 3). Notably, the intuitions developed are relevant not only to the specific model itself (as would be the case with phenomenological models) but also to the general understanding of the function of different ion channels in sculpting neuronal dynamics since the models directly incorporate the experimentally derived dynamics of specific channels. In summary, despite the simple and artificial nature of the step current, it is more successful in constraining the dynamics of the neuron than the synaptic-like noisy stimuli that more closely mimic the conditions a neuron might encounter in-vivo. Thus, we point out that the similarity to natural conditions should not be the only reason for selecting stimuli. Indeed, one must in addition consider how the stimuli might be used to uncover the underlying biophysical dynamics.
Mapping the portion of parameter space [29] corresponding to solutions consistent with a given stimulus provides a both intuitive and quantitative view of the effect of different stimuli on model reliability. Different stimuli carve-out different shaped zones in parameter space (see Figure 3). The degree to which two zones overlap is an indication of how well the models will generalize from one to the other, as only those models found in the intersection area are consistent with both. Thus, if one of the stimuli is chosen to train the model, the portion of the area found outside of the intersection area corresponds to models that will fail to generalize to the other stimulus. By combining different stimuli in the training set we obtain different intersections of these zones. Ultimately, we are interested in finding effective intersections that will reduce the space of solutions as efficiently as possible to the intersection of all stimuli measured. Naturally, as we add more and more stimuli at some point the zones will fail to intersect any longer, indicating that we have tasked our models too far and must either choose a different model or less ambitious requirements. Notably, we believe that this provides a very useful framework for tackling the problem of non-uniqueness in the solution space. Importantly, this will allow more detailed exploration of the spread and composition of different membrane channel conductances for a given neuron type and even comparisons between the same neuron type at a different stage of neuronal maturation, or between different neuron types and different species.
In summary, we have demonstrated that, given the experimental response to different stimulus types (and several repetitions of each) and based on the theoretical framework presented here, we can construct faithful CBMs of different neuron types that can accurately predict the responses to both simple and noisy current injections that were not used during model construction. This suggests that the models generated indeed capture the neuron's dynamics. We emphasize that modeling studies should report not only the similarity of models to the data used in their generation (training error) but should also reserve some of their data for examining the generalization (or predictive) quality of the models. We note however that there is no guarantee that models that generalize well to a certain stimulus will also generalize well to other stimuli and this issue requires more careful exploration with many stimuli. Our development of a framework to quantitatively test the utility of different stimuli and our finding that some stimuli are more advantageous in constraining CBMs than other stimuli has prompted us to start exploring experimentally and theoretically the effectiveness of more sophisticated stimulus protocols in constraining neuronal models. Ultimately, the goal is to find the optimal (and minimal) set of stimuli that ensure accurate generalization to a wide set of diverse stimuli. There are numerous possible options for the different forms of stimuli that could be injected within a fixed time, for instance frequency sweeps that explore frequency response and resonant properties of neurons [32], [41] or more complicated noisy stimuli that alternate between different noise parameters [42], [43]. This is a subject that is presently under active pursuit.
Wistar rats (17–19 days old) and one x98 mouse [44] were quickly decapitated according to the Swiss national and institutional guidelines.
The brain was carefully removed and placed in ice-cold artificial cerebrospinal fluid (ACSF). 300 mm thick parasaggital slices were cut on a HR2 vibratome (Sigmann Elektronik, Heidelberg, Germany). Slices were incubated at 37°C for 45–60 min and then left at room temperature until recording. Cells were visualized by infrared differential interference contrast videomicroscopy utilizing a VX55 camera (Till Photonics, Gräfeling, Germany) mounted on an upright BX51WI microscope (Olympus, Tokyo, Japan). Cells were patched in slices ∼1.8 mm lateral to the midline and above the anterior extremity of the hippocampus ±0.8 mm, corresponding to the primary somatosensory cortex [45], [46], [47]. Thick tufted layer 5 PCs (rat and mouse) were selected according to their large soma size and their apparent large trunk of the apical dendrite. Layer 6 fast-spiking interneurons were selected according to their multipolar soma shape. Care was taken to use only “parallel” slices, i.e. slices that had a cutting plane parallel to the course of the apical dendrites and the primary axonal trunk. The cell type was confirmed by biocytin staining revealed by standard histochemical procedures [48].
Slices were continuously superfused with ACSF containing (in mM) 125 NaCl, 25 NaHCO3, 2.5 KCl, 1.25 NaH2PO4, 2 CaCl2, 1 MgCl2, and 25 D-glucose, bubbled with 95% O2 – 5% CO2. The intracellular pipette solution (ICS) contained (in mM) 110 K-gluconate, 10 KCl, 4 ATP-Mg, 10 phosphocreatine, 0.3 GTP, 10 N-2-hydroxyethylpiperazine-N9-2-ethanesulfonic acid (HEPES), and 13 biocytin, adjusted to a pH 7.3–7.4 with 5 M KOH. Osmolarity was adjusted to 290–300 mosm with D-mannitol (35 mM). The membrane potential values given were not corrected for the liquid junction potential, which was approximately −14 mV. All chemicals were from Sigma-Aldrich (Steinheim, Germany) or Merck (Darmstadt, Germany).
Whole cell recordings (1–3 cells simultaneously) were performed with Axopatch 200B amplifiers (Molecular Devices, Union City, CA) in the current clamp mode at a bath temperature of 34±1°C during recording. Data acquisition was performed with an ITC-1600 board (Instrutech Co, Port Washington, NY), connected to a Macintosh running a custom written routine under IgorPro (Wavemetrics, Portland, OR). Sampling rates were 10 kHz, and the voltage signal was filtered with a 2 kHz Bessel filter. Patch pipettes were pulled with a Flamming/Brown micropipette puller P-97 (Sutter Instruments Co, Novato, CA) and had an initial resistance of 3–4 MW. During recording the series resistance was 10, 10, 11, 17, or 22 MW and bridge balanced. Miniature excitatory postsynaptic potentials (mEPSPs) were blocked with 10 mM CNQX and occasionally with 40 mM AP5.
Three different types of stimuli were applied. Stimuli were scaled with a constant factor k ∈ (1, 2, 2, 2.5, 3) so that the cells fired with moderate mean frequencies of 2–20 Hz, high enough to obtain enough spikes for analysis, yet low enough not to over stimulate the cells and shorten their life span. Six depolarizing step currents of 2 s duration and increasing amplitudes (100–225×k pA) were applied. Five depolarizing ramp currents, 2 s rising phase (from 0 to 125–250×k pA) and symmetrically decaying falling phase, were injected (only rising phase was used in this study). In addition, we apply Ornstein-Uhlenbeck [49] (OU) colored noise processes that are considered to represent the current that might arrive at the soma of a cell as a result of the summation of the activation of many synapses in the cell's dendritic arbor [50]. We employ two different 20 s long OU processes with identical correlation time (2 ms) but different statistics. One is generated with a mean of 50×k pA and SD of 100×k pA (hereby referred to as noise type 1). The other mirrors this process by having a mean of 100×k pA and SD of 50×k pA (hereby referred to as noise type 2). We repeatedly inject the different currents in order to measure response variability. Each stimulus was repeated 10–20 times.
All simulations were performed in the NEURON simulation environment [51]. The morphology of 5 cortical neurons from rat and mouse somatosensory cortex was derived from reconstruction of in-vitro stained cells. The number of compartments employed differed from cell to cell, all cells contained more than a hundred compartments. Specific axial resistance was 150 Ωcm and capacitance was 1 µF. The following ion channels were assumed to be present in the membrane of the modeled soma: Transient sodium channel-Na, Delayed potassium channel-Kd, Slow inactivating persistent potassium channel-Kp, fast non-inactivating potassium channel Kv3.1 channel, high-voltage-activated calcium channel Ca, calcium dependent K channel - SK, Hyperpolarization-activated cation current – Ih, M-type potassium channel Im, for full details see ref. [27]. The dynamics of these channels were described using Hodgkin and Huxley formalism [1]. As all the experimental recordings in this work were performed from the cell's somata and for the sake of simplicity, the modeled dendrites were assumed to be passive. The maximal conductance of all eight channels along with the leak reversal potential and leak conductance in the soma and dendrite served as free parameters, yielding a total of eleven free parameters in the model. The allowed range for the conductances can be found in ref. [27].
An overview of the procedure by which we generate conductance-based models (CBMs) from an experimental data set is presented in Figure S1. We begin by recording the responses (Figure S1a) of the cell to intracellular current injection. Responses are then analyzed by the extraction of a set of features (Figure S1b), which are used to generate feature-based distance functions (see below). Next, we use the reconstructed morphology (Figure S1c) to generate the compartmental model of that cell and assume a set of 8 ion channels to be present in the soma membrane of the model cell. Together the reconstructed morphology and the assumed ion channels compose the model skeleton. When combined with a set of specific values for the free parameters they together constitute a single CBM for that neuron.
A stochastic optimization procedure is employed to constrain the parameters of the model in accordance with the experimental data. We employed a multiple objective optimization (MOO) algorithm which operates by genetic algorithm optimization [52]. The algorithm evaluates 300 sets of parameter values in parallel and iteratively seeks to reduce the error, which measures the discrepancy between model and experiment (Figure S1d). As the algorithm is stochastic in nature, we repeat the optimization procedure ten times in order to reduce the chance that the optimization procedure fails to converge. Thus, at the end of the optimization procedure, 3000 parameter sets, i.e. 3000 tentative models of the cell are obtained along with their corresponding error values. We then choose only those models that pass the acceptance criterion of a model-experiment mismatch no greater than two SD in each feature (Figure S1e). Ultimately, we end up with a set of models that closely match the experimental voltage responses (Figure S1f).
The discrepancy between the target experimental data (a train of spikes in response to a set of current stimuli) and model simulation of the response was measured using feature-based distance functions [27]. Features to be fitted were extracted from the firing response of the neuron (e.g. number of action potentials (APs), spike height). The value of each feature was derived from the experimental responses. The model response to the same stimulus was then analyzed in an identical fashion. The model-to-experiment distance value, for this feature, was measured by the distance of the model feature value from the experimental mean, in units of experimental SD. These distance functions have two main advantages. First, they address experimental variability by considering the distance of a model in relation to the experimental SD. Second, they are expressed in well defined, not arbitrary, units.
For step pulses, we employ a set of six features: the number of action potentials (APs) during the pulse, the time to the first AP from stimulus onset, the accommodation index (a measure of the accommodation in the rate of APs during the stimulus [27]), the width of an AP at half height, the average height of an AP, and the average depth of the after hyperpolarization (AHP) as defined by the minimal voltage point. For ramp currents, as the height of APs decreases during the stimulus response, we considered not only the average height of APs and the depth of AHPs but also the slope of a linear fit to the change as an additional feature. For the noise stimuli we do not use feature-based distance functions, but rather the gamma coincidence factor [30], [31] - an index measuring the coincidence of AP times in relation to the neuron's intrinsic reliability. The index is normalized from 0 to 1, a value of 0 indicates that a model does no better than a Poisson train and a value of 1 indicates that the model and experimental repetitions have as many coincident spikes on average as do two experimental repetitions. Note, that in this context the objective of optimization is to maximize this value.
In order to assess the utility of different stimuli in generating neuron models that generalize well both within stimulus and across stimuli we generate models with training sets that are equally matched in terms of the length of the experimental data. Namely, we consider four different training sets: step current pulses only (four intensities of two second long step currents), ramp current pulses only (four intensities of two second long ramp currents), combined step and ramp currents (two intensities of two second long step currents and two intensities of two second long ramp currents) and noise currents (eight seconds of OU noise process current injection). For each of these training sets, we test the generalization of the model to four different conditions: step currents, ramp currents and two different noise currents. Five intensities of step and ramp currents can be potentially employed to both train and test generalization. Stimuli used during the parameter constraining process (e.g. the four step currents used by the first training set) are excluded from the generalization test.
As we typically employ several feature-based distance functions per stimulus and we often use more than one stimulus for the optimization, we obtain multiple distance function values for each model-experiment comparison. To obtain a single value a weight vector is used to sum all the different distance functions. Here we employ a different approach termed multiple objective optimization (MOO) [53]. This approach maintains the multiple distance measures and does not employ a weight vector. Instead, the relation between distance measures is that of domination: solution i is said to dominate solution j if for all distance functions the values of solution i are no greater than those of solution j and for at least one distance function the value of solution i is strictly lower than that of solution j. The purpose of a multiple objective optimization procedure is to find the best possible tradeoffs between the distance functions, termed the Pareto front.
We employ a genetic algorithm (GA) based optimization algorithm designed for multiple objective optimization named NSGA-II [52]. This algorithm is an elitist (GA) with a parameter-less diversity preserving mechanism. We custom implemented this algorithm in NEURON. We find that the algorithm almost always converges after 1000 iterations of evaluation of the full set of parameter values. As a safety factor, 1500 iterations were used. We repeated each given optimization ten times.
The spread of successful solutions in parameter space for a given stimulus type can be explored by simply marking the location of each point corresponding to a solution. However, it is difficult to determine in this fashion whether a certain region of parameter space is consistent with more than one stimulus as the points themselves will almost surely not coincide. An additional disadvantage is that many of the solutions are the result of the same optimization run and thus contain artificial correlations due to the closely linked nature of solutions generated by a single optimization run. To overcome these two difficulties we complement our analysis by additional simulation of the response of a large set of points placed on a high-dimensional grid [29] to all (step and ramp) stimuli used in the experiments. This approach is extremely computationally expensive. However, it overcomes the above-mentioned difficulties: as the same points are simulated for all conditions, one can easily ascertain which are the conditions consistent with each point. Secondly, as all points are generated on the grid there are no unknown artificial correlations between them. Lastly, this approach allows visualization of projections of the space of solutions consistent with each stimulus (see Figure 2).
|
10.1371/journal.ppat.1006957 | Down-regulation of microRNA-203-3p initiates type 2 pathology during schistosome infection via elevation of interleukin-33 | The type 2 immune response is the central mechanism of disease progression in schistosomiasis, but the signals that induce it after infection remain elusive. Aberrant microRNA (miRNA) expression is a hallmark of human diseases including schistosomiasis, and targeting the deregulated miRNA can mitigate disease outcomes. Here, we demonstrate that efficient and sustained elevation of miR-203-3p in liver tissues, using the highly hepatotropic recombinant adeno-associated virus serotype 8 (rAAV8), protects mice against lethal schistosome infection by alleviating hepatic fibrosis. We show that miR-203-3p targets interleukin-33 (IL-33), an inducer of type 2 immunity, in hepatic stellate cells to regulate the expansion and IL-13 production of hepatic group 2 innate lymphoid cells during infection. Our study highlights the potential of rAAV8-mediated miR-203-3p elevation as a therapeutic intervention for fibrotic diseases.
| Schistosomiasis is a serious but neglected tropical infectious disease. Hepatic fibrosis caused by lodged eggs from the parasite is the primary cause of morbidity and mortality from this disease. Type 2 immune response, featured by the T helper 2 cell associated cytokines such as IL-4 and IL-13, is the central regulator of disease progression in schistosomiasis, but the signals that induce it after infection remain elusive. Aberrant expression of miRNAs underlies a spectrum of human diseases, including infectious diseases. In this study, using a well-studied murine model of human schistosomiasis, we show that Schistosoma infection down-regulates the miR-203-3p expression, and that IL-33, an inducer of type 2 immunity, is a direct target of miR-203-3p in hepatic stellate cells. The reduced miR-203-3p leads to elevated levels of IL-33, promoting the expansion and IL-13 production of hepatic group 2 innate lymphoid cells and thus initiating type 2 pathology. Importantly, rAAV8-mediated elevation of miR-203-3p in liver tissues protects mice against lethal schistosome infection by alleviating hepatic fibrosis. Thus, our study highlights the crucial role of miR-203-3p in the initiation of type 2 pathology during schistosome infection, and suggests miR-203-3p as a potential target for fibrotic diseases.
| Schistosomiasis is a serious but neglected tropical infectious disease, affecting more than 230 million people worldwide [1]. Hepatic granuloma and secondary fibrosis caused by lodged eggs from the parasite are the primary cause of morbidity and mortality from this chronic and debilitating disease. Elucidating the mechanisms that initiate hepatic schistosomiasis has been a major research objective for decades, and it is now well-established that hepatic schistosomiasis is an immune pathological disease [2,3]. A major breakthrough was the identification of type 2 immune response, characterized by the T helper 2 cell associated cytokines such as interleukin 4 (IL-4) and IL-13, as a central regulator of disease progression in schistosomiasis [2,3]. However, the signals that induce type 2 immune response after infection remain elusive.
Quiescent hepatic stellate cells (HSCs) are located in the subendothelial space, between the anti-luminal side of sinusoidal endothelial cells and the basolateral surface of hepatocytes, and are characterized by their cytoplasmic vitamin droplets [4]. When liver injury occurs, quiescent HSCs are activated to become proliferative, contractile, and fibrogenic myofibroblasts [5]. Activated HSCs produce excessive extracellular matrix (ECM) that is deposited in the liver, and are the main effector cells in various types of hepatic fibrosis, including fibrosis induced by schistosome infection [6]. In addition, more recent studies have uncovered the fundamental role of HSCs in hepatic inflammation and immunity [7,8].
MicroRNAs (miRNAs) are endogenous, small noncoding RNAs which control the activity of more than 30% of protein-coding genes through target mRNA degradation or translational inhibition [9,10]. Increasing evidence has demonstrated that miRNAs are involved in regulating almost every cellular process, and aberrant miRNA expression is a hallmark of many human disorders, including infectious diseases [11,12]. Several studies by ours and other groups have shown that miRNAs play a crucial role in the pathogenesis of schistosomiasis and may serve as useful therapeutic targets [13–15]. In particular, one of our previous studies has shown that depletion of a single miRNA, miR-21, in the liver protects hosts from lethal infection through attenuation of hepatic fibrosis [15].
In this study, we used a murine model of Schistosoma japonicum (S. japonicum) to investigate the role of miR-203-3p, a miRNA down-regulated following infection in the progression of hepatic schistosomiasis [15]. We found that recombinant adeno-associated virus 8 (rAAV8) mediated elevation of miR-203-3p in the liver protected mice from lethal infection through alleviating type 2 pathology. Importantly, our data indicate that miR-203-3p targets IL-33, an inducer of type 2 immunity [16,17], in HSCs to regulate the expression of IL-13 in hepatic group 2 innate lymphoid cells (ILC2s) during infection.
We previously identified more than thirty deregulated host miRNAs by expression profiling during the progression of hepatic schistosomiasis. This includes miR-203-3p as the most down-regulated [15]. To examine the role of miR-203-3p in schistosomiasis in vivo, mice were first challenged with a lethal dose of S. japonicum cercaria and then intravenously injected with either rAAV8-pri-miR-203-3p vector sustainedly expressing the miRNA, control vector, or PBS at day 10 post-infection. We found that a single dose of rAAV8-pri-miR-203-3p protected infected mice from the lethal effect of schistosomiasis. Six of ten mice receiving rAAV8-pri-miR-203-3p survived to the end of the study (i.e. 80 days; Fig 1A). In contrast, the majority of mice receiving control vector (n = 10) or PBS (n = 10) died within 9 weeks post-infection (Fig 1A).
Hepatic granuloma and fibrosis, induced by host type 2 immune response resulting from liver-trapped parasite eggs, are the primary cause of morbidity and mortality from this disease [2,3]. Thus, we next investigated if the rAAV8-pri-miR-203-3p-mediated intervention was indeed through effective elevation of miR-203-3p activity that in turn attenuated the type 2 pathology. To this end, mice were exposed to a mild dose of parasites and then treated with vectors or PBS. Our data revealed that the level of miR-203-3p in the rAAV8-pri-miR-203-3p treated group was significantly higher than in the control groups (Fig 1B). Excessive ECM deposition is the main feature of hepatic fibrosis. By 6 weeks post-infection, mice receiving rAAV8-pri-miR-203-3p displayed a significant reduction in ECM deposition as shown by hydroxyproline quantification (Fig 1C) and Masson’s trichrome staining (Fig 1D and 1F), whereas the size of hepatic granulomas in all groups was similar as shown by H&E staining (Fig 1E and 1F). Reduction of fibrosis was further confirmed by qPCR-based quantification of fibrosis associated gene expression in the livers of infected mice. Amounts of Col1α1, Col3α1, and α-Sma mRNA were dramatically reduced in livers of mice treated with rAAV8-pri-miR-203-3p (Fig 1G, 1H and 1I). In addition, we detected the alteration of the cytokines that are associated with type 2 immune response in liver tissues after elevation of miR-203-3p. Consistent with the antenuated fibrotic phenotype, a strong reduction in mRNA levels of Il13 and Tgf-β1 was detected in livers of mice treated with rAAV8-pri-miR-203-3p (Fig 1J and 1K). However, levels of other cytokines, including interferon-γ (Ifn-γ), tumor necrosis factor-α (Tnf-α), Il4, and Il5, were not significantly altered (S1A Fig). Consistent with our previous study, the virus vector did not affect the survival and egg production of parasites in the hosts (S1B Fig), and virus delivery in the liver showed no significant differences between groups (S1C and S1D Fig).
HSCs are the predominant cellular source of ECM during hepatic fibrosis. Thus, we investigated whether the anti-fibrotic effect of rAAV8-pri-miR-203-3p intervention directly modulated the activity of HSCs. To this end, we isolated primary HSCs from infected mice after administration with vectors to quantify mRNA levels of miR-203-3p, Col1α1, Col3α1, and α-Sma. Our data showed that significantly decreased miR-203-3p expression and increased collagen and α-Sma expression were observed in the infected mice without rAAV8-pri-miR-203-3p treatment compared with uninfected mice (S2 Fig). As expected, miR-203-3p expression was clearly elevated, while collagen and α-Sma expressions were distinctly reduced in HSCs after rAAV8-pri-miR-203-3p intervention (S2 Fig), suggesting that miR-203-3p could modulate the activation of HSCs in vivo.
To investigate whether elevation of miR-203-3p in the liver can reverse the parasite egg-induced hepatic fibrosis, mice were infected with a mild dose of parasites. At 42 days after infection, when hepatic fibrosis was clearly manifest, mice were treated with praziquantel to kill the parasite, then injected with either rAAV8 vectors or PBS, and necropsied at 70 days post-infection (Fig 2A). Again, the expression of miR-203-3p was significantly increased in the rAAV8-pri-miR-203-3p treated mice (Fig 2B). Importantly, hydroxyproline quantification and Masson’s trichrome staining revealed that fibrosis in rAAV8-pri-miR-203-3p treated mice was markedly reduced compared with controls (Fig 2C, 2D and 2F), but the size of hepatic granulomas in all groups was similar as shown by H&E staining (Fig 2E and 2F). This was confirmed by reduced expression of Col1α1, Col3α1 and α-Sma in these mice (Fig 2G, 2H and 2I). Of cytokines tested, only Il13 mRNA was reduced (Fig 2J and 2K), and livers showed no significant change in egg burden (Fig 2L).
Considering that elevation of miR-203-3p attenuates type 2 pathology, we speculated that miR-203-3p could regulate the initiation of type 2 immunity after infection. IL-33, an IL-1-related cytokine, is an inducer of type 2 immunity in several organs [18], and is a potential target of miR-203-3p, predicted by TargetScan database. To analyze the relationship between miR-203-3p and IL-33, we evaluated their expression during the progression of hepatic schistosomiasis. We found that expression of miR-203-3p began to decrease in the liver at day 32 post-infection, reaching its lowest levels at day 42 and 52 (Fig 3A); in contrast, the level of Il33 mRNA was maintained during the early stage of infection, then significantly elevated by day 42 post-infection (Fig 3B). In addition, we investigated the expression of miR-203-3p and Il33 mRNA in different hepatic cell compartments, including hepatocytes, HSCs, and Kupffer cells (KCs). Our data showed that, similar to the expression pattern in whole liver, the expression of miR-203-3p in hepatocytes and HSCs began to decrease at day 42 post-infection, while the level of Il33 mRNA was elevated at the same time (Fig 3C and 3D). However, the expression of both miR-203-3p and Il33 mRNA in KCs was unchanged during the observed time (Fig 3C and 3D). We also characterized the relative abundance of miR-203-3p and Il33 mRNA in different hepatic cell compartments, and found that, in both the uninfected and infected livers, miR-203-3p was selectively expressed in hepatocytes and HSCs (Fig 3E), whereas Il33 was primarily expressed in HSCs (Fig 3F and 3G). To further validate that activated HSCs could be a source of IL-33 in infected livers, we carried out immunochemistry staining for IL-33 and α-SMA, and we observed that both factors were mainly located in the periphery of egg granulomas (S3 Fig). Double staining using immunofluorescence displayed a co-localization of IL-33 and α-SMA staining (Fig 3H), indicating that activated HSCs express IL-33 in vivo. In addition, we detected the expression of miR-203-3p and Il33 during the progression of HSC activation in vitro. Resting HSCs will be automatically activated when cultured on a plastic surface [4]. We found that, when primary HSCs from uninfected mice were cultured on plastic plates, expression of miR-203-3p in these cells was significantly reduced, while the level of Il33 mRNA was significantly elevated, accompanied by a dramatic increase in collagen expression (S4 Fig). Taken together, these results suggest that the activated HSCs in infected livers are a source of IL-33, and that IL-33 is a potential target of miR-203-3p in HSCs.
To test whether IL-33 is a direct target of miR-203-3p, we first generated a reporter construct that contains the firefly luciferase gene fused to the 3’ UTR from Il33 mRNA containing a putative miR-203-3p target site (Fig 4A). This construct was transiently transfected into 293T cells together with miR-203-3p mimics or a negative control miRNA. We observed a marked reduction in luciferase activity in cells transfected with miR-203-3p mimics together with Il33-UTR (Fig 4B). In contrast, mutation of 5 nt in the miR-203-3p seed sequence led to a complete abrogation of reporter inhibition (Fig 4B). We transfected miR-203-3p mimics or inhibitors into primary HSCs from uninfected mice, and quantified the level of IL-33 by qPCR or western blot. Our data revealed that, at both the mRNA and protein levels, elevation of miR-203-3p distinctly reduced the expression of IL-33, while depletion of miR-203-3p significantly increased the expression of IL-33 (Fig 4C and 4D). Finally, we analyzed IL-33 levels in primary HSCs from infected livers treated with rAAV8-pri-miR-203-3p, and found that IL-33 expression was markedly reduced (Fig 4E and 4F). Taken together, these data indicate that IL-33 is a direct target of miR-203-3p in HSCs. In addition, we noticed that the target site of miR-203-3p in the 3’UTR of Il33 gene is not conserved from mouse to human. However, we provided evidence that human IL-33 is also a direct target of miR-203-3p in the HSCs (S5 Fig).
Though a number of cell types were suggested as sources of IL-13 in response to IL-33 stimulation, a recent study has proved that ILC2s, instead of lymphocytes, basophils, or mast cells, are the predominant source of IL-13 in the liver after IL-33 stimulation [19]. Having found that IL-33 is a target of miR-203-3p and that elevation of miR-203-3p leads to a reduction of IL-13 in the liver, we hypothesized that down-regulation of miR-203-3p during the progression of hepatic schistosomiasis could lead to higher expression of IL-13 in ILC2s via increased levels of IL-33 in HSCs. To address this, we first analyzed the number of hepatic ILC2s and production of IL-13 by hepatic ILC2s during infection using flow cytometry (S6 Fig). Our data indicated that ILC2s were a primary source of IL-13 production in the infected livers (S7 Fig), and both the number of ILC2s and production of IL-13 in these cells were increased by day 32 post-infection, peaking at day 42 post-infection (S8A Fig). This initial elevation occurred prior to the elevation of IL-33 in HSCs (Fig 3D and S8B Fig). STAT6 phosphorylation, a marker of IL-13 pathway activation, also began to increase in HSCs, peaking at the same time point, and the production of collagen in HSCs was dramatically elevated at day 42 post-infection (S8B Fig). Moreover, we observed that, following elevation of miR-203-3p in HSCs using rAAV8-pri-miR-203-3p vectors, the number of ILC2s and production of IL-13 in these cells were markedly reduced (Fig 5A and 5B). The expression of IL-33 (Fig 4E and 4F), the phosphorylation of STAT6 (Fig 5C), and production of collagen (S2 Fig) in HSCs were all significantly decreased. Finally, we found that purified primary HSCs, stimulated with recombinant IL-13 ex vivo, responded by phosphorylation of STAT6 and production of collagen in a time-dependent manner (Fig 5D and 5E). Taken together, these data indicate that miR-203-3p regulates the expression of IL-13 in ILC2s by targeting IL-33 in HSCs, thus modulating the expression of ECM by HSCs, during the progression of hepatic schistosomiasis.
In this study, we demonstrate that efficient and sustained elevation of miR-203-3p in liver tissues, using the highly hepatotropic rAAV8, protects mice against lethal schistosome infection by alleviating hepatic fibrosis. Importantly, we show that miR-203-3p targets IL-33, an inducer of type 2 immunity, in HSCs to regulate the expansion and IL-13 production of hepatic ILC2s during infection.
Type 2 immune response, featured by elevation of IL-4 and IL-13 levels, plays a crucial role in host protection as well as pathological tissue fibrosis after helminth infection, including schistosome infection, but the signals that induce type 2 immunity are poorly understood. Recently, numerous studies have highlighted that tissue damage, which induces the release of cytokine alarmins such as IL-33, is a potent mechanism driving type 2 immunity, particularly in the context of helminth infection [20,21]. The role of IL-33 in schistosomiasis has also been intensively studied, but published reports have been inconsistent. Mchedlidze et al. reported that, in the animal model of Schistosoma mansoni (S. mansoni) infection, IL-33 was critical for inducing the development of IL-13-dependent hepatic fibrosis [19]. This phenomenon was also observed in the animal model of S. japonicum infection [22]. However, a recent study showed that IL-33 needed to synergize with two other cytokine alarmins, thymic stromal lymphopoietin (TSLP) and IL-25, in the regulation of IL-4/IL-13-dependent inflammation or fibrosis after S. mansoni infection [23]. The inconsistency might be due to the difference of intervention time: when IL-33 is depleted in the embryo stage, the function of IL-33 might be compensated by other factors; but when IL-33 is inhibited during the progression of diseases, the role of IL-33 in disease progression become obvious. In this study, our data also indicated that IL-33 is crucial for inducing the progression of type 2 pathology after infection, as significant reductions in hepatic fibrosis and IL-13-producing ILC2s were observed upon down-regulation of IL-33 in the liver, which resulted from rAAV8-mediated elevation of miR-203-3p. These studies, including our current study, have established that IL-33 is a crucial mediator in the maintenance of type 2 pathology induced by schistosome infection.
Importantly, our study revealed other important aspects of the role of IL-33 in the regulation of type 2 pathology after infection. Although IL-33 is involved in the progression of many human diseases, little is known about the regulation of its expression. Our study, for the first time, revealed that IL-33 is regulated by miRNAs. IL-33 is a direct target of miR-203-3p in HSCs, and downregulation of miR-203-3p leads to elevated levels of IL-33 in HSCs, initiating type 2 pathology after infection. Previous studies have demonstrated that miR-155 can regulate the expansion and IL-13 production of ILC2 in the context of IL-33 [24], and miR-29a can regulate IL-33 effector function via targeting its decoy receptor sST2 [25]. Thus, miRNA could be an important regulator in the initiation of type 2 pathology. Activated HSCs are a major source of IL-33 in infected livers. Our findings confirmed that IL-33-producing cells are located in the periphery of egg granulomas where activated HSCs produce excess collagen, and that expression of IL-33 in primary HSCs is significantly elevated after infection. These findings are consistent with previous studies, which demonstrated that activated HSCs and pancreatic stellate cells (PSCs) are major sources of IL-33 in the fibrotic liver and pancreas, respectively, of both mice and human [26–28]. These studies suggest that HSCs or PSCs might be the sentinel cells in the organs, which detect the injury signals and promote wound healing response by releasing damage-associated molecular pattern such as IL-33. Our data showed that the initial elevation of IL-13 in hepatic ILC2 cells (day 32 post-infection) occurs prior to the elevation of IL-33 in whole livers and HSCs (day 42 post-infection). It is reported that cell death by necrosis or active necroptosis, instead of active secretion, might be the dominant mechanism by which IL-33 reaches the extracellular milieu [29]. Therefore, we speculate that the source of IL-33 that activates hepatic ILC2s at day 32 after infection derives from the necrosis or active necroptosis of HSCs.
Here, we created a schematic diagram showing the molecular mechanism underpinning the regulation of type 2 pathology after infection by miR-203-3p (Fig 6): The toxic challenge derived from parasite eggs trapped in the liver tissue induces the down-regulation of miR-203-3p in HSCs, which relieves the inhibition to IL-33. Sequential elevation of IL-33 is released into the liver tissue and stimulates the proliferation and IL-13 production of hepatic ILC2s. IL-13 then activates HSCs to produce excessive ECMs through activation of STAT6 pathway. Thus, our study highlights the crucial role of miR-203-3p and its target IL-33 in the initiation of type 2 pathology during schistosome infection. It is noteworthy that IL-33 expresison in HSCs begins to elevate at day 42 after infection, thus, this mechanism mainly exerts its role in the progression of Th2 pathology after 42 days. In addition, IL-33 is reported to promote the development of fibrosis in many organs, including liver [19], lung [30], kidney [31], heart [32], skin [33], and other organs [34]. Therefore, miR-203-3p might serve as a useful target in the treatment of these fibrotic diseases.
All animal experiments were performed in strict accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, and were approved by the Animal Ethics Committee of Second Military Medical University (laboratory animal usage number FYXK (Shanghai) 2014–0003). To minimize pain and discomfort, all animal surgeries were undertaken under sodium pentobarbital anaesthesia.
Male BALB/c mice (6 week) were obtained from the experimental animal center of Second Military Medicine University, and were housed under specific pathogen-free conditions and fed with autoclaved food and water as needed. To establish the animal model of schistosomiasis, mice were exposed percutaneously to 16 or 30 S. japonicum cercariae.
For parasite perfusion, the portal vein was dissected at the root, then the thoracic cavity of the mouse was opened and the circulatory system was perfused via the aorta with sterile PBS. Parasites were collected and counted in a sterile petri dish containing medium. Subsequently, the liver was removed and snap-frozen in liquid nitrogen. For egg counting, part of the liver was digested overnight with 4% potassium hydroxide, then the total number of schistosome eggs was counted, and the liver egg burdens were defined as 104 eggs per gram of liver tissue.
The size of hepatic granuloma was measured from Mayer’s H&E staining of sections using a calibrated measuring eyepiece, and the extent of fibrosis was analysed by Masson’s trichrome staining of sections as described previously [15]. All granulomas within each section were scored for blue density on a scale of 1–4, and a second measurement of area involved was also determined using the same scale. The total fibrosis score was determined by multiplying the density and area for each granuloma (a score of 16 would be the maximum). The hydroxyproline content in the liver was detected using a colourmetric assay kit according to the manufacturer’s instructions (Nanjing Jiancheng Bioengineering Institute, Nanjing, China).
The procedure was performed as described previously [15]. In short, HSCs were first isolated by density-gradient centrifugation and then further purified using negative selection with magnetic CD11b antibody beads (MACS, Miltenyi, Auburn, CA); Kupffer cells were first isolated by density-gradient centrifugation and then further purified using positive selection with magnetic CD11b antibody beads.
Primary HSCs were cultured on plastic dishes in DMEM supplemented with 10% fetal bovine serum (Hyclone), 4 mmol/L L-glutamine, penicillin (100 IU/ml), and streptomycin (100 mg/ml). Cells were maintained at 37°C, 5% CO2 in a humidified atmosphere. For transfection, HSCs were transfected with 40 nM miR-203-3p mimics (Qiagen), miR-203-3p inhibitors (Qiagen), or negative controls at day 3 after seeding using Lipofectamine 3000 (Invitrogen).
Total RNA was isolated using Trizol reagent (Invitrogen) according to the manufacturer’s protocol. Real-time PCR was performed as described previously [35]. The levels of miR-203-3p, Col1α1, Col3α1, α-Sma, Ifn-γ, Tnf-α, Tgf-β1, Il4, Il5, Il10, Il13, and Il33 were detected using the SYBR Green Master Mix kit (Roche). U6 snRNA or Gapdh was used as an internal control, and the fold change was calculated by the 2-ΔΔCt method. Sequences of primers used in this study are shown in S1 Table.
Total cell protein was extracted on ice using RIPA lysis buffer in the presence of freshly added protease and phosphatase inhibitors (Thermo), then quantified by the BCA method (Pierce). A total of 30 μg protein extract per lane was loaded onto a 14% SDS-polyacrylamide gel and transferred to nitrocellulose membranes (Pierce). Nonspecific binding was blocked with 5% nonfat milk in PBS. The membrane was incubated with rat anti-IL-33 (R&D) or rabbit anti-phospho-STAT6 (Cell signaling) antibody overnight at 4°C. IRDye 800CW goat anti-rabbit IgG or goat anti-rat IgG (LI-COR) was used as secondary antibody, and rabbit anti-GAPDH antibody (Abcam) was used as an internal standard.
Immunohistochemistry was performed on formaldehyde-fixed, paraffin-embedded mouse livers. After hydration, liver sections were incubated with rat anti-IL-33 (R&D) or rabbit anti-α-SMA (Abcam) antibody for 1 hour at 37°C, and HRP or fluorescence conjugated secondary antibody (Abcam) was used to display the signals.
Nonparenchymal cells isolated from liver were stimulated with PMA (50 ng/mL), Ionomycin (1 μg/mL), and BFA (3 μg/mL) for 4 hours. Cells were surface stained with FITC conjugated lineage cocktail (CD3 / Gr-1 / CD11b / CD45R / Ter-119 / Siglec-f / CD11c / NK1.1) (Biolegend), PerCP/Cy5.5 conjugated ICOS (Biolegend) and APC conjugated ST2 (Biolegend), permeabilized with 0.1% spaonin buffer for 15 minutes, and further stained with PE conjugated IL-13 (eBioscience) before acquiring with FACS Calibur. ILC2s are defined as Lineage(-) ST2(+) ICOS(+) cells in this study. Data were analyzed in Flowjo.
The human or mouse wild-type or mutant 3’ UTRs of IL-33 containing the predicted miR-203-3p binding sites were synthesized (South Gene Technology, China) and cloned into the pGL3.0-control vectors according to the manufacturer’s instructions (Promega). 293T cells were seeded in 24-well plates, then transfected with 40 nM miR-203-3p or a negative control (Qiagen) and co-transfected with 0.8 μg per well wild-type IL-33 3’ UTR-luc or mutant IL-33 3’UTR-luc, using Lipofectamine 3000 (Invitrogen) according to the manufacturer’s instructions. pRL-TK vectors (0.1 μg per well) were co-transfected as endogenous controls for luciferase activity. After 24 h, cells were lysed, and luciferase activities were measured using a dual-luciferase assay kit (Promega).
To express miR-203-3p, pri-miR-203-3p fragment was amplified by PCR from C57/B6 mouse genomic DNA using primer pri-miR-203-3pF (5´AACAGGTCCTCGCACAGAGTGCAGCCCGGC 3´) and pri-miR-203-3pR (5´AACAGGTCCTCCACCCCCGCGCCCCTCTCA3´), then cloned into the PpuMI restriction site in the intron of pscAAVCBPI GLuc plasmid [36]. The identity of pri-miR-203-3p was verified by sequencing. rAAV8 vectors used in this study were generated, purified, and tittered as described [37].
All analyses were carried out with the SPSS 19.0 software. Data were shown as mean ± s.d. The significance of difference between two groups was identified using a Student’s t-test. Multiple comparisons were performed by one-way ANOVA, and followed by Bonferroni post test for comparison between two groups. Survival between different groups was compared by Kaplan–Meier survival curves with log-rank test. P values less than 0.05 were considered significant.
|
10.1371/journal.pntd.0003883 | Echinococcosis: An Economic Evaluation of a Veterinary Public Health Intervention in Rural Canada | Echinococcosis is a rare but endemic condition in people in Canada, caused by a zoonotic cestode for which the source of human infection is ingestion of parasite eggs shed by canids. The objectives of this study were to identify risk factors associated with infection and to measure the cost-utility of introducing an echinococcosis prevention program in a rural area. We analyzed human case reports submitted to the Canadian Institutes for Health Information between 2002 and 2011. Over this 10 year period, there were 48 cases associated with E. granulosus/E. canadensis, 16 with E. multilocularis, and 251 cases of echinococcosis for which species was not identified (total 315 cases). Nationally, annual incidence of echinococcosis was 0.14 cases per 100 000 people, which is likely an underestimate due to under-diagnosis and under-reporting. Risk factors for echinococcosis included female gender, age (>65 years), and residing in one of the northern territories (Nunavut, Yukon, or Northwest Territories). The average cost of treating a case of cystic echinococcosis in Canada was $8,842 CAD. Cost-utility analysis revealed that dosing dogs with praziquantel (a cestocide) at six week intervals to control cystic echinococcosis is not currently cost-effective at a threshold of $20,000-100,000 per Quality Adjusted Life Year (QALY) gained, even in a health region with the highest incidence rate in Canada ($666,978 -755,051 per QALY gained). However, threshold analysis demonstrated that the program may become cost-saving at an echinococcosis incidence of 13-85 cases per 100,000 people and therefore, even one additional CE case in a community of 9000 people could result in the monetary benefits of the program outweighing costs.
| In Canada, Echinococcus spp. tapeworms cycle primarily among wildlife hosts. People are infected with this parasite when they accidentally consume microscopic eggs spread by canids (e.g. dogs, wolves, coyotes, and foxes), and develop larval cysts, often in the liver or lungs. Echinococcosis can be a life-threatening medical condition with long-term health consequences and can be an economic burden for infected individuals and for the public health system. We analysed national health records to measure echinococcosis incidence and risk factors in Canada, and then used this information to determine if a program that facilitated dog deworming to prevent human infection might be economically feasible. Our model suggested that treating infected individuals is currently less expensive than preventing infection, even in the highest risk regions of Canada. However, deworming dogs might be feasible in small rural communities where at least one case was identified. Furthermore, the prevention program has many add-on benefits that contribute to overall community health, but are not measured by our model.
| Echinococcosis, also known as hydatid disease, is a potentially fatal condition caused by zoonotic cestodes of the genus Echinococcus [1]. Two species are endemic to Canada: E. multilocularis, which causes alveolar echinococcosis (AE), and E. canadensis (formerly known as the G8 and G10 genotypes, or the cervid/sylvatic strains of E. granulosus), which causes cystic echinococcosis (CE) [1–3]. Human cases of echinococcosis are considered rare, resulting in approximately 0.72 hospitalizations per million people per year [4]. Domestically-acquired cases are thought be almost exclusively CE (caused by E. canadensis), and appear to occur more commonly in northern latitudes (>55°), in women, and in groups of Indigenous descent [2,4–6]. Foreign-acquired cases of echinococccosis could be caused by other species not present in Canada (e.g. E. granulosus sensu strictu), and may be associated with more severe disease. Echinococcosis is under-diagnosed in people due to an often prolonged disease progression, asymptomatic or nonspecific symptoms, and the difficulty of definitive diagnosis—especially in northern areas where medical imaging services are limited [1,7,8]. It is also under-reported, as there is no formal requirement to report human cases to national public health authorities in Canada. Recent studies highlight the need to better determine the incidence and health care burden associated with human echinococcosis in Canada, especially in rural, remote, and Indigenous communities [4,9,10].
The life cycle of E. canadensis is indirect, and utilizes wild cervids such as moose (Alces alces), elk (Cervus canadensis), and caribou (Rangifer tarandus) as intermediate hosts. Canids such as wolves (Canis lupus), coyotes (C. latrans), and dogs (C. familiaris) serve as definitive hosts [11–13]. Neither intermediate hosts nor definitive hosts are thought to suffer serious adverse effects as a result of infection; however, intermediate hosts may be at higher risk of predation due to decreased pulmonary function [14,15]. In Canada, E. multilocularis utilizes canid definitive hosts (e.g. coyotes, wolves, foxes [Vulpes spp.], and dogs) and rodents (arvicoline and neotomine) as normal intermediate hosts [2]. In contrast to E. canadensis, intermediate hosts of E. multilocularis experience more severe adverse effects [16]. People become infected by accidentally ingesting eggs shed by definitive hosts. Dogs have been identified as high risk reservoirs for human exposure to both species of Echinococcus, especially in areas where dogs can access offal or scavenge rodents as a food source, and where poverty is prevalent [1,17,18].
Worldwide, echinococcosis affects 2–3 million people per year, at an estimated cost of $750 million USD [19]. In countries where livestock strains of E. granulosus are highly prevalent, this disease represents a significant economic burden to healthcare systems, as well as to animal production systems [1,20,21]. For all forms of echinococcosis, there is the possibility for recurrence and of long-term sequelae following treatment, further increasing the burden of disease [1,22]. Multiple countries have implemented programs with various strategies to control and/or eliminate CE [1]. The most effective strategy is generally believed to be treatment of dogs with praziquantel (PZQ), a cestocide effective against Echinoccocus spp., at six week intervals in concert with surveillance of people, dogs and livestock (1). Dog population reduction can also factor into these programs. Few studies have calculated the cost effectiveness of CE prevention programs and none have been done in Canada where the status quo is simply to treat infected people [23–25]. The goals of this paper are to 1) report the incidence of echinococcosis based on existing national datasets, and 2) determine the cost-utility of using a CE prevention strategy (6-week PZQ dog dosing) in comparison to status quo, for a high risk health region in Canada using both public pay and societal perspectives.
We obtained case records for Canadians diagnosed with echinococcosis from the Discharge Abstract Database (DAD) and National Ambulatory Care Reporting System (NACRS) for 2002–2011 through the Canadian Institute for Health Information (CIHI). Nationally, DAD captures all acute hospital inpatient cases, including deaths, discharges and hospital transfers; while NACRS collects ambulatory cases through voluntary submissions from day surgery, outpatient clinics and emergency department visits. Cases were coded using version 10 of the International Classification of Diseases (ICD-10) coding system of the World Health Organization. DAD did not report data from one province (Quebec -QC) unless a resident was treated out of province, but did report data from all other provinces and territories (BC—British Columbia, AB—Alberta, SK—Saskatchewan, MB—Manitoba, ON—Ontario, NL—Newfoundland and Labrador, NS—Nova Scotia, NB—New Brunswick, YT—Yukon Territories, NT—Northwest Territories, NU—Nunavut). Other omissions included MB and NB data for 2002/2003 and NB data for 2003/2004 due to delays in transitioning from ICD-9 to ICD-10 coding. NACRS abstract submissions were biased towards ON, as this was the only province with mandated reporting during the study period. Due to the small population in the 3 northern territories (YT, NT, NU), these cases were grouped together to avoid identifying patients or communities. Our dataset did differentiate between international patients and citizens, but did not report travel history or whether a person had recently immigrated to Canada. Cost and length of stay estimates were only available for 2009/2010 and 2010/2011.
Anonymized patient records from the NACRS and DAD databases were analyzed using SPSS statistical software (version 20; Chicago, Illinois, USA). Individual health card identification numbers (issued by provinces and territories to individuals enabling free access to insured health care services) were assigned a Meaningless But Unique Number (MBUN), which we used to ensure that each individual was counted only once over the study period. Treatment costs and length of stay estimates of individuals hospitalized multiple times were combined for that individual. Rural/urban and neighbourhood quintile income classifications were based on an individual’s postal code, and geographic location was reported by health region (according to a patient’s health card). Rural/urban residence categories adhered to Statistics Canada definitions: (1) Rural (outside or fringe of Census Metropolitan Areas (CMAs) or Census Agglomerations (CAs); (2) Urban Core (large urban area with ≥50 000 people for CMA or ≥10 000 people for CA); (3) Urban Fringe (small urban areas inside CMA or CA but separated from the urban core); (4) Urban areas outside CMAs/CAs (small towns with a population of 1000–10000 people and population density of ≥400 persons/km2) [26]. Other variables included age (categorized as <14, 15–64 and >65 years), gender, province where treatment occurred, and discharge status (i.e. a patient’s health status or anticipated location after leaving the hospital). Population proportions of infection were compared for risk factors using the Z-test, with statistical significance reported at the P<0.05 level. Only individuals over 14 years of age were included in the rural/urban comparison because the comparison dataset provided by Statistics Canada is limited to this age group. Incidence was reported as the median rate over 10 years [27].
For this evaluation, we focused on CE caused by E. canadensis, which is thought to be the primary species endemically transmitted in Canada [2]. We conducted a cost-utility analysis comparing a strategy for CE prevention, PZQ dog dosing, with the status quo (no prevention). Cost-utility analysis presents the cost per Quality Adjusted Life Year (QALY) and therefore captures the costs associated with CE, along with its impact on quantity and quality of life. We modelled one cohort, representing the health region with the highest incidence rate (Kelsey Trail, SK) and included residents of all ages, over the lifetime of the patient, to fully represent the long-term consequences of the disease and the possibility for recurrence. Consistent with the Canadian Agency for Drugs and Technologies in Health (CADTH) guidelines, we used a public payer perspective to represent all the public sector costs associated with CE and the PZQ dog dosing program, and thereby characterised the interests of both the prevention program and the health care system funders. Furthermore, we presented the societal perspective to represent the indirect costs, such as loss of productivity and travel expenses. We considered a cost per QALY between $20,000–100,000 as cost-effective, <$20,000 as very cost-effective and ≤$0 per QALY gained as cost-saving as per [28].
The Kelsey Trail Health Region is home to 42 218 people (2013–2014 estimate), of which approximately half reside in population centres with veterinary services [29]. These centres also have animal control by-laws that prohibit dogs from running freely, require all owners to register their dogs annually, and impose fines on animal owners who do not remove animal waste from public areas. The PZQ dog dosing program considered in this paper included re-homing unwanted dogs from rural/remote communities, as per [30], as well as the following:
CE Surveillance: People and dogs monitored to identify high priority communities
PZQ dosing at 6 week intervals:
Population centres with dog bylaws and veterinary clinics—dog owners given PZQ tablets free of charge at annual registration
Rural/remote communities with no animal bylaws or veterinary clinics—program veterinarian injects dogs with PZQ 2–8 times annually depending on logistical constraints (e.g. accessibility, road conditions) and canine echinococcosis prevalence
Education: Echinococcosis teaching materials provided to primary and secondary school teachers
We used decision analysis to construct a Markov cohort simulation model within Treeage to determine incremental cost-utility. The model ran for 79 years or the average life expectancy of a resident of the Kelsey Trail Health Region [31], with Markov cycles occurring at one year iterations. For both the PZQ program and status quo, the model considered the transition between five CE health states (Healthy, Sick, Sequelae, Fully Recovered, Dead), each with associated costs and utilities. Transition probabilities determined the likelihood of moving between states.
Three types of costs were considered: CE treatment costs, CE indirect costs, and PZQ prevention program costs (S1 Appendix). We estimated the average treatment costs per CE case using estimates provided by CIHI and adding physician costs. Physician costs were estimated based on the Saskatchewan Ministry of Health Payment Schedule for Insured Services Provided by a Physician and expert physician opinion. PZQ program costs included a veterinarian salary [32], vehicle use for travel in and between communities, and the wholesale costs of PZQ. Finally, indirect costs included loss of production due to treatment (one month lost earnings), loss of production due to mortality (average income lost from death until life expectancy), and travel costs (car travel and hotel) [33]. All costs were provided in 2011 Canadian dollars.
We used utilities, which quantify the health wellbeing of an individual, to value the outcomes observed in each health state, with dead having a utility of zero and the healthy average Canadian having a utility of 0.93. Utility scores are weights representing preferences for different health states. The more preferred health states receive higher weights. Utilities are measured on a scale of 0–1, where 0 indicates death and 1 indicates perfect health [34]. Utility scores/weights for different health states could be obtained using Quality of Life instruments [35] and the Standard Gamble approach [36]. The sickness utility represents both the time when an individual is sick and when they were undergoing treatment. Our CE sickness utility (0.72) was based on estimates for hepatic resection and liver cancer, as these illnesses have similar treatments and outcomes [37,38]. Post treatment, those who fully recovered were assumed to return to the healthy state utility of 0.93, whereas those with sequelae had a slightly lower utility of 0.89. The sequelae utility was based on a SF-36 quality of life study of echinococcosis patients who had already undergone treatment but still experience effects of the disease [20,39].
We calculated the risk of developing CE in Kelsey Trail Health Region using the incidence of hospitalization from the CIHI databases (AE cases were excluded; Table 1). The course of disease and the likelihood of different outcomes, including the risk of recurrent echinococcosis, risk of sequelae, fatality rates and all-cause mortality rates were derived from the literature [1,22,40–42]. Costs and utilities were both discounted at a rate of 5% as recommended in the CADTH guidelines, with sensitivity analysis at 0% and 3% discount levels [43].
We calculated the baseline relative risk (RR) of acquiring CE using a PZQ dosing strategy versus status quo from CE incidence estimates in Chile before and after the implementation of a similar PZQ program [22]. In the regions where the Chilean PZQ program was applied, the CE incidence in people decreased from ~60 per 100,000 to 11.8 per 100,000 within 10 years (a RR = 0.19) [22]. Therefore, to represent changes in CE risk following PZQ dosing while taking into account the time-lag between PZQ treatment and impact on human health, we used a table function that decreased the RR every year, from 0.919 the first year, until it reached the base rate of RR = 0.19 after 10 years. For the rest of the model the RR was held constant at 0.19 to represent the possibility that the program missed some infected dogs. We chose the Chilean prevention program and the associated RR as the base estimate for our study because it targeted a similar pathogen (E. granulosus sensu strictu) to that observed in Canada, as well as being one of few programs conducted on a continent rather than an island.
To ensure the validity of the model and the robustness of the findings, we conducted one-way sensitivity analyses of key variables, including risk of echinococcosis, dog-to-human ratio, fatality rate and discount rate. Plausible ranges were derived from 95% confidence intervals, inter-quartile ranges or the literature. Finally, a threshold analysis was conducted to determine the level at which PZQ dosing would be cost-saving, and to identify the minimum incidence rate that would result in a cost-effectiveness of <20,000 per QALY (Quality Adjusted Life Year).
This project was reviewed and approved by the University of Saskatchewan Biomedical Ethics Review Board (REB protocol number 13–51), which adheres to national standards set out by the Tri-Council for research involving humans. We report data at the level of the public health region (or pooled for the sparsely-populated northern territories) to avoid inadvertently identifying individual patients or communities.
Between 2002 and 2011, 384 discharge abstracts were submitted to the DAD and NACRS databases for patients under-going treatment for echinococcosis. Of these, 69 abstracts were removed from descriptive analyses either because they were duplicates (the same individual obtaining medical care on multiple occasions), or because they lacked sufficient information to be assigned an MBUN. The median annual incidence rate was 0.14 cases per 100 000 people (range: 0.12–0.25 cases per 100 000). The median age of echinococcosis cases was 46 years. The highest frequency of cases was observed in females, those aged 15–64, those residing in an urban core, and those residing in neighbourhoods with the lowest income quintile ranking (Table 2). Relative to the 2006 Census estimates of female:male ratios [46], the proportion of female cases was significantly higher than the proportion of male cases at the national level and in three provinces (BC- 13:6, P = 0.012; AB- 25:6, P = 0.032, ON- 18:11, P = 0.001). The proportion of cases in the top age category (<65 years) was significantly higher than the other two categories at the national level, and in AB, ON, and MB (P = 0.001, P<0.001, and P = 0.01, respectively). In BC the proportion of cases in the top age category was significantly higher than the proportion of cases in the youngest age category (P = 0.02), but not the middle age group.
We observed no significant difference in urban versus rural incidence among patients older than 14 years. The highest proportion of cases were observed in the territories (NU, NT, YK), while the lowest were observed in Atlantic Canada (NL, NS, PE, and NB). The proportion of cases in these provinces and territories were all significantly different from the proportion of cases in ON (the most populous province in Canada). Our data suggests that the majority of echinococcosis patients were treated within their province of residence (311/323, 96%), except for those residing in the territories who were all treated out of territory (NL, MB, or AB). Annual incidence rates were highest in the Kelsey Trail Health region (1.7 cases/100 000) in SK and the Norman Regional Health Authority (1.2 cases/100 000) in MB (Fig 1).
Abstracts in this dataset reported cyst location within the body in 60% of cases (188/315), and the species of Echinococcus in 20% of cases (64/315; Table 3). For CE, the most commonly reported cyst location was lung, followed by liver, multiple sites, and bone; whereas liver and multiples sites were the most common descriptors for AE. Of the 305 cases that described discharge disposition, 2.3% ended in fatality.
We based our model on the Kelsey Trail Health Region (SK), which had the highest CE incidence rate in Canada (1.7 cases/100 000). The average cost to treat a single CE case was $8,841.68. The prevention program has a significant yearly cost, approximately $654,033, due to high numbers of dogs in rural areas and high drug costs (S1 Appendix). Furthermore the analysis showed a very small utility gain from using a prevention program compared to status quo (Incr. QALY = +0.00031870). All these factors resulted in a very high incremental cost-utility ratio (ICUR) for the base public pay case ($755,051 per QALY gained). Moreover, the societal perspective did not drastically change the outcome, with a cost per QALY gained of $666,978 (Table 4).
One-way sensitivity analysis demonstrated that none of the cost-utility ratios for any of the plausible variable ranges were under $100,000 per QALY. The best cost-utility ratio came at the high range of the plausible risk of CE (0.0000316 or incidence of 3.16 per 100,000) using the societal perspective with a cost per QALY of approximately $311,143. Varying other data inputs did not significantly change the outcomes, most likely because the starting incidence was so low, thereby making other probabilities irrelevant. Sensitivity of the analysis to the risk of developing CE prompted us to conduct a threshold analysis to determine at what incidence the prevention program might be considered cost-effective. This analysis found at an incidence between 10–37 cases per 100,000 the cost per QALY would be approximately $20,000, while an incidence of 13–85 per 100,000 (risk = 0.000014) would result in the program becoming cost-saving (Table 5).
We report an echinococcosis incidence rate of 0.14 cases per 100 000 annually, which is slightly higher than a previous Canadian estimate, likely because we included cases where echinococcosis was not the primary diagnosis. This is lower than CE incidence rates in other endemic countries including Spain, Portugal, Italy, Greece, and China; but is higher than New Zealand or the island of Tasmania which are provisionally free following the success of control programs [1,22]. We believe that our incidence rate under-estimates the true incidence because CE is not nationally notifiable to public health authorities in Canada, cases were removed from analysis due to incomplete identifier data, the dataset does not include all emergency room discharges or private clinics, and because up to 60% of CE cases (especially those caused by the sylvatic form in Canada) are thought to be asymptomatic [1]. Furthermore, over 150 cases of unspecified liver disease were reported annually during the study time period, suggesting that under-diagnosis of echinococcosis may occur [47]. Based on the best data currently available, we were not able to determine what proportion of cases were foreign-acquired; however, the universal nature of health care in Canada means that costs of treatment of foreign-acquired cases are still incurred.
Although the majority of case reports did not differentiate between CE and AE, highlighting another weakness in reporting, our findings suggest that most CE cases were likely to be domestically-acquired. The geographic distribution of CE cases (Fig 1) is very similar to the known range of E. canadensis in cervids and wolves in Canada (prevalent in the north, absent in the Atlantic provinces) [2, 12, 13], further supporting that these cases are likely endemically acquired. The highest incidence rates were in northern areas of SK, MB and the territories(YK, NU and NT) as opposed to health regions where large metropolises are present (Fig 1). All of the top primary, secondary and tertiary immigration destinations in English-speaking Canada (Toronto, Vancouver, Calgary, Edmonton, Winnipeg, Hamilton, Ottawa, Saskatoon, Victoria, Regina, and Halifax) had very low incidence rates [48]. This emphasizes the need for veterinary public health efforts and improved awareness of Echinococcus transmission in northwestern Canada. Sixteen individuals were diagnosed with AE over the ten year study period, which could be explained by E. multilocularis emergence or incorrect use of ICD codes by physicians. In Canada, AE cases are generally thought to be foreign-acquired, as no autochthonous cases have been reported in Canada since 1928 [49]. However, six of these individuals resided in northern health regions of BC, AB, ON, and in the northern territories (YK, NU, NT), where immigration rates are presumably low. Echinococcus multilocularis is has been observed in wildlife in BC, AB, SK, MB, NT, and NU, and European strains of this tapeworm were recently detected in a domestic dog (as AE) and in wild canids (as adult cestodes) [13,47,50,51]. European strains may have greater zoonotic potential than strains of the parasite long established in the southern parts of the western Canadian provinces (AB, SK, and MB), and this may be supported by the recent emergence of AE in dogs in Canada [49, 50, 52], which is more typically seen in regions of Europe highly endemic for E. multilocularis. Heightened surveillance for AE is warranted, as it generally results in worse health outcomes and significantly higher treatment costs than for CE [53].
Our findings that echinococcosis diagnosis occurred more commonly in females and older adults are comparable with other Canadian studies [4,6]. Cases of CE were most likely to have pulmonary or hepatic involvement, which is a common finding for the cervid strains in people. These findings of gender, age and tissue predilection site are risk factors shared by wildlife cervid hosts for E. canadensis [6,12,54]. The highest frequency of CE cases occurred in low income neighborhoods but we were unable to determine if the proportion of cases relative to other income quintiles was significantly different. Low income individuals might be at higher risk of CE if they fed raw offal to pets and were unable to afford regular cestocidal dosing for dogs.
We report a CE treatment cost that is similar to that in the UK ($10 215 USD), but far higher than that in other countries such as Jordan ($524 USD) [41]. At an ICUR of $755,051 per QALY gained, the dog dosing prevention program was not cost-effective relative to other funded health care programs and current willingness-to-pay guidelines [28]. The current incidence and drug costs, including indirect (societal) costs, yields an ICUR of $666,978 per QALY gained, which is also not cost-effective. To date, few CE/AE prevention programs have been evaluated from an economic perspective [53]. This is a major gap in the literature as these programs may be very cost-effective in higher incidence countries. In fact, although PZQ dosing is not currently cost-effective at the health region level, it may be cost-effective at a smaller community level. The average population per community in Kelsey Trail Health Region is 660 people; therefore, even one CE case per year in a community could warrant a prevention strategy in that community, especially if the source of CE is linked back to the dog population, rather than directly from wildlife. According to the threshold analysis, even one CE case in a community of 9000 people a year could be potentially cost saving for society. Furthermore, as CE incidence increases, the cost-effectiveness of a prevention program becomes more and more dependent on the indirect costs, especially productivity loss. A similar PZQ dog dosing program, delivered concurrently with sheep and goat vaccination, was thought to be cost-effective or even cost saving in Shiqu County (China), where the incidence of human CE (caused by E. granulosus livestock strains) and AE is extremely high [55]. Therefore, it is important that other CE/AE endemic countries engage in evaluations to determine the cost-effectiveness of echinococcosis prevention programs using domestic estimates for incidence, cost, and targeting the strains endemic in their region.
Important considerations exist that might further impact the cost-utility and feasibility of the PZQ dosing program. First, veterinarians use PZQ to treat dogs against a wide range of cestode species in addition to E. canadensis, including Diphyllobothrium spp., Taenia spp., Dipylidium caninum, and Mesocestoides spp., some of which can infect people and/or livestock. Other treatments to prevent zoonotic diseases in dogs, such as nematocides or rabies vaccination, could easily be added to the PZQ program infrastructure at a lower cost than administering all treatments separately. Second, the World Health Organization suggests that the control of echinococcosis go through multiple phases: 1) planning; 2) attack (costly and intensive control measures implemented); 3) consolidation (only high risk animals and people targeted); and 4) maintenance [1]. While the complete eradication of echinococcosis in Canada is not feasible due to wildlife reservoir hosts, there remains the possibility of future cost reductions of the prevention program. For example, after echinococcosis rates in people and animals decreased, this low risk status could be maintained through cheaper methods (e.g. education, owner-administered PZQ, screening high risk dogs). Third, our threshold analysis demonstrated that an increased incidence of echinococcosis could markedly impact the cost-utility of a prevention program. Fourth, this program would target under-served and vulnerable populations that have poorer health outcomes, and therefore, the benefit of preventing disease in these risk groups may help to reduce health inequalities. Lastly, our results indicated that all YK, NU and NT patients travelled out of territory for treatment, which can be very expensive for the health care system, as this generally involves travel by air. In NU, more than 25% of the operations budget is spent sending patients to southern referral centres to obtain care that is unavailable in the north, which drastically increases the costs of treating CE and demonstrates the benefits of a prevention program [56].
Our study provides baseline human echinococcosis data that is otherwise unavailable for Canada, since there is no national reporting or targeted surveillance. Improvements to echinococcosis surveillance could include development of serological tests that are optimal for Canadian strains (e.g. E. canadensis G8 and G10), improved classification of echinococcosis by physicians (i.e. to species and/or genotype level, which can help determine pathogenicity and if cases are endemically or foreign acquired), increased awareness of echinococcosis among physicians in areas where this parasite is prevalent, and addition of this parasite to the list of nationally notifiable pathogens. Improving surveillance would allow policy-makers and governments to make informed decisions about implementing control programs, with the knowledge that increasing incidence greatly improves the cost-utility of an echinococcosis prevention program. Although PZQ dosing did not appear to be cost effective under current conditions at the level of the health region in Canada, it might still be warranted in high risk communities, especially as there are added benefits to people, pets and wildlife in controlling Echinococcus and other zoonotic cestodes.
|
10.1371/journal.pgen.1004785 | Evolution of DNA Methylation Patterns in the Brassicaceae is Driven by Differences in Genome Organization | DNA methylation is an ancient molecular modification found in most eukaryotes. In plants, DNA methylation is not only critical for transcriptionally silencing transposons, but can also affect phenotype by altering expression of protein coding genes. The extent of its contribution to phenotypic diversity over evolutionary time is, however, unclear, because of limited stability of epialleles that are not linked to DNA mutations. To dissect the relative contribution of DNA methylation to transposon surveillance and host gene regulation, we leveraged information from three species in the Brassicaceae that vary in genome architecture, Capsella rubella, Arabidopsis lyrata, and Arabidopsis thaliana. We found that the lineage-specific expansion and contraction of transposon and repeat sequences is the main driver of interspecific differences in DNA methylation. The most heavily methylated portions of the genome are thus not conserved at the sequence level. Outside of repeat-associated methylation, there is a surprising degree of conservation in methylation at single nucleotides located in gene bodies. Finally, dynamic DNA methylation is affected more by tissue type than by environmental differences in all species, but these responses are not conserved. The majority of DNA methylation variation between species resides in hypervariable genomic regions, and thus, in the context of macroevolution, is of limited phenotypic consequence.
| DNA methylation is an epigenetic mark that has received a great deal of attention in plants because it can be stably transmitted across generations. However, the rate of DNA methylation change, or epimutation, is greater than that of DNA mutation. In addition, different from DNA sequence, DNA methylation can vary within an individual in response to developmental or environmental cues. Whether altered characters can be passed on to the next generation via directed modifications in DNA methylation is a question of great interest. We have compared how DNA methylation changes between species, tissues, and environments using three closely related crucifers as examples. We found that DNA methylation is different between roots and shoots and changes with temperatures, but that such changes are not conserved across species. Moreover, most of the methylated sites are not conserved between species. This suggests that DNA methylation may respond to immediate fluctuations in the environment, but this response is not retained over long evolutionary periods. Thus, in contrast to transcriptional responses, conserved epigenetic responses at the level of DNA methylation are not widespread. Instead, the patterns of DNA methylation are largely determined by the evolution of genome structure, and responsive loci are likely short-lived accidents of this process.
| Cytosine methylation is a heritable epigenetic modification found in the genomes of organisms spanning the eukaryotic phylogeny [1], [2], [3], [4]. It occurs in three nucleotide contexts, CG, CHG, or CHH (where H is any nucleotide except G) [5], and is enriched in the repeat rich heterochromatic regions of genomes, in nucleosome linkers, and at CG sites in the exon sequences of genes (gene body methylation) [4], [6], [7], . Repeat-localized DNA methylation plays a role in transposon silencing [12], [13], but the direct relationship between transcription of protein coding genes and DNA methylation remains unclear. In contrast to repeat methylation, gene body methylation is associated with moderately transcribed sequences [6], [7], [14], [15], [16], and has been proposed to stabilize gene expression levels by excluding H2A.Z [17]. Nevertheless, DNA methylation can vary between tissues and environments [18], [19], [20], and in a handful of cases changes in methylation state contribute to heritable phenotypic variation, although the majority have been linked to structural differences near the affected genes [21], [22], [23], [24], [25], [26], [27]. These observations suggest that DNA methylation may regulate developmental processes and that it could potentiate phenotypic variation during evolution.
Unlike mutational processes acting on DNA sequences, our understanding of the factors contributing to meiotically stable variation in DNA methylation is in its infancy [28]. The different molecular mechanisms governing DNA methylation constitute one factor impacting stability and subsequent inheritance at symmetric and asymmetric sites. In the plant Arabidopsis thaliana, initiation and maintenance of methylation at CG and CHG sites is divided primarily between DNA METHYLTRANSFERASE 1 (MET1) and CHROMOMETHYLASE3 (CMT3) [29], [30], [31]. During DNA replication these two enzymes copy symmetrically methylated cytosines onto the newly synthesized DNA strand using the parental strand as a template [32], [33]. Unlike symmetric cytosine methylation, CHH methylation cannot be replicated from the template strand [34]. Instead, methylation at newly synthesized CHH sites is established after cell division by the RdDM RNA-directed DNA methylation pathway through the concerted action of small RNAs (sRNAs) produced from the methylated locus and the de novo DNA methyltransferases DRM1/DRM2 (DOMAINS REARRANGED METHYLTRANSFERASE1/2) [34], [35], [36], [37]. In addition, RdDM-independent asymmetric DNA methylation relies on DDM1 (DECREASE IN DNA METHYLATION1) and CMT2 [38].
The extent to which DNA methylation varies at individual sites across generations, or the epimutation rate, has only recently been characterized in isogenic plant lines [39], [40]. Repeat-associated methylation was remarkably stable over 30 generations, but some variability arose outside of repeats in euchromatic sequence [39], [40]. Changes in DNA methylation accumulated non-linearly, indicating that a subset of methylated sites is particularly prone to spontaneous changes in methylation and, as a result, the absolute DNA methylation differences quickly reach saturation [39], [40]. Variation of methylation across generations has been linked to the transgenerational cycling of transposon and repeats between methylated and unmethylated states in the germline [41].
Armed with the knowledge of within-species epimutation rate, the degree of epigenome stability over short evolutionary periods, within a single species, for example, can be addressed [18]. Using A. thaliana, intraspecific variation in methylation was surveyed in 140 geographically diverse accessions [18]. Most single site and RdDM-derived regional epimutations were rare, occurring in only a few of the 140 accessions [18]. The lack of intermediate frequency epimutations in these categories is consistent with the view that the vast majority of new methylation variants within a species may only exist for brief periods during evolution. Not too surprisingly, a significant subset of both rare and intermediate frequency RdDM-derived regional epimutations were associated with previously unknown structural variants [18]. Expansion and contraction of repeat-associated sequences leads to intraspecific structural variation; therefore, as a result of RdDM silencing, such structural variants should be linked to methylation variation.
Over longer evolutionary periods, broad similarities in DNA methylation are observed across a variety of genomic features. Large-scale patterns of methylation are shared across flowering plants, including extensive methylation of heterochromatic transposon and repeat-associated sequences [6], [7], [8], [9], [10], [11] likely due to conservation of the RdDM machinery in plants. Over shorter divergence times, similar levels of gene body methylation have been observed at orthologous genes within the grasses [11], [42]. Similarly, in vertebrates, where most of the CG sites in the genome are methylated, absence of methylation at so-called CpG islands is usually found in all species examined [43]. Regardless of organism, the degree of DNA methylation conservation depends on both the evolutionary time scale under consideration and on the genomic feature of interest.
Here we compare at single base resolution DNA methylation in three closely related Brassicaceae - Capsella rubella, Arabidopsis lyrata, and Arabidopsis thaliana. These three species, which diverged about 10 to 20 million years ago [44], vary in genome size and architecture [45], [46], [47]. Both C. rubella and A. lyrata have a Brassicaceae typical set of eight chromosomes, while A. thaliana has only five chromosomes [48], [49]. Both the A. lyrata and C. rubella genomes are about 50% larger than that of A. thaliana, but for very different reasons. Expansion of centromeric, heterochromatic regions has enlarged the C. rubella genome, but predominantly euchromatic regions have expanded in A. lyrata, driven by insertions of transposable elements (TEs) adjacent to genic sequences [46], [47]. Reflecting these differences in genome architecture, the reference genome assemblies represent about 85% of the entire genome in A. lyrata, about 75% in A. thaliana, and about 60% in C. rubella (Table S1) [46], [47], [50], [51], [52], [53], [54]. We show that the difference in genome structure is a major factor influencing the evolution of DNA methylation in these species. Furthermore, while overall DNA methylation is similar between species at many sites, dynamic DNA methylation responses between environments and tissues are rarely conserved. Using a comparative framework we were able to disentangle the contribution of genomic, environmental, and developmental factors to DNA methylation variation between species.
Using a factorial design, we subjected seedlings of the inbred reference strains, A. thaliana Col-0, A. lyrata MN47, and C. rubella MTE, to either a control or 23-hour cold treatment and separately harvested root and shoot tissues. This design provides the opportunity to determine conservation of DNA methylation as well as dynamic changes between and within species. In addition to extracting DNA for bisulfite-sequencing in duplicate, we also extracted RNA in triplicate for RNA-seq.
Bisulfite-treated samples were sequenced to an average of 20× strand-specific coverage (Table S2). With this coverage, over 97.5% of the cytosines in the non-repetitive portion of the reference genome of each species could be interrogated (99.5% for C. rubella, 97.5% for A. lyrata, and 98.7% for A. thaliana). With a minimum coverage of three, we confidently estimated methylation rates at two thirds to three quarter of cytosines (62% for C. rubella, 65% for A. lyrata, and 75% for A. thaliana). Sites with significant methylation levels were identified using a binomial test [39]. False positive rates, determined from incomplete conversion of exogenous unmethylated phage lambda DNA, were very low (Table S3).
Global patterns of DNA methylation in A. lyrata and C. rubella are similar to those reported before for A. thaliana, with highest levels in regions near the centromeres, which are populated by TEs and repeats, but contain few genes [6], [14], [15] (Fig. 1). There is little correlation between DNA methylation density and gene expression at the 500 kb scale (Fig. 1). Centromeric regions are plagued with TEs, and as expected, methylation is found preferentially at sites annotated as residing in TEs (Fig. 2A). Methylation at CHG and CHH sites, which account for over half of methylated sites in all three species, occurs almost exclusively in TEs (Fig. 2A).
Methylation patterns in the three species reflect their genome architecture. While we mapped a similar number of methylated cytosines in A. thaliana and C. rubella, consistent with the almost equal size of euchromatic sequences in both species, we identified almost three times as many methylated cytosines in A. lyrata, even though its reference genome assembly is only 50 to 75% longer than that of the other two species. The larger number of methylated cytosines in A. lyrata has led to an elevation in the methylation rate at a number of genomic features (Fig. 2B). This increase has only occurred at CHG and CHH sites, hallmarks of RdDM at TEs, and is especially evident in introns, correlating with the invasion of introns by TEs in this species (Fig. 2B, C). Almost one third of intronic bases in A. lyrata overlap with a TE or repeat, compared to fewer than 10% in the other two species (Fig. 2C), with the expansion found for all TE classes (Fig. 2D). Intron-inserted TEs are frequently found in non-expressed genes (Fig. S1) and are associated with increased methylation in flanking intronic and exonic sequences (Fig. S2), potentially due to pseudogenization or incomplete annotation of repeats. However, when a TE is inserted into the intron of an expressed gene, elevation of CHG and CHH methylation of exon sequences is not evident (Fig. S2, S3). Despite TE expansion in A. lyrata, the level of A. lyrata gene body methylation is comparable to that of C. rubella, which has few TEs in its introns (Fig. 2E). However, species-specific differences in methylation patterns are evident in flanking UTR and intergenic sequence (Fig. 2E). In these regions A. lyrata is the most highly methylated in all contexts (Fig. 2E). Depending on context, C. rubella displays methylation levels either similar to A. thaliana or intermediate between the two other species (Fig. 2E).
Arabidopsis thaliana lost three centromeres relative to A. lyrata and C. rubella, and this loss has been estimated to account for about 10% of the genome size reduction in A. thaliana [46]. Using orthologous genes, it is possible to reconstruct the gene, repeat, and methylation density using the ancestral chromosome positions (Fig. 3). As expected, repeat density and cytosine methylation next to these degraded centromeres is reduced in A. thaliana, while gene density is higher (Fig. 3). Particularly notable is the decrease in CG gene body methylation (Fig. 3). Although gene body methylation is positively correlated with gene expression in several species [6], [7], [14], [15], [16], gene expression is not noticeably different in these regions between the three species (Fig. 3). Thus, the elimination of centromeres has had a measurable impact on repeat and methylation distribution in A. thaliana, but did not strongly affect the expression of ancestrally pericentromeric genes.
Methylation of plant genomes is driven to a large extent by TEs, which are silenced via either the sRNA-mediated RdDM pathway [36] or the RdDM-independent pathway which relies on DDM1 [38]. Using a Hidden Markov Model algorithm, we identified methylated regions (MR) in each genome, which have a median length of 300 to 530 bp and cover between 26 and 73 Mb (Table S4). MRs are preferentially found in heterochromatic sequence next to centromeres, as they are enriched for TEs (Fig. S4, Fig. 4A). Since TEs are rapidly turned over, we expected MRs to be only poorly conserved. To test this assumption, we identified nearly 60 Mb of sequences with a 1∶1∶1 relationship in whole-genome alignments (Table S5) [47]. Less than 1% of the MR space is contained in the alignable portion of the genomes (Fig. 4B). In the rare cases where an MR spans alignable sequences, such sequences are almost always methylated in only one of the three species (Fig. 4C). We conclude that DNA methylation targets primarily the variable portion of the genome, which is subject to species-specific expansion and contraction of TEs.
To determine whether specific orthologs tend to be associated with methylation in all species, even in the absence of MR sequence conservation, we analyzed orthologs that contained a MR overlapping or within 1 kb of their coding region. Again, we found that the presence of MRs is rarely conserved (Fig. 4D, Table S6), although MR sharing is seen more often than expected by chance (Fig. 4D, Tables S7, S8). This could, however, be simply due to genes near centromeres being more often associated with MRs because they are in an MR-rich genome environment.
In contrast to RdDM of TEs and other repeats, the function of CG gene body methylation is still enigmatic, although it correlates positively with gene expression and negatively with mean normalized expression variance, or the coefficient of variation, across tissues and treatments (Fig. S5) [6], [7], [14], [15], [16], [17]. CG gene body methylation is found in the majority of genes (Table S9), and its rate is highly correlated between orthologs, while CG methylation up- and downstream of genes is much less correlated (Fig. 5).
CHG and CHH methylation in gene bodies is often indicative of transcriptionally inactive pseudogenes, paralogs, or transposons wrongly annotated as protein coding genes [14], [15], [55]. Between 10 and 20% of genes exhibit CHG or CHH methylation, most of which were not expressed in our samples (Table S9). Genes with CHG or CHH methylation are underrepresented in the orthologous gene set, where their fraction drops to less than half of their fraction among all genes, supporting the assertion that CHG and CHH methylation point to a tendency toward pseudogenization (Table S9). Moreover, CHG and CHH methylation are generally not conserved, suggesting that these marks arise in a lineage-specific fashion.
We used the cross-species alignments to identify 15.1 million conserved CG, CHG and CHH sites, which are located particularly in exons (Fig. 6A, Table S5). Although only a small portion, 2%, had significant methylation, most were shared between at least two species, with A. thaliana having the fewest methylated sites, reflecting the general decrease in global DNA methylation in this species (Fig. 6B–D, Table S10). Sites methylated in multiple species are further enriched in exons, with very few of these conserved sites being CHG or CHH sites (Fig. 6B,C, Fig. S6).
Sites that differ in methylation between species can be used to study gain and loss of methylation. We consider sites that are methylated only in a single species as lineage-specific gains, and absence of methylation in only one species as lineage-specific losses. We found that the number of gains and losses reflect the differences in genome architecture between the three species (Fig. 6 B,D). The many methylation losses in A. thaliana appear to be the result of genome shrinkage, and this species has also the fewest gains. In contrast, A. lyrata has the most gains, likely reflecting recent TE expansion (Fig. 6 B,D). The density of variable sites across the genome (in 10 kb windows) illustrates that gains and losses are not randomly distributed (Fig. 6D). Species-specific gains, which occur in all three sequence contexts, are concentrated in a subset of windows that are strongly enriched for TEs (Fig. 6D,E), but are also frequently found in exons (Fig. S6). That methylation gains are particularly likely in first and last exons suggests that methylation spreading from nearby TEs makes an important contribution to newly methylated sites, regardless of TE class (Fig. 6F, S7) [56], [57], [58].
Lineage-specific losses are more evenly distributed, without any signature of TE association. In addition, sites that are conserved in not only two, but all three species occur across a similar spectrum of genomic features (Fig. S6). Together these results indicate that unlike gains, losses occur in a random fashion, with the proviso that there is an overall global loss of methylation in A. thaliana (Fig. 6D). Though centromere elimination contributes to the different methylation pattern in A. thaliana, this explains only a minority of these losses (Fig. S8). It appears more likely that they are caused by the global reduction in TE content. We also attempted to understand what factors might contribute to conservation of DNA methylation over time. Sites found in more than one species are enriched in exons of conserved length and are more frequent in the center of exons (Fig. S9, S10).
Because several studies have shown that DNA methylation can change between tissues and in response to external stimuli [19], [20], we wanted to address whether these responses are conserved. Principal component analysis on the four types of samples, control shoots, cold-treated shoots, control roots and cold-treated roots, for all three species according to global RNA-seq measurements revealed that tissue is the most important factor, with over 7,000 genes being differentially expressed between roots and shoots (Fig. 7A, S11). Tissue-specific differences in gene expression are the largest source of expression variance in this data set (Fig. 7A). In contrast, species is the most important factor for differences in DNA methylation and explains 80% of the variance in our data (Fig. 7B, Fig. S12). Moreover, PC2 places A. lyrata closest to C. rubella instead of its congener A. thaliana, reflecting the methylation losses in A. thaliana (Fig. 7B).
To evaluate the degree to which within-species DNA methylation changes are conserved, we first estimated significant differential methylation at site and region levels. Four biologically appropriate comparisons were performed for each species to minimize multiple testing problems. Two tests identified differentially methylated positions (DMPs) between roots and shoots, and two tests identified DMPs between cold and control conditions regardless of tissue type. In each species, ten times as many DMPs were found between tissues than between treatments (Figure 8A, Table S11). Similar to DMPs, 20 to 50 times as many differentially methylated regions (DMRs) were detected between tissues than between treatments (Fig. 8B, Table S12).
Importantly, DMPs and DMRs do not necessarily coincide (Fig. S4, S13). DMPs in all contexts are rarely found within DMRs, indicating that significant regional changes in methylation are not just the extension of single base differences (Fig. 8C). CHG and CHH DMPs reside mainly within MRs (Fig. 8C); since these are almost exclusively found in the non-alignable portions of the genome, including TEs (Fig. 4A, Fig. 8D), the positions of DMPs and DMRs are typically not conserved between species (Fig. 8E). In the rare case that DMPs or DMRs can be found in the portion of a species' genome that can be aligned with the genomes of the other two species (Fig. 8E), they are only variable in a single species (Fig. 8F). Methylation variation at both the site and region level is therefore not conserved across species.
In the absence of sequence conservation at DMRs, we looked for conservation of their presence at orthologous genes. When only considering orthologs, fewer than 700 genes coincide with a DMR (405 in C. rubella, 652 in A. lyrata, and 221 in A. thaliana) (Table S13). Orthologs only rarely shared the presence of an overlapping or adjacent DMR, similar to what we see for MRs. Despite the rarity of such cases, they occur more often than expected by chance for a subset of genomic features and species comparisons (Fig. S14, Table S14, Table S15). Lack of sequence conservation together with minimal overlap of DMR presence at orthologs supports the transitory nature of methylation variation during genome evolution.
We also asked whether differential methylation in or near coding sequences is correlated with changes in gene expression. DMP and DMR overlap with genes was analyzed separately for those that overlapped with exons, introns, 5′ UTRs, 3′ UTRs and 1 kb upstream regions (Table S13, S16). DMPs occur in many genes in all three species, and most of them are expressed in our samples (9,631 in C. rubella, 12,216 in A. lyrata, and 6,345 in A. thaliana), but there is no evidence for correlation between DMPs and gene expression. This holds true for tissue as well as treatment DMPs (average Spearman rank correlation coefficient tissue = −0.04, treatment = 0.02, Table S17). Only a small number of DMRs overlap with expressed genes (529 in C. rubella, 801 in A. lyrata, and 284 in A. thaliana). Again, there is no correlation with gene expression (average Spearman rank correlation coefficient for CG DMRs = −0.16, CHG DMRs = −0.06, CHH DMRs = 0.00, Table S18).
Although DMPs and DMRs are not conserved across species, there is consistently more variability between root and shoot samples at a number of genomic features. Importantly, the methylation profile across transposons is quite different between tissues. Transposons are consistently more highly methylated in all sequence contexts in shoots (Fig. 9A). A similar trend is apparent for CHG and CHH sites in intergenic regions in A. lyrata, reflecting that TEs are closer to genes in this species (Fig. 9B) [46].
DNA methylation is an ancient epigenetic modification that appears in the genomes of organisms throughout the eukaryotic phylogeny [1], [2], [3]. This mark is associated with a number of cellular processes including transposon silencing and host gene regulation, but the cause-and-effect relationship between gene expression and DNA methylation remains unclear [6], [7], [12], [13], [14], [15], [16]. From an evolutionary standpoint, it is useful to consider methylated cytosines from two differing perspectives, either as a non-canonical nucleotide or as a molecular phenotype akin to transcription, and each perspective has important implications for the interpretation of its evolutionary dynamics.
As a molecular phenotype, many characteristics of DNA methylation are conserved between the species we examined. DNA methylation is generally associated with the repeat-dense sequences found in the centromeres, with CG methylation being in addition present at high levels in exonic sequences [6], [7], [8], [9], [10], [11], [14], [15], [16]. Furthermore, gene body methylation levels are conserved in orthologous genes indicating that DNA methylation rate may be subject to purifying selection, a finding consistent with previous wider evolutionary comparisons [42]. The close relationship of the species used in our experiments allows us to make inferences at base pair resolution. Given the substantial rate of epimutation in non-repetitive sequences [39], [40], we were surprised to discover that a large fraction of sites is methylated in more than one species. These sites were predominantly found in gene bodies, providing additional evidence for selective constraint. While gene body methylation is poorly understood, there is some evidence that it is correlated with nucleosome positioning in exons [14], [59]. If nucleosome position is conserved, it could potentially explain long-term conservation of DNA methylation at some sites.
An additional proposed feature of DNA methylation as a molecular phenotype is the ability to respond to external stimuli or internal developmental cues. In theory, such variation could control changes in gene expression. We found evidence for DNA methylation variation in all three species across both tissue type and environment. The changes in DNA methylation were in all three species much greater between tissues, and consistently resulted in lower methylation levels in the root [19]. Differences between the root and shoot tissues also explain a majority of the expression variation in the transcriptional data, but these changes are not directional. We found no evidence that changes in DNA methylation across tissues is associated with changes in gene expression. In fact, a large proportion of methylation changes were found in repetitive sequences. This pattern may result from the increased stringency of transposon silencing in the shoot, which includes the plant germline [60].
While transcriptional responses are highly conserved across all three species, we found no evidence for conservation of DNA methylation response at the sequence level. MRs and DMRs are predominantly found in the rapidly evolving repeat-rich regions of the genome and rarely reside in or near the same orthologous gene in more than one species. In many of the classical epimutants, epigenetic regulation of nearby transposon insertions can impact neighboring genes and cause phenotypic variation [21], [24], [25], [26]. This additional regulation is in some cases beneficial; for example, for genes specifically expressed in the pollen [41], [61]. The data presented here demonstrates that these events are both rare and likely lineage-specific. It is possible that the reported cases of differential methylation as a regulator of transcription are short-term innovations that are eventually replaced by genetically encoded regulation.
The mode of inheritance of symmetrically methylated cytosines motivates the interpretation of DNA methylation as a molecular modification that increases the complexity of the genetic code. While mutational processes affecting DNA sequence are well described, epimutational processes are poorly understood. DNA mutations rarely revert and occur in a largely random fashion throughout the genome [62]. In contrast, recent studies have shown that the transgenerational stability of DNA methylation is very context dependent [39], [40]. Over short evolutionary times, epimutations are more likely to occur in euchromatic sequences and are biased away from heavily methylated repetitive sequences [39], [40].
Over the longer evolutionary times examined here, we find that changes in genome content and structure are the major contributors to DNA methylation variation. While the majority of single site and regional methylation is found in repetitive sequences that are unlikely under evolutionary constraint, the remaining observed patterns in euchromatic sequence reflect lineage-specific evolution of transposons. This is particularly obvious in A. lyrata, which has experienced a recent invasion of transposable elements into euchromatic sequences [46] and subsequent elevation in the methylation rate of euchromatic features, particularly introns.
Large-scale structural changes that have perturbed the genome-wide DNA methylation landscape have also occurred in A. thaliana [48], [49]. Loss of three repeat-rich centromeres in A. thaliana caused a decrease in DNA methylation in sequences flanking the ancestral centromeres. The impact of lineage-specific transposon evolution and subsequent methylation is similarly evident in genic sequences. Approximately 40% of methylation in conserved exon sequence is species-specific. These sites are non-uniformly distributed near the 5′ or 3′ edges of genes, likely due to spreading from adjacent transposons [56], [57], [58]. These observations support the hypothesis that surveillance of transposons is the primary contributor to the genomic distribution of DNA methylation in plants. Since transposon content and genome structure vary extensively even over short evolutionary time periods, DNA methylation appears to be similarly variable. This is supported by the poor resolution of species relationships in a principal component analysis of DNA methylation and a nearly ten-fold increase in divergence between A. lyrata and A. thaliana when comparing DNA methylation as opposed to nucleotide sequence [46]. Together, these results indicate that DNA methylation as a non-canonical nucleotide is very rarely conserved over intermediate evolutionary times scales.
Despite the fact that we can estimate the epimutation rate of methylated cytosines and other parameters related to nucleotide mutations, it is misleading to equate DNA methylation changes to nucleotide substitutions. Our results indicate that the rapid evolution of repeat sequences is the major contributor to the equally rapid changes in the genomic distribution of DNA methylation. In this respect, it is more reasonable to regard DNA methylation primarily as a molecular phenotype resulting from the underlying genetic sequences. Although a few “pure” epialleles have been identified in nature, the majority of natural epimutations are linked to nearby transposon insertions or other genetic changes [21], [24], [25], [26]. Fast evolution of repeat-sequences can, however, provide opportunities for lineage-specific cooption of DNA methylation for regulation of endogenous genes in response to various stimuli.
Seeds from the reference strain for each species (A. thaliana Col-0, A. lyrata MN47, C. rubella MTE) were sterilized with a 15 minute treatment of 30% bleach and 0.1% Triton X-100. Sterilized seeds were plated onto 0.5× MS 0.7% agar plates with 1% sucrose. Each plate represented a single replicate consisting of 20 seedlings. In total, 7 replicates were sown and randomized into a 3×2×2 factorial design. The three factors in this experiment were species, tissue, and cold treatment. After sowing, plates were stratified in the dark at 4°C for 8 days, before being shifted to 23°C short-day conditions (8 hr light∶16 hr dark). Plates were oriented vertically. After 6 days in 23°C, half of the plates were exposed to 4°C short-day conditions for 23 hours. At the end of the cold treatment, both control (23°C) and treated (4°C) samples were harvested. Root and shoot tissues were harvested independently. Plants were cut just above and below the root-shoot junction to separate the tissues and avoid cross contamination of tissue types. To minimize daily collection times, replicates were blocked by day.
Total RNA was isolated from three replicates of each factor combination using the Qiagen RNAeasy Plant Mini Kit (catalog # 74904). An on-column DNase digestion was included (catalog # 79254). Total RNA integrity was confirmed on the Agilent BioAnalyzer. Illumina TruSeq RNA libraries were constructed using 3 µg of total RNA. Samples were randomized before library construction. The manufacturer's protocol was followed with one exception - 12 PCR cycles were used instead of the recommended 15. Libraries were quantified on an Agilent BioAnalyzer (DNA 1000 chip). Samples were normalized to 10 nM library molecules and then pooled for sequencing. Three pools were constructed, each consisting of 12 random samples. Each pool was sequenced across three lanes of an Illumina GAII flowcell.
DNA was extracted from two replicates of each factor combination using the Qiagen DNeasy Plant Mini Kit (catalog # 69104). DNA was quantified using the Qubit BR assay (Life Technologies, catalog # Q32853). Bisulfite libraries were confstructed using modifications to the Illumina TruSeq DNA kit and published bisulfite library protocols [15], [39]. Depending on the sample, starting material ranged from 200 ng to 1 µg. Changes to the manufacturer's protocol will be noted here. After shearing of genomic DNA with a Covaris S220 instrument, sheared lambda DNA was spiked into each sample (1∶0.001 sample∶lambda ratio) as a control., for accurate estimation of failure to bisulfite convert non-methylated cytosines. Samples were randomized before library construction. During the ligation step, the amount of adapter was adjusted based on the amount of starting material in each sample. For 1 µg of input DNA, 2.5 µl of adapter were used. Adapter input was scaled linearly for samples with less starting DNA. For the second AMPure bead clean up after the ligation step, the ratio of sample to beads was adjusted to 1∶0.74. A final elution volume of 42.5 µl was used for this step. After ligation, 40 µl of eluate was transferred to a new tube for subsequent bisulfite treatment.
The Qiagen Epitect Plus Kit (catalog # 59124) was used for bisulfite treatment. The manufacturer's protocol for ‘low concentrated and fragmented samples’ was followed, using 85 µl of bisulfite mix for conversion. Clean up of the bisulfite reaction included ethanol as a final wash step. The sample was eluted in 17 µl. After bisulfite treatment samples were amplified using Pfu Cx HotStart Polymerase from Agilent (catalog # 600410) instead of the supplied PCR mix. Reaction conditions are all follows: 32.9 µl of water, 5 µl of 10× Pfu Cx Buffer, 5 µl of 2 mM dNTP, 1.6 µl of Illumina PCR Primer Cocktail, 0.5 µl of Cx Polymerase (2.5 U/µl), 5 µl of bisulfite-treated DNA eluate. Three PCR reactions were pooled for each bisulfite-treated sample. The following cycling conditions were used: 98°C - 30 seconds; 18 cycles of 98°C - 10 seconds, 65°C - 30 seconds, 72°C - 30 seconds; 72°C - 5 minutes. An AMPure bead clean up was used to purify the final PCR product (1∶1 sample to bead ratio). Samples were eluted in 32.5 µl of Illumina supplied Resuspension buffer. 30 µl of the final eluate was transferred to a new plate for subsequent quantification and sequencing. Libraries were quantified using the Agilent BioAnalyzer (DNA 1000 chip). Libraries were diluted to 10 nM and then pooled. Samples were pooled based on genome size - and each pool consists of 2 random samples from each species. Four pools were constructed and each was sequenced across three lanes of the Illumina HiSeq 2000.
We sequenced bisulfite-converted libraries with 2×101 base pair paired-end reads on an Illumina HiSeq 2000 instrument with conventional A. thaliana DNA genomic libraries in control lanes. Each sample contained 0.1% lambda DNA as an unmethylated control. We pooled six different samples in each lane. The Illumina RTA software (version 1.13.48) performed image analysis and base calling.
Reads were filtered and trimmed as previously described [39]. Subsequently, trimmed reads were mapped against the corresponding reference genomes (Crubella_183, Alyrata_107, Athaliana_167 (TAIR9) [46], [47], [50], [51]. The lambda genome sequence was appended to each species genome sequence in order to estimate the false methylation rates of each sample. All reads were aligned using the mapping tool bismark v0.7.3 [63]. Applying the ‘scoring matrix approach’ of SHORE as previously described [39], we retrieved unique and non-duplicated read counts per position. Read and alignment statistics can be found in Table S2. All command line arguments are listed in Text S1. Raw reads are deposited at the European Nucleotide Archive under accession number PRJEB6701.
We used published methods [39], with a few exceptions. Here we retrieved incomplete bisulfite conversion rates, or false methylation rates (FMRs), from the alignments against the lambda genome rather than the chloroplast sequence. False methylation rates are found in Table S3. In addition, we combined the read counts of replicate samples after removing sites that were differentially methylated between replicates. The methylation rates for combined replicates were used for all subsequent analyses. The number of DMPs detected between replicates can be found in Table S19. In each species we required a methylation rate of at least 20% in one of the four tissue-treatment combinations in order for a site to be considered significantly methylated.
To identify DMPs we followed published methods [39], but we required positions to have a methylation rate of at least 20% in one of the treatment combinations before performing Fisher's exact test. This increased statistical power by reducing the number of multiple testing corrections. Pairwise tests were not performed between all treatment combinations, instead only relevant comparisons were performed within each species (Root-23°C vs Shoot-23°C, Root-4°C vs Shoot-4°C, Root-23°C vs Root-4°C, Shoot-23°C vs Shoot-4°C).
To detect contiguously methylated parts of the genome we modified a Hidden Markov Model (HMM) implementation [64]. Briefly, each cytosine can be in either an unmethylated or methylated state. The model trains methylation rate distributions for each state and sequence context (CG, CHG, CHH) independently using genome-wide data. In addition, transition probabilities between the states are trained. To make the original HMM implementation applicable to plant data, three different (beta binomial) distributions were estimated for each state (methylated and unmethylated) instead of just the single distribution used in mammals, which have almost only CG methylation [64]. To prevent identification of regions over uncovered bases, the genome was split at locations that lacked a covered cytosine position for 50 adjacent base pairs. On each of these segments, the most probable path through the methylation states was estimated after genome-wide parameter training. Transitions between states demarcated the methylated regions (MR). Replicates of each treatment combination were combined for this analysis. The combined read counts at cytosines were used to calculate methylation rates, train the HMM, and identify methylated regions. As a result, there is a single segmentation of the genome per treatment combination. Methylated regions were trimmed on both 5′ and 3′ ends by removing positions with a methylation rate below 10%. Further details will be described in a manuscript by Hagmann, Becker et al. [65].
Based on the MRs identified for each sample using the HMM algorithm described above, we selected regions of variable methylation state between samples to test for differential methylation. Due to the very large number of MRs, it was critical to reduce the number of tests performed to identify DMRs. By filtering MRs using the criteria outlined in a forthcoming manuscript by Hagmann, Becker et al. [65], we reduced the number of MRs four fold in each species. For each identified region, pairwise statistical tests were performed for the relevant comparisons listed above. The statistical test approximates the context-specific beta binomial distribution for the region of interest. Individual and joint distributions are approximated for two samples being compared. The statistical test compares the individual sample distributions to the joint distribution using a log-odds ratio. This ratio is compared against a chi-squared distribution to obtain confidence values. For each identified region, samples were assigned to groups by separating the samples with statistically significant methylation. To confirm groupings, we first combined read counts from treatment combinations in the same group. With the combined data, the same statistical test as described above was performed to test for differential methylation. Groups were confirmed in this way to identify and filter potentially erroneous DMRs. After false discovery rate (FDR) correction using Storey's method [66], regions with an FDR below 0.01 were defined as differentially methylated regions (DMRs). To resolve overlapping DMRs, we retained the non-overlapping regions containing the maximum number of samples with statistically significant differential methylation. Apart from the criterion used to resolve overlapping DMRs, the methods follow those that will be described in detail in a manuscript by Hagmann, Becker et al. [65].
We identified conserved sites using a published three-way whole genome alignment [47]. For CG sites, identical context was required while substitutions at the H positions were allowed in degenerate contexts as long as they did not mutate to G. Sites that transitioned contexts were not considered. Methylation rates for significantly methylated sites were then extracted from each species, tissue, and treatment combination for subsequent analysis.
Three-way orthologs were identified using the reciprocal-best blastp hit approach as implemented in the multiParanoid pipline (inParanoid v. 4.1, blast v. 2.2.26) [67].
We sequenced each RNAseq library with 101 base pair single-end reads on the Illumina GAII instrument. We pooled twelve different samples in each lane. Each pool was sequenced over three lanes. The Illumina RTA software (version 1.13.48) performed image analysis and base calling.
Reads were trimmed using the shore import function in SHORE version 0.9.0 [68]. Command line arguments can be found in Text S1. This function simultaneously trims reads and separates samples by barcode. Since all samples were sequenced over three lanes, after lanes are de-multiplexed sample reads were combined. Due to variable annotation qualities between species, only sequences annotated as CDS annotations were used to map RNA-seq reads. The following representative gene model annotation versions were used for each species: Crubella_183, Alyrata_107, Athaliana_167 (TAIR10) [46], [47], [50], [51]. Reads were aligned with one allowed mismatch to the appropriate annotation using bwa version 0.6.1 [69]. Read counts were obtained for each gene using a custom perl script. In summary, the script identified uniquely aligned read with a mapping quality score above 30 and stored the total read count for each target sequence. Read and alignment statistics can be found in Table S20. Raw reads are deposited at the European Nucleotide Archive under accession number PRJEB6701.
Differentially expressed genes were identified using the R package edgeR (3.4.2) with minor modifications [70]. Using edgeR, we estimated the dispersion parameter for each gene using estimateGLMTagwiseDisp(). Next, we fit a negative binomial generalized linear model (GLM) using glmFit(). Significance testing for differential expression was performed using a custom GLM. Significance testing in edgeR was done via term-dropping of each factor level (likelihood ratio test), and as a result performed more statistical tests than necessary. To minimize multiple testing problems, we implemented a negative binomial GLM that tested for differential expression significance using an ANOVA [71]. Dispersion estimates from edgeR were provided to the modified GLM. Using this model, differential expression analysis was performed in two ways. First, expression analysis was performed within species. There were 12 samples consisting of three replicates and four unique treatment combinations. All representative gene models were considered. The following custom GLM model was used: expression∼tissue*treatment. This included the main effects of tissue and treatment as well as their interaction. Secondly, we performed differential expression analysis between all species simultaneously. In this case, there are a total of 36 samples consisting of three replicates of each species, tissue, and treatment combination. Only 1∶1∶1 orthologous gene pairs were considered (14,395 in total). The following custom GLM model was used: expression∼species*tissue*treatment. This includes the main effects of species, tissue, and treatment as well as all two and three-way interactions. Corrections for gene length were performed, but this did not impact the results and was subsequently ignored.
Transposon and repeat annotations for all three species were derived from the Capsella rubella genome paper [47], [72], [73].
|
10.1371/journal.pntd.0006082 | Canine visceral leishmaniasis: Diagnosis and management of the reservoir living among us | This article reviews essential topics of canine visceral leishmaniasis (CVL) due to Leishmania infantum infection. It focuses on the current serological and molecular diagnostic methods used in epidemiological research and veterinary clinics to diagnose CVL and includes new point-of-care (POC) tests under development. The efficacy of different treatment regimens on the clinical improvement and infectiousness of dogs is also addressed. In the last section, the review provides a critical appraisal of the effectiveness of different control measures that have been implemented to curb disease transmission.
| Dogs are the principal reservoir hosts of L. infantum and consequently play a critical role in the transmission cycle of urban VL, which also affects humans. This review provides updated information on important topics such as diagnostic tests and dog treatments that improve dog health and decrease their transmission efficacy to insect vectors. A critical review of control measures is also provided.
| CVL negatively impacts society from medical, veterinary, and societal standpoints. Studies on risk factors for human infection with L. infantum have yielded opposing results, but a meta-analysis suggested that owners of infected dogs and household members could be at high risk of infection, at least in the Americas [1]. Based on this generalized concept, strategies to comply with public health guidelines typically lead to difficult or expensive decisions. In developed countries, infected dogs are subjected to different treatment protocols that improve the clinical condition but do not clear L. infantum, while in developing countries, the recommended euthanasia of infected dogs generates societal conflicts [2–4]. Herein, we review current serological and molecular tools for the diagnosis of CVL, the impact of treatment on infectiousness, and control strategies to prevent infection and disease development.
The selection of studies referenced in this review was based on searches in PubMed, Library of Congress, Web of Science, Scielo, and Google Scholar, with no specific year range. The search strategy included combinations of the following key words: “canine visceral leishmaniasis,” “leishmaniasis,” “diagnosis,” “molecular diagnostic methods,” “serology,” “treatment,” “xenodiagnosis,” “infectivity,” “prevention,” “control,” “dog culling,” and “vaccination.” Paper selection was grounded in the specific information each article provided to the different sections of this review.
The analytical sensitivity of molecular tests suggests they can detect between 0.001 and 0.1 parasite/reaction [5–8]. However, determining the actual diagnostic efficacy at different infection stages could be problematic due to the relatively heterogeneous clinical criteria used in different studies. Before guidelines were established [9], categorization as oligosymptomatic or polysymptomatic animals according to the number of CVL signs and symptoms could vary between studies, making them difficult to compare.
The seroconversion to parasite antigens can be as early as one month after an infective phlebotomine bite [32]. Active CVL is usually associated with significant antibody titers of all classes, whereas low antibody levels are characteristic of subclinical infections or exposed but uninfected dogs [33]. Although the use of different antibody isotypes was proposed for improving serological evaluations [34–36], the most recent data from a cohort of 134 dogs suggest that isotype responses have no major predictive value [37]. Increased levels of immunoglobulin G2 (IgG2) were associated with protective responses, while rising IgG1 production was considered as bad prognosis [35]. IgE and IgA seem to be detected mostly in active CVL but with low predictive value in asymptomatic dogs [35,38,39].
Clinical and epidemiological studies have used serological tests involving whole parasites, soluble parasite extracts, or recombinant proteins derived from genes of interest. The recent development of chimeric antigens with relevant protein epitopes demonstrated the capacity of this method to detect specific antibodies during active disease or asymptomatic infection [40,41].
The direct agglutination test (DAT) is based on the agglutination of trypsinized Coomassie-stained Leishmania promastigotes by anti-Leishmania antibodies. It was the first serological test developed for field use. It is simple, cheap, and reliable, with proven clinical accuracy (S1 Table) [42,43]. Moreover, it can be performed in laboratories not requiring electrical equipment and has up to 2 years of shelf life. DAT has long incubation times and requires some level of expertise to run and read the test [43,44]. It has a sensitivity and specificity of 91% to 100% and 72% to 100%, respectively, yet subjective reading of end-point titers leads to interobserver discrepancy [43,45]. Despite those drawbacks, DAT is well accepted as a routine serologic test usually applied to a large number of samples [46]. The fast agglutination screening test (FAST) is a modification of DAT based on a single serum dilution above the cutoff point of normal sera. It requires shorter incubation times and has been optimized for screening large dog populations [46].
The immunofluorescence antibody test (IFAT) against Leishmania promastigotes is the reference qualitative serological method for CVL diagnosis [47]. The use of IFAT is restricted to laboratory settings because it needs specialized equipment and trained personnel [48]. The specificity and sensitivity are close to 100% in symptomatic animals. Some notable limitations are the cross-reactivity with other pathogens such as trypanosomes [47,48] and the significantly lower sensitivity for identifying asymptomatic dogs compared with ELISA [41].
ELISA allows screening large numbers of samples utilizing antigen-coated microplates and a spectrophotometer that determines antibody titers by optical density. The potential for absolute antibody quantification renders ELISA as a powerful tool that is less susceptible to operator bias. One of its strengths is the possibility of using combinations of multiple antigens, thereby increasing the sensitivity and/or specificity of the method (S2 Table) [49,50].
Flow cytometry (FC) is an emerging technology [51] that quantifies antibodies against Leishmania surface antigens, avoiding cross-reactivity against more conserved intracellular structures. Using amastigotes or promastigotes, this method achieved high levels of sensitivity and specificity and has been shown to distinguish serological profiles of infected but clinically healthy versus sick dogs [52].
In Europe, the treatment of CVL has been almost exclusively limited to the use of pentavalent antimony meglumine antimoniate. The recommended regimen of 35 to 50 mg/kg subcutaneously twice daily for 4 to 6 weeks [9] demonstrated good clinical efficacy but without clearing the infection.
The combination with allopurinol showed better response to treatment of sick dogs, with good clinical recovery and improvement of hematological and biochemical abnormalities [73]. Allopurinol administration (10 mg/kg by mouth twice daily) for 6 to 12 months after an antimonial course was highly leishmaniostatic, maintaining treated dogs in long-term clinical remission [74].
Other leishmanicidal drugs such as miltefosine at a dosage of 2 mg/kg by mouth once daily for 4 weeks, in combination with allopurinol, demonstrated leishmanicidal efficacy in naturally infected dogs [75]. Other drugs against CVL have been studied in vivo or in vitro such as aminosidine, pentamidine, enrofloxacine, and marbofloxacine, but further controlled clinical trials are needed [76,77]. Some of them might be used as an alternative when first line therapy fails or renal function is altered [78]. A new therapeutic trend is the combination of parasiticidal–parasitostatic drugs and immunomodulators aimed at reducing parasite burden and establishing an appropriate immune response (domperidone or Protein Aggregate Magnesium-Ammonium Phospholinoleate-Palmitoleate Anhydride (P-MAPA) [79,80].
Nevertheless, most dogs remain infected and might relapse, becoming infectious to healthy dogs and other hosts, including humans. Until new therapies that consistently clear L. infantum are found, dog infectivity can be managed by applying topical repellents that can reduce transmission risks to near zero (see section “Are prevention and control strategies working?”).
Xenodiagnosis (sand fly feeding on host) is the best alternative to determine dog infectiousness, but this method can only be applied in specialized research centers [81]. The first research using posttreatment xenodiagnosis that demonstrated significant reduction of dog infectiousness was published by Gradoni et al. [82]. Afterwards, Alvar et al. [83] evaluated six dogs treated with meglumine antimoniate in combination with allopurinol, reporting that all dogs were noninfective for a “few months” after chemotherapy. Guarga et al. [84] evaluated dogs (n = 10) treated with meglumine antimoniate and found a significant reduction in dog infectivity, which persisted until the end of the study (120–180 days). Another study used Lutzomyia longipalpis to evaluate the effectiveness of a liposome formulation of meglumine antimoniate [85]. Dogs were treated parenterally with antimonials, empty liposomes, or isotonic saline solution. A significant reduction of dog infectivity was found in the group treated with antimonials as determined 150 days post treatment. More recently, Miró et al. [86] evaluated 32 dogs with CVL that were subjected to three different treatments: antimonials, antimonials plus allopurinol, or allopurinol alone. They showed a considerable reduction of infectivity of all three groups and significant decrease of parasite burden in bone narrow. A study conducted in Brazil by da Silva et al. [87] included 52 dogs distributed in six treatment groups: liposomal formulation of antimonials, allopurinol, liposomal formulation of antimonials plus allopurinol, empty liposomes plus allopurinol, empty liposomes, and saline solution. The negative xenodiagnosis and qPCR quantification of L. infantum in the skin below the infectious threshold indicated that antimonials plus allopurinol was the most efficient regime to decrease infectivity to sand flies.
A recent study in Brazil (n = 36 dogs) assessed the infectivity of sick dogs after a conventional miltefosine treatment. After three months of treatment, there was a significant reduction of parasite load in the bone marrow, lymph nodes, and skin. These results correlated with the xenodiagnoses in which 74.2% of dogs were noninfectious for sand flies [88].
In conclusion, treatment of sick dogs in endemic areas decreases canine infectiousness, thus diminishing the epidemiological risks for humans and other uninfected dogs. Assessment of parasite loads in the ear skin by qPCR has been proposed as surrogate marker of infectiousness [89] and may be used whenever xenodiagnosis is not available. Nevertheless, further studies on posttreatment infectiousness of canines using different drugs are still needed. Despite not being a simple method, xenodiagnosis is a useful tool to assess the infectious capacity of dogs treated with new drugs and/or new treatment regimens.
Several strategies have been proposed for preventing and controlling CVL at both individual and population levels. Prevention of infection can be achieved by applying insecticide-impregnated collars or spot-on products (e.g., deltamethrin, permethrin, flumethrin, fipronil) on dogs, whereas the risk of disease development can be reduced by vaccination or immunomodulation. The effectiveness of some of these strategies has been assessed by recent systematic reviews and meta-analyses [90,91]. The conclusions based on either parasitological or serological evidence were that repellents and prophylactic medication (i.e., domperidone) tended to reduce the proportion of dogs infected with L. infantum [90]. Nonetheless, Wylie et al. [91] also concluded that well-designed, adequately powered, and properly reported randomized clinical trials are needed to clearly establish the efficacy of vaccines against CVL.
Culling of infected dogs has been recommended as a control strategy in many endemic countries. A cluster-randomized trial in Northeast Brazil reported a low to moderate effectiveness of dog culling and concluded that there is an urgent need for revision of the Brazilian VL control program [92]. González et al. [93] reviewed two trials from Brazil that evaluated the effects of culling infected dogs compared to no intervention or indoor residual spraying. Although these trials reported a reduction in seroconversion over 18-month follow-up, they did not measure or report effects on clinical disease in humans [93]. Mathematical models suggest that dog culling alone is not effective in areas of high transmission [94]. According to Costa et al. [94], the indiscriminate culling of healthy, seropositive dogs may jeopardize the effectiveness of the control program if low specificity tests are used, thus increasing the chance of generating outrage in the population and reducing the adherence to the program. In Iran, the utilization of insecticide-impregnated dog collars in nine villages showed significantly decreased seroconversion in children (odds ratio [OR] 0.57; 95% CI 0.36–0.90; p = 0.017) and dogs (OR 0.46; CI 0.30–0.70; p = 0.0003) compared with nine nonintervened villages [95].
While the above-mentioned strategies (e.g., repellents, vaccination, and immunomodulation) may work at the individual level, their effectiveness at the population level still needs to be demonstrated when the intervention is transferred to the communities. Indeed, the effectiveness of strategies like community-wide application of insecticide-impregnated collars is directly dependant on coverage and loss rate [96]. This is particularly unrealistic in developing countries, considering the limited economic and human resources, especially in periods of crisis and political upheavals. Moreover, in these countries, stray dogs will not be targeted by such strategies and may function as infection reservoirs. As for any zoonosis, stray dog population management should be part of any VL control programs.
Molecular tools are mainly restricted to research laboratories, but progress is being made towards field applicability. The costs of real-time PCR machine and the need for sophisticated laboratory infrastructure highlight the importance of validating and implementing POC molecular tests that could be used in-clinic and in the field. The veterinarians should rely also on complementary serological information to make the best decision regarding both the animal’s health and the epidemiological risk it entails. The remaining challenge for serological tests is to improve the capacity to differentiate clinically healthy but infected and vaccinated dogs. The evaluation of new antileishmanial drugs should be complemented by standardized follow-up that includes the infectiousness status of dogs at different times post treatment. Nowadays, veterinarians and dog owners have different options to decrease infection risks, of which repellents are still the most important prevention tools as demonstrated by laboratory and field studies. New vaccines that could reduce the risk of disease development and infectiousness of vaccinated dogs are urgently needed.
|
10.1371/journal.pgen.0030012 | A Multi-Step Pathway for the Establishment of Sister Chromatid Cohesion | The cohesion of sister chromatids is mediated by cohesin, a protein complex containing members of the structural maintenance of chromosome (Smc) family. How cohesins tether sister chromatids is not yet understood. Here, we mutate SMC1, the gene encoding a cohesin subunit of budding yeast, by random insertion dominant negative mutagenesis to generate alleles that are highly informative for cohesin assembly and function. Cohesins mutated in the Hinge or Loop1 regions of Smc1 bind chromatin by a mechanism similar to wild-type cohesin, but fail to enrich at cohesin-associated regions (CARs) and pericentric regions. Hence, the Hinge and Loop1 regions of Smc1 are essential for the specific chromatin binding of cohesin. This specific binding and a subsequent Ctf7/Eco1-dependent step are both required for the establishment of cohesion. We propose that a cohesin or cohesin oligomer tethers the sister chromatids through two chromatin-binding events that are regulated spatially by CAR binding and temporally by Ctf7 activation, to ensure cohesins crosslink only sister chromatids.
| Complexes containing members of the structural maintenance of chromosomes (Smc) family regulate higher order chromosome architecture in diverse aspects of DNA metabolism including chromosome condensation, sister chromatid cohesion, DNA repair, and global control of transcription. Smc complexes are thought to regulate higher order chromosome folding by tethering together two strands of chromatin. However, the mechanism of tethering is poorly understood in part because of a poor understanding of the function of the core Smc subunits. To gain insight into the structure and function of Smc subunits, we developed a novel strategy of mutagenesis called random insertion dominant negative (RID), which generates informative alleles with high efficiency and should provide an effective tool to study any multi-subunit complex. Using RID we generated novel alleles of a Smc subunit from the cohesin complex. The cohesin complex tethers together newly replicated chromosomes (sister chromatids). The analyses of these RID mutants suggest that the tethering activity of cohesin (and possibly other Smc complexes) is generated by two sequential chromatin-binding events (first the capture of one piece of chromatin followed by the capture of the second piece of chromatin), which are regulated both spatially and temporally. We speculate that the spatial and temporal regulation of cohesin ensures that it tethers together only sister chromatids rather than randomly crosslinking the entire genome.
| Proper transmission of eukaryotic chromosomes during cell division requires DNA replication and three other DNA-dependent processes: recombination-dependent DNA repair, sister chromatid cohesion, and chromosome condensation. Each of these diverse processes requires protein complexes containing two members of the highly conserved structural maintenance of chromosomes (Smc) family of proteins [1–3]. Smc complexes likely share a common core activity of chromosome crosslinking, either within a chromosome, as in chromosome condensation, or between chromosomes, for sister chromatid cohesion and recombination-dependent DNA repair. How Smc complexes mediate chromosome crosslinking is unknown.
Smc molecules are composed of five structural domains (Figure 1A) [4,5]: a globular N-terminal domain containing a Walker A motif, a globular C-terminal domain with Walker B and Signature motifs, two long α-helical domains, and a globular Hinge domain. Smc monomers fold in half at the Hinge domain, allowing the two α-helices to form a long antiparallel coiled-coil domain [6]. This folding juxtaposes the N- and C-terminal globular domains and the Walker A and B motifs, creating an Smc head domain with ATPase activity. Folded Smc monomers resemble a flexible dumbbell, with the Hinge and head domains separated by ∼40 nm of coiled coil [6,7].
Smc complexes are composed of two Smc molecules, a kleisin subunit, and at least one accessory protein [6,8,9]. Smc monomers dimerize primarily through interactions between their Hinge domains [6,10]. The head domains of Smc molecules are also tethered together through the shared binding of a single kleisin subunit and two ATP molecules [6,11,12]. These interactions at both the head and Hinge domains give Smc dimers the potential to form large rings that have been observed in preparations of purified Smc complexes [13].
One of the most studied Smc complexes is cohesin, which mediates sister chromatid cohesion. Cohesin is composed of Smc1, Smc3, Scc3, and the kleisin subunit Mcd1/Scc1 [14–18]. The association of cohesin molecules with chromatin requires the integrity of all cohesin subunits and the ability of the Smc molecules to bind and hydrolyze ATP [19–22]. Multiple cohesins bind proximal to each centromere, forming a large pericentric domain. Cohesins also bind along chromosome arms. In budding yeast, these cohesin-associated regions (CARs) extend over approximately 1 kb of DNA and are spaced at roughly 10-kb intervals [23–25].
Several auxiliary factors contribute to the establishment, maintenance, and eventual dissolution of sister chromatid cohesion. The binding of cohesin to chromosomes at any phase of the cell cycle requires the loading factors Scc2 and Scc4 [26]. However, only cohesin binding in S phase, coupled with the function of Ctf7, results in the establishment of cohesion [16,27]. Pds5 binds cohesin and helps maintain cohesion during S and G2 phases of the cell cycle [28–30]. Finally, separase promotes removal of cohesin at the onset of anaphase by cleavage of Mcd1 [31].
These observations have led to two distinct models to explain how cohesin molecules crosslink sister chromatids. The embrace model posits that cohesin rings encircle chromosomes prior to replication and make no specific contacts with chromatin [6,20]. A topological interaction between cohesin and chromatin is supported by the fact that cohesin can be released from chromatin either by a single cleavage of the DNA or a single proteolytic cleavage of a cohesin subunit [19,20,32]. Passage of the replication fork through cohesin rings leaves both sister chromatids trapped inside, establishing sister chromatid cohesion. In contrast, oligomerization models, on the basis of observations of other Smc complexes, posit that cohesins bind to both sister chromatids. Then, cohesins on one sister chromatid oligomerize with cohesins on the other sister chromatid to generate cohesion [33–36]. To resolve these and other models will require a better understanding of how cohesins bind chromatin and the relationship between this chromatin binding and establishment of cohesion.
Mutagenesis of cohesin subunits provides one approach to gain insight into the chromatin binding of cohesin mutants. Indeed, site-directed mutagenesis of conserved Smc1 residues in the Hinge, Walker A, Walker B, and Signature motifs demonstrated that Smc1 must bind Mcd1 and Smc3, as well as bind and hydrolyze ATP in order for cohesin to bind chromatin [7,21,22,37]. While informative, these mutational analyses of cohesin–chromatin association leave key questions unanswered. Do these mutations define all the domains of cohesin subunits required for chromatin binding? Is chromatin binding of cohesin anywhere on chromatin prior to DNA replication sufficient to generate cohesion? Are additional constraints on cohesin necessary to generate cohesion?
To address these types of questions, one needs a way to identify rare mutant forms of cohesin that modulate rather than eliminate its activity. In the past, knowledge of a protein's structure (sites for modification or interaction with other subunits) has been used to make dominant negative mutants that alter its activity or incorporation into a fully functional complex. We rationalized that the reciprocal would also be true; surveying an entire polypeptide chain for rare insertions with a dominant negative phenotype should provide a highly efficient means to identify precise regions of a protein important for its activity and/or assembly with other subunits. Furthermore, the study of these alleles should be highly informative for elucidating the molecular mechanism of a protein/complex. With this in mind, we screened a library of random insertion mutations in SMC1 for those that cause a dominant negative phenotype in the budding yeast, Saccharomyces cerevisiae. This strategy is henceforth referred to as random insertion dominant negatives (RID). Here, we successfully use RID to identify rare SMC1 alleles that are highly informative in dissecting Smc1′s role in the assembly and function of cohesin. We also use these mutants along with a ctf7 mutant to provide important insights into cohesin and sister chromatid cohesion.
As the basis of our RID mutagenesis of SMC1, we constructed a minichromosome containing the SMC1-TAP gene under control of the galactose-inducible GAL1 promoter (see Materials and Methods). The level of Smc1-TAP protein expressed from the GAL1 promoter relative to the endogenous SMC1 promoter decreases 0.5-fold under uninducing conditions and increases 100-fold under inducing conditions (unpublished data). The Smc1-TAP protein is functional, as it restores viability and normal growth rate to cells carrying the temperature-sensitive smc1–2 allele strain deleted for the essential SMC1 gene under repressing, uninducing, and inducing conditions (Figure S1 and unpublished data). The SMC1-TAP gene was mutagenized by a Tn7-based in vitro system that resulted in five amino acid insertions at random positions within the Smc1 protein. For brevity, the initial SMC1-TAP gene is henceforth referred to as the SMC1 or wild-type allele, and the insertion derivatives are named based upon the position of the insertion in the amino acid sequence.
Minichromosomes harboring the mutagenized library of SMC1 were transformed into budding yeast and assayed for their effects on cell viability and cohesion. A total of 13 transformants out of 2,500 candidates showed a marked decrease in cell viability under inducing conditions (Figure S1). The fact that these transformants are viable under uninducing conditions, where significant mutant smc1 expression occurs, indicates that these mutants are not dominant under low expression. To examine whether the mutant proteins are functional, minichromosomes containing inducible wild-type SMC1 or dominant negative smc1 alleles were introduced into cells carrying the temperature-sensitive smc1–2 allele at the endogenous locus [17]. Under uninducing or inducing conditions, the temperature sensitive growth of the smc1–2 strain is complemented by expression of wild-type SMC1, but not the insertion alleles (Figure S1B), indicating that the products of the insertion alleles are defective for Smc1 function under all conditions. The smc1–2 strains were then used to test the ability of the smc1 insertion alleles to generate cohesion. All smc1 insertion alleles are dramatically impaired for the establishment and maintenance of sister chromatid cohesion (unpublished data) (Figure 1B–1D).
All 13 insertions mapped to the SMC1 ORF, and 11 of them were unique (Figure 2A and 2B). A total of five insertions map to regions of known functional importance: the Walker A motif (35W), Signature motif (1129S), globular Hinge region (657H), and C terminus of Smc1 (1192C and 1215C) [7,21,22]. The remaining six insertions are either in 209L1–1 and 209L1–2 or cluster around Loop1 (189L1, 191L1–1, 191L1–2, and 235L1). Loop1 was previously defined through bioinformatics as one of three small regions within the α-helical domains of Smc molecules that are predicted to disrupt coiled-coil formation [38]. Since the insertions within and around Loop1 exhibit similar phenotypes (see below), these insertions define a new functional region of Smc1, which we call the Loop1 region.
We wanted to determine the efficacy of RID mutagenesis to define regions of Smc1 important for cohesin assembly. For this purpose we analyzed the ability of the 11 RID smc1 mutants (expressed at physiological levels) to coimmunoprecipitate with Mcd1 and Smc3. By comparison with wild-type Smc1, four RID mutants are dramatically reduced in their ability to coimmunoprecipitate with Mcd1, while all can coimmunoprecipitate Smc3 (Figure 3A and 3B). The failure to identify insertions that block Smc3 association is not surprising, since mutants defective in Smc1/Smc3 dimerization do not interact with any other cohesin subunit [21,22,39] and, therefore, are not likely be dominant negative.
The four Smc1 insertions defective for Mcd1 binding lie within the Walker A motif (35W), the Signature motif (1129S), the HH helix (1192C), or the S15 β strand (1215C) (Figures 2B and 3C). These are four of the five motifs known to be required for Mcd1 binding. The Walker A and Signature motifs, through ATP binding, tether together the Smc1 and Smc3 head domains so that they can both associate with a single Mcd1 [21,22,39]. The helix HH and S15 β strand provide two of the three major contacts between Smc1 and Mcd1 (Figure 3C) [39]. These motifs are widely spaced within the polypeptide chain and constitute a target of only ∼2% of the total residues. The efficient identification of these small disperse motifs by RID validates it as an extremely efficient means to identify regions of proteins important for complex assembly.
In addition to being a powerful tool to dissect cohesin assembly, we anticipated that RID would also efficiently identify informative alleles for cohesin function. The remaining seven RID alleles in the Hinge and Loop1 regions encode mutant Smc1 proteins that assemble with Smc3 and Mcd1 (Figure 3B) as well as Scc3 (unpublished data). Since they assembled with all known cohesin subunits, they were candidates for alleles that blocked cohesin function. All previously published alleles of cohesin subunits block chromatin binding. Therefore, we tested whether cohesins with these RID smc1 mutants were competent for chromosome binding. Minichromosomes were generated that express wild-type Smc1, 657H, or 209L1–2 fused to a 3XHA epitope, again under control of the GAL1 promoter. These alleles will henceforth be referred to as Smc1-HA, 657H-HA, and 209L1–2-HA, respectively. Cultures of smc1–2 cells containing these minichromosomes were released from G1 under conditions that inactivate the smc1–2 protein and induce expression of the galactose-regulated allele. These cells were then arrested prior to anaphase. Nuclei from these cells were spread on slides and processed for immunofluorescence to detect the chromosome association of different Smc1 proteins (Figure 4A). Like wild-type Smc1-HA, the insertion alleles 657H-HA and 209L1–2-HA associate with chromosomes.
We also monitored the chromosome localization of epitope-tagged Mcd1, a cohesin subunit, and Pds5, a cohesin accessory factor whose chromatin binding is mediated by cohesin (Figure 4B) [29,30]. Epitope-tagged Mcd1 and Pds5 localize to chromosomes in cells expressing the wild-type 657H or 209L1–2 allele, but not in cells with only inactivated smc1–2 (empty vector). The fact that the 657H and 209L1–2 alleles mediate Mcd1 and Pds5 localization to chromosomes suggests that these smc1 alleles bind chromosomes as part of the cohesin complex.
While it is clear that upon induction, the Hinge and Loop1 mutant cohesins can associate with chromosomes, their levels of chromosomal staining are reduced compared to induced wild-type cohesin (Figure 4C). Thus, the first question we wanted to ask was whether this reduction was sufficient to explain the cohesion defect of the Hinge and Loop1 mutants. To assess the level of cohesin binding to chromosomes, cells expressing Smc1, 657H, or 209L1–2 were arrested in mitosis. Nuclear spreads were prepared, and the levels of Smc1 and Pds5 immunostaining on chromosomes were quantified (Materials and Methods). The levels of chromosome binding for induced 657H-HA and its associated Pds5 are nearly identical to the binding of Smc1-HA and Pds5 under uninducing conditions (Figure 4C), a level of binding that is sufficient to generate cohesion and normal cell growth. Therefore, the level of chromosome association for 657H complexes should be sufficient to generate cohesion. The levels of 209L1–2-HA binding and its associated Pds5 are reduced by a maximum of 40%. This reduction may be below a threshold needed to generate cohesion. However, 50% of chromosome-bound cohesins in yeast meiosis and 90% in mammalian mitosis can be removed without eliminating cohesion [40,41]. Therefore, the level of chromosome association for 209L1–2-HA complexes is also likely to be sufficient to mediate at least partial sister chromatid cohesion, yet this seems not to be the case (Figure 1C).
If the quantity of chromatin binding for the mutant complexes is sufficient to generate cohesion, then the quality of their chromatin binding must be defective. The mutant complexes may bind chromatin by a nonphysiological mechanism. Previous studies have shown that the binding of cohesin to chromatin requires Mcd1 and ATP. The ATP dependence is mediated through the Walker A, Walker B, and Signature motif. To test whether the mutant complexes also bound in an ATP-dependent manner, we constructed Hinge and Loop1 mutants that carried a second insertion allele in the Signature motif. While the protein products of these double mutants are stable (unpublished data), they failed to associate with the chromatin (Figure 4A), indicating that the mutant complexes bind by ATP-dependent mechanism. One method to address whether the mutant cohesin complexes require Mcd1 to bind chromosomes would be to inactivate the temperature sensitive mcd1–1 in strains harboring the smc1–1 and the smc1 loop or hinge mutants. Unfortunately, mcd1–1 and smc1–1 are synthetically lethal. As an alternative, we followed the chromosome association of the Hinge and Loop1 mutant complexes through a cell cycle. Cohesin binds chromosomes only after G1, when Mcd1 is expressed [14,15], and dissociates from chromosomes upon the onset of anaphase, when Mcd1 is cleaved [19]. Cohesin with 657H-HA or 209L1–2-HA is absent from chromosomes in G1, localizes to chromosomes in metaphase, and is lost by the following G1, a pattern identical to cohesin with Smc1-HA (Figure 4D), strongly suggesting that the mutant complexes, like wild-type complexes, require Mcd1. The fact that, like wild- type, the Hinge and Loop1 mutants bind chromatin in an ATP- and Mcd1-dependent manner suggests that they all bind chromatin by a similar mechanism.
Despite these similarities to wild-type chromatin binding, the mutant complexes could still fail to generate cohesion if their binding is too late in the S phase to establish cohesion or too unstable to maintain cohesion until M. To test if the mutant cohesins bind chromatin in a timely manner, cells were released synchronously from a G1 arrest and analyzed at different intervals for DNA content and chromosome association of wild-type and mutant cohesins (Figure 5A). The timing of chromosome association during S phase for 657H-HA or 209L1–2-HA cohesins is very similar to that seen for uninduced levels of Smc1-HA cohesin, which is sufficient to generate cohesion. Also, the amount of cohesin bound per nuclei during S phase increased with similar kinetics for Smc1-HA, 657H-HA, and 209L1–2-HA cohesins (unpublished data). Hence, cohesin containing 657H or 209L1–2 exhibits a normal timing of chromosome association.
To examine the stability of chromosome binding for cohesin, nuclei were prepared from mitotic cells expressing Smc1-HA, 657H-HA, or 209L1–2-HA and spread in the absence of fixative. Spread nuclei were incubated in ∼50 ml buffer with varying amounts of KCl for 30 min. After incubation, fixative was added, and the association of cohesin with chromosomes was examined by immunofluorescence (Figure 5B). No change in chromosome binding is observed for Smc1-HA in the presence of 150 mM KCl, corroborating previous biochemical analyses that infer stable association between cohesin and chromosomes [32]. Similarly, no significant change in chromosome binding is observed for 657H-HA. Both wild-type and mutant cohesins are extracted completely from spread chromosomes at 250 mM KCl (Figure 5C, unpublished data). Similar results were obtained with the 209L1–2-HA mutant (unpublished data). Therefore, cohesin complexes with wild-type, 657H-HA, or 209L1–2-HA appear to exhibit the same stability of chromosome binding under these conditions. Together, our analyses of the Hinge and Loop1 mutants reveal a new type of cohesin complex. Like wild-type, these mutant complexes bind stably to chromatin, bind chromatin with proper cell cycle timing, and require ATP and Mcd1. However unlike wild-type, these mutant complexes fail to generate cohesion. These mutants also suggest that binding, per se, prior to DNA replication is not sufficient to generate cohesion.
Since we could not explain the cohesion defect of these mutants by changes in their general chromosome-binding properties, we tested the specificity of their chromatin binding by examining their enrichment at CARs using chromatin immunoprecipitation (ChIP) [23–25]. Cohesin enrichment was analyzed initially at the centromere of Chromosome III and CARC1, a CAR approximately 11 kb from the centromere (Figure 6A and 6B). Under uninduced levels of Smc1 expression, Mcd1-6HA is enriched at CARC1 and around CEN3 as reported previously [24,25]. Conversely, Mcd1-6HA enrichment at CARC1 and CEN3 is eliminated for cohesin complexes containing 657H or 209L1–2. Similar results were obtained for CARL1 on Chromosome XII (unpublished data). Therefore, cohesin complexes with 657H or 209L1–2 fail to be enriched at CARs.
Because these mutant complexes do bind to chromatin as assayed by chromosome spreads, they apparently are bound to sites other than CARs. One possibility is that the mutant cohesins bind to chromatin through the Scc2/Scc4 loading factor, but are trapped in a nonproductive preloading complex. To test this, we examined cohesin binding at Scc2/Scc4 chromatin-binding sites [42], but we observed no enrichment of the mutant cohesins at these sites (Figure S2). The failure to observe enrichment of the mutant cohesins in our ChIP experiments does not exclude the possibility that they bind randomly within these regions. The enrichment of wild-type cohesin at CARs is only 10-fold above background, and CARs are spread at approximately every 10 kb, hence the dispersal of this signal to random sites would dilute the signal to background levels. While the position of the ectopic binding remains to be elucidated, the fact that the Hinge and Loop1 mutants cause ectopic binding indicates that Smc1 plays an active role in the specific localization of cohesins to CARs and pericentric regions. Furthermore, since this mislocalization is the only severe cohesin defect we have been able to observe in the Hinge and Loop1 mutants, it suggests that cohesin localization to CARs is critical for cohesion.
Our studies show that the Hinge and Loop1 regions of Smc1 are required for the establishment of cohesion and the enrichment of cohesin at CARs and pericentric regions. The accessory protein Ctf7/Eco1 is also required for the establishment of cohesion [16,27]. In a ctf7 mutant, cohesins pellet with chromatin [16]. This technique does not distinguish between ectopic and specific chromatin binding. Indeed, if ctf7, like Hinge and Loop1 mutants, showed ectopic binding, then this would suggest that Ctf7 interacts with the Hinge and Loop1 regions of Smc1 to ensure specific binding to CARs.
To test the function of Ctf7 in cohesin enrichment at CARs, we compared the localization of cohesin on chromatin in wild-type and ctf7 temperature-sensitive strains. Wild-type and ctf7 mutant cells expressing Mcd1-HA as the sole source of Mcd1 function were synchronized at the permissive temperature of 23 °C in G1. These cells were released to the the nonpermissive temperature of 37 °C in media containing hydroxyurea (HU) to inactivate mutant ctf7 protein and block DNA replication [27], respectively. Cohesin association to chromatin was monitored by ChIP (Figure 7A and 7B). In wild-type cells arrested in HU, cohesin is enriched at CARs and pericentric regions as described previously [24,25]. In ctf7 strains arrested in HU at the nonpermissive temperature, cohesin is again enriched at CARs and pericentric regions. Thus Ctf7 is not needed to localize cohesins to CARs or pericentric regions.
The above results indicate that Ctf7 function is distinct from that of the Hinge and the Loop1 regions of Smc1. Previous experiments showed that Ctf7 function is required during S phase; however, this activity was not ordered relative to cohesin loading at CAR sites. To test whether Ctf7 is required coincident with or after cohesin binding to CARs, wild-type and ctf7 mutant cells were allowed to progress from G1 to early S (HU arrest) at the nonpermissive temperature and then shifted to the permissive temperature for the remainder of the cell cycle. The results show that cell viability was 93% ± 4% for wild-type cells and 88.2% ± 5% for ctf7 mutant cells, indicating that Ctf7 function is required after cohesin loads to CARs. This result is strongly supported by cell cycle mapping studies that show inactivation of ctf7 from early S (HU arrest) and prior to the end of DNA replication leads to cell death [16,27,43], indicating that Ctf7 performs its essential function during this window of the cell cycle. Thus, Ctf7 appears to function in the establishment of cohesion after the Smc1-, Loop1-, and Hinge-dependent localization of cohesin to CARs and pericentric regions, and during S phase.
Our results validate RID as a very efficient strategy to identify Smc1 alleles that provide important structural and functional information about Smc1 and cohesin. First, of the 11 smc1 RID alleles, ∼40% blocked Smc1 association with Mcd1 but not Smc3. Thus, RID efficiently identifies alleles that trap partially assembled cohesin. Furthermore, these Smc1 insertions lie within four of the five small structural elements (each element is approximately ten residues each spread throughout a total of 1,300), which are required for Mcd1 binding. The efficacy and remarkable precision of RID suggests that it can be used effectively on less well-characterized proteins for the de novo identification of candidate regions of that protein that mediate its binding to interacting partners. Second, the remaining seven RID alleles of Smc1 identify a new functional domain (the Loop1 region), and a new function for the Hinge beyond its established function in dimerization. These Hinge and Loop1 alleles also generated cohesin complexes that unlike any previous cohesin mutants retain the ability to bind chromatin. The isolation of these unusual alleles underscores the combined power of using random mutagenesis, which allows one to avoid the inherent biases of directed mutagenesis, and the imposition of the dominant negative phenotype, which allows rapid identification of rare partially functional alleles in a sea of common null mutations caused by misfolding and truncations. Given the success of the RID strategy for Smc1/cohesin, RID should be a useful tool to dissect the structure and function of many multi-subunit complexes.
Our study of RID smc1 alleles has provided important new insights into cohesin binding to chromatin. First, our results show that in vivo the Hinge and Loop1 regions are needed for binding to CARs, two domains not implicated in chromatin binding by either the embrace or snap model. The joint requirement for these two domains is even more surprising given their apparent physical separation (∼40 nm) based upon electron micrographs of cohesin [13]. Interestingly, in vitro analyses of bacterial Smc complexes have implicated the hinge domain in single-stranded and double-stranded DNA binding [37,44] and DNA-dependent stimulation of the ATPase activity of the head [44]. Thus, our in vivo study combined with these in vitro studies support DNA-dependent functional interactions between the opposite ends of Smc complexes.
The second interesting feature of cohesins containing the RID smc1 mutants is that they appear to bind chromatin ectopically. This conclusion is based upon the observations that cohesins exhibit general chromatin binding as assayed by chromatin spreads but are no longer enriched at CARs. It is possible that the cohesin binding is compromised at CARs in some way that they bind there but subsequently dissociate. Consistent with this, the level of binding of overexpressed RID mutant proteins is reduced compared to overexpressed wild-type protein. However, under our assay conditions using chromatin spreads, the Hinge, Loop1, and wild-type cohesins appear to be bound with similar stability. Furthermore, the level of chromatin binding of cohesin, when the Hinge mutant is overexpressed, is still comparable to the level of binding for uninduced wild-type, which is sufficient for cohesion function. Alternatively one could argue that the ectopic binding observed by chromatin spreads reflects some artifact of this assay. This is extremely unlikely, as chromatin binding observed by spreads for the RID mutants share many of the chromatin-binding features of wild-type cohesin, including Mcd1 dependence, ATP dependence, and proper cell cycle loading in S and unloading in M. Together these results suggest that cohesins are capable of general chromatin binding, mediated by Mcd1 binding and ATP functions, but are then targeted to CARs through an additional function(s) mediated by the Hinge and Loop1 regions.
In one scenario the Loop1 and Hinge regions may target cohesins to CARs by binding targeting factors that recognize specific chromatin features; the RID mutants perturb the binding of the targeting factors. There are precedents for this idea, since the Hinge and Loop1 regions mediate interactions with non-SMC factors in other Smc-related proteins [45–47]. In addition, histone modifications have already been demonstrated to be critical to target cohesin binding to heterochromatin and double strand breaks [48–50]. In this model, cohesin binding to chromatin resembles RNA polymerase; both are targeted to specific sites through specificity factors that recognize changes in chromatin, and in their absence load with lower efficiency at abundant cryptic sites.
As an alternative, cohesin itself may be capable of recognizing specific features of chromatin. Indeed the MutS mismatch repair complex, which shares similarities to Smc complexes, forms a ring that topologically traps generic DNA, while residues within the ring specifically recognize mismatch DNA [51–53]. In this scenario, the Smc1, Hinge, and Loop1 mutant subunits allow a topological interaction of cohesin with chromatin, but perturb its ability to make intimate chromatin/DNA interactions. Interestingly, in vitro studies have implicated residues on the interior of the bacterial Smc ring for DNA binding and DNA-dependent stimulation of the ATPase [44].
The chromatin-binding properties of cohesin in ctf7 and RID mutants also provide important insights into the mechanism of sister chromatid cohesion. First, the binding of cohesin to chromatin in early S at CARs (ctf7 mutants) or ectopic sites (Hinge and Loop1 mutants) is not sufficient to generate cohesion. Second, cohesin binding to CARs and centromeres appears necessary for cohesion. Third, an auxiliary factor, Ctf7 enables cohesins already bound at CARs to generate functional cohesion. These observations all contradict a one-step mechanism for cohesion like the simple embrace model, which requires only chromatin binding of cohesin anywhere on the chromosomes followed by DNA replication.
Rather, our results support tethering of sister chromatids by two chromatin-binding events, which require specific binding of cohesins at CARs followed by a Ctf7-dependent step. Given these new constraints, we propose the following working model for the establishment of cohesion: cohesins bind to CAR sites as they emerge from the replication fork and are subsequently activated by Ctf7 to initiate the capture of the homologous CAR on the sister chromatid. This capture could occur by activating a second chromatid binding event by a single complex (for example, a second embrace) or by activating oligomerization of cohesins bound to each CAR (the oligomerization models) [33–36]. In either version of this two capture model, the specific binding to CARs and Ctf7 steps would be critical. Because of local proximity during replication, a cohesin molecule bound to a CAR on one sister chromatid will find the sister CAR and/or cohesin bound to that site before finding other sites/cohesins in the genome. We speculate further that the activation by Ctf7 is programmed to be local/transient. As a result, cohesin bound to a random site on a chromatid will be unlikely to remain active long enough to find a random CAR or another randomly bound cohesin. Thus, the targeting of cohesin by CAR binding and its local activation by Ctf7 would provide spatial and temporal regulation of cohesin to activate the second capture and ensure that cohesin generates a crosslink only between only sister chromatids and not random chromatin. Finally, the specificity of tethering by Smc complexes in other DNA processes may be achieved through specific chromatin binding coupled with Ctf7-like activators.
Yeast strains were grown in YEP, SC-URA, or SC-URA-TRP media [54] supplemented with 2% dextrose (D), 2% raffinose 2% galactose (RG), 3% glycerol 2% lactic acid (LA), or 3% glycerol 2% lactic acid 2% galactose (LAG), as indicated. Glucose, raffinose, and galactose were purchased from Sigma-Aldrich (http://www.sigmaaldrich.com), glycerol from EMD Biosciences (http://www.emdbiosciences.com), and lactic acid (40% v/v stock, [pH 5.7]) from Fisher Scientific (http://www.fishersci.com). A PCR-based strategy was used to generate a complete genomic replacement of SMC1 with the Schizosaccharomyces pombe His5+ gene [55], creating the yeast strain Ymm1. Ymm1 is dependent on the plasmid pMM26 for viability. The genotypes of strains used in this work can be found in Table S1. Yeast transformation and genetic methods were as described previously [56].
The plasmids described were generated through use of the Echo Cloning System (Invitrogen, http://www.invitrogen.com), which results in the expression of genes fused to COOH-terminal V5 epitopes under control of the GAL1 promoter. In all cases, the V5 epitope was replaced by a TAP tag using a PCR-based tagging strategy [57,58]. The plasmid pMM14 contains the SMC1 ORF. Restriction digestion by SmaI and PmeI, followed by religation, destroyed a unique PmeI site and generated AMH4.
The plasmid pMM26 was derived from pMM14. GAL1 was replaced with a 412-basepair fragment immediately 5′ of the SMC1 ORF using AgeI and XhoI restriction sites, creating pMM24. Loss of the URA3 gene by BsgI digestion, followed by T4 blunting and religation, generated pMM26. Expression of Smc1-TAP from pMM24 and pMM26 is capable of rescuing growth of a temperature sensitive smc1–2 strain (1360–7C) and a strain deleted for endogenous SMC1. The plasmid pMM25-3HA expresses Smc1 fused to a COOH-terminal TAP-3HA tag under control of the GAL1 promoter. Through a PCR-based strategy, a BamHI site within the TAP coding sequence of pMM14 was destroyed by a silent mutation and replaced by a BamHI site immediately 5′ of the TAP stop codon [57]. TRP1 was lost by SmaI digestion, followed by religation, resulting in pMM14-PIBS. The XbaI/SmaI fragment of pMM14-PIBS was subcloned into pRS305, resulting in pMM27. A 3XHA coding sequence was inserted in frame at the BamHI site of pMM27, resulting in pMM27-3HA. Replacing the XbaI/SmaI fragment of pMM25 with the XbaI/SmaI fragment of pMM27-3HA resulted in pMM25-3HA. The plasmid pMM25-3HA is able to rescue viability of an SMC1 deleted strain under either uninduced or induced expression conditions. The plasmids 209L1–2-HA and 657H-HA were generated by replacing the XhoI/XbaI fragment of pMM25-3HA with those from 209L1–2 and 657H.
Insertion mutagenesis of AMH4 was performed using the GPS-Linker Scanning system (New England Biolabs, http://www.neb.com). In vitro mutagenized AMH4 was transformed into Escherichia coli. Approximately 3,100 colonies were scraped from plates, and plasmid DNA was isolated using a Plasmid Maxi kit (Qiagen, http://www.qiagen.com). The majority of plasmids received only a single insertion (unpublished data). Plasmid DNA in the primary library was linearized by PmeI digestion, gel purified, religated, and transformed into E. coli. Plasmid DNA from this secondary library was isolated from approximately 4,000 colonies from plates using the Plasmid Maxi kit.
The secondary library was transformed into YMM1/pMM26, with transformants selected for on SC-URA-TRP-D plates at 23 °C. Individual transformants were picked and patched to new SC-URA-TRP-D plates and grown at 23 °C. Patches were replica plated to an identical plate and a SC-URA-TRP-RG plate and grown for 3–5 d at 23 °C. From patches that were impaired for growth on SC-URA-TRP-RG plates, plasmids were isolated from the identical patch growing on the SC-URA-TRP-D plate and retested. For plasmids that retested, the insertion mutations were mapped by restriction endonuclease digestion and sequenced. To isolate functional alleles, the secondary library was transformed into an smc1–2 strain, 1360–7C. Transformants were selected for on SC-URA-D plates at 23 °C, patched to new SC-URA-D plates, and grown at 23 °C. Patches were replica plated to an SC-URA-RG plate and grown at 37 °C. Plasmids were isolated from patches that grew at 37 °C and retested. For plasmids that retested, the insertion mutations were mapped by restriction endonuclease digestion and sequenced.
Exponentially dividing cell cultures were initially grown in SC-URA-LA at 23 °C. α-Factor (1.5 × 10−8 M [Sigma]) was added to cultures in mid-log phase (approximately 0.5 × 107 cell/ml). To simultaneously inactivate smc1–2 and release them from G1 arrest, cells were washed twice in either 37 °C YEP-LA or 37 °C YEP-LAG containing 0.1 mg/ml Pronase (Sigma). Cells were resuspended in either 37 °C YEP-LA or YEP-LAG and grown at 37 °C. To arrest G1 released cells in metaphase, nocodazole was added to a final 15 μg/ml (1.5 mg/ml in DMSO stock [Sigma]) and cultures were grown for 3 h. To arrest in early S phase, G1 released cells were grown in the presence of 0.2 M hydroxyurea (HU) (Sigma) for 3 h. Cell cycle arrest was assessed by flow cytometry and cell morphology [27].
Cells were processed to visualize GFP by microscopy or to measure DNA content by flow cytometry as described previously [27,59]. Protein extracts for coimmunoprecipitation and immunoblotting were prepared as described previously [15,20] from cultures grown in YEP-D media. ChIP was performed as described [60]. Information about the primers used in this study is available upon request. PCR and data analysis for ChIP was performed as described [49]. All experiments were done at least twice and a representative dataset is shown.
Chromosome spreads and indirect immunofluorescence of spread nuclei were performed as described previously except spheroplasting was done at 25 °C for 30 min [41]. To assess the level of cohesin association per chromosome, the average pixel intensity for each chromosome mass was determined. The background pixel intensity for each slide was determined by measuring the average pixel intensity for areas similar in size to spread nuclei. Subtracting the background intensity from each chromosome mass gave relative pixel intensity. At least 100 nuclei were analyzed per slide to generate an average relative pixel intensity per chromosome mass. To assay the stability of cohesin association with chromosomes, nuclei were spread on multiple slides in the absence of fixative with 0.25% Triton X-100 in PHEM buffer (60 mM Pipes, 25 mM Hepes, [pH 6.95], 10 mM EGTA, and 4 mM MgCl2) and incubated for 10 min. Each slide was then placed in a single coplin jar containing ∼50 ml 0.25% Triton X-100 in PHEM buffer with varying amounts of KCl for 30 min, while gently shaking. Following this incubation, nuclei were fixed by 4% PFA with 0.25% Triton X-100 in PHEM buffer, as before. Immunofluorescence was performed as above. |
10.1371/journal.pgen.1002073 | The stb Operon Balances the Requirements for Vegetative Stability and Conjugative Transfer of Plasmid R388 | The conjugative plasmid R388 and a number of other plasmids carry an operon, stbABC, adjacent to the origin of conjugative transfer. We investigated the role of the stbA, stbB, and stbC genes. Deletion of stbA affected both conjugation and stability. It led to a 50-fold increase in R388 transfer frequency, as well as to high plasmid loss. In contrast, deletion of stbB abolished conjugation but provoked no change in plasmid stability. Deletion of stbC showed no effect, neither in conjugation nor in stability. Deletion of the entire stb operon had no effect on conjugation, which remained as in the wild-type plasmid, but led to a plasmid loss phenotype similar to that of the R388ΔstbA mutant. We concluded that StbA is required for plasmid stability and that StbA and StbB control conjugation. We next observed the intracellular positioning of R388 DNA molecules and showed that they localize as discrete foci evenly distributed in live Escherichia coli cells. Plasmid instability of the R388ΔΔstbA mutant correlated with aberrant localization of the plasmid DNA molecules as clusters, either at one cell pole, at both poles, or at the cell center. In contrast, plasmid molecules in the R388ΔΔstbB mutant were mostly excluded from the cell poles. Thus, results indicate that defects in both plasmid maintenance and transfer are a consequence of variations in the intracellular positioning of plasmid DNA. We propose that StbA and StbB constitute an atypical plasmid stabilization system that reconciles two modes of plasmid R388 physiology: a maintenance mode (replication and segregation) and a propagation mode (conjugation). The consequences of this novel concept in plasmid physiology will be discussed.
| The ability of bacteria to evolve and adapt to new environments most often results from the acquisition of new genes by horizontal transfer. Plasmids have a preponderant role in gene exchanges through their ability to transfer DNA by conjugation, a process that transports DNA between bacteria. Besides, plasmids are autonomous DNA molecules that are faithfully transmitted to cell progeny during vegetative cell multiplication. In this study, we report a system composed of two proteins, StbA and StbB, which act to balance plasmid R388 physiology between two modes: a maintenance mode (vertical transmission) and a propagation mode (horizontal transmission). We demonstrate that StbA is essential to ensure faithful assortment of plasmid copies to daughter cells. In turn, StbB is required for plasmid R388 adequate localization for conjugation. This is the first report of a system which reconciles plasmid segregation and conjugation. Furthermore, R388 belongs to the IncW family of conjugative plasmids, which are of particular interest due to their exceptionally broad host range. We show that the StbAB system is conserved among a wide variety of conjugative plasmids, mainly broad host range plasmids. Thus, the Stb system could constitute an interesting therapeutic target to prevent the spread of adaptive genes.
| Transmissible plasmids contribute greatly to the plasticity of bacterial genomes and to the acquisition of genetic traits by host cells through the collective carriage of adaptive genes, including antibiotic resistance and virulence genes, and through the ability to disseminate them by conjugation [1]. Horizontal gene transfer may thus increase the adaptability of bacteria to changing environmental conditions, which is dramatically exemplified by the emergence and spread of multiple antibiotic-resistance plasmids in and between potentially pathogenic bacteria. Conjugative plasmids are transmitted both vertically to daughter cells and horizontally to other strains or species. Vertical transmission requires timely controlled replication and faithful assortment (segregation) of sister plasmid copies to daughter cells. Segregation occurs by a range of different mechanisms including control of copy number, resolution of multimeric plasmid molecules and, in the case of most low copy number plasmids, active segregation (partition). Partition (Par) systems ensure efficient distribution of plasmid molecules to each daughter cell during division (for reviews: [2]–[5]). They are composed of a cis-acting centromere-like site and two proteins, a nucleotide-binding cytomotive protein and a centromere-binding adaptor protein. Stable inheritance requires that these proteins form a partition complex on the centromere. The par centromere locus and the Par proteins are encoded by sets of homologous genes in various plasmids, phages, and chromosomes.
The mechanism of plasmid conjugation in gram negative bacteria has been well characterized (for a review: [6]). The overall process is accomplished by two functional multiprotein complexes encoded by two gene clusters: the set of mobility genes (MOB), involved in conjugative DNA processing, and the mating pair formation cluster (MPF), encoding the nucleoprotein transport apparatus. These two systems are connected by the coupling protein (T4CP). The MPF is a type IV protein secretion system and implies the assembly of a multiprotein complex at the bacterial membrane [7]. Conjugal DNA processing involves the formation of a nucleoprotein complex called relaxosome at the cognate origin of conjugative transfer (oriT). The relaxase catalyses a DNA strand- and site- specific cleavage at the nic site of oriT, and remains covalently attached to the 5′ end of the cleaved single-stranded (ss) DNA [6]. Complementary-strand synthesis is initiated from the free 3′ end of the cleaved strand through rolling-circle replication. The relaxase-ssDNA complex interacts with the T4CP, which guides the transferred strand through the DNA transfer apparatus formed by the MPF proteins throughout the membrane and into the recipient cell [8], [9]. Fundamental questions remain unanswered concerning the spatiotemporal coordination of the DNA substrate processing, its recruitment to conjugative pores, and the subsequent DNA translocation reactions. More relevant to this work, how conjugation is integrated with plasmid maintenance functions such as replication and segregation is still poorly understood.
The low-copy number conjugative plasmid R388 is the prototype of the IncW family. It represents the smallest broad host range conjugative plasmid [10], [11]. Its mechanism of segregation is presently unknown, and no par system was annotated in its DNA sequence [12]. R388 carries a cluster of two operons, which are transcribed divergently from the region containing oriT. One contains the MOB genes, and the other includes a cluster of three genes, stbA, stbB and stbC, whose functions have not been analyzed previously. Their relative position to oriT makes these three genes the first to enter the recipient cell during plasmid conjugation. Here, we investigated the role of the stbABC operon in plasmid R388 transfer and stability in E. coli. By using a fluorescent protein to tag plasmid molecules, we found that R388 plasmid foci, most of which contain a single copy of the plasmid, are evenly distributed within E. coli cells. In contrast, a derivative of R388 lacking stbA mislocalized as clusters at the cell poles or at the cell center, and correlated with plasmid instability. In addition, we show that StbB, a putative ATPase protein, is strictly required for R388 conjugative transfer and the occurrence of plasmid foci close to the membrane cell poles. Taken together, these results suggest that stbA and stbB constitute a balancing system that integrates plasmid conjugation with the functions that ensure the efficient vertical transmission of the plasmid.
To study the role of the three stb genes in plasmid R388, we first analyzed the stb operon by protein sequence comparison. We found that the most conserved protein is StbB, whereas StbA is poorly conserved. Besides, StbC is an orphan protein, without significant homologs in any other system. We thus used R388 stbB gene as template to search for homology. StbB homologs were usually included in operons of three genes at the leading region of conjugative plasmids of MOBF11, MOBP11, MOBP6 and of mobilizable plasmids MOBP13/P14, belonging to several Inc groups (IncW, IncN, IncP-1, IncP-9, IncQ and IncI-2). Synteny conservation of the stb and MOB regions of representative plasmid groups is shown in Figure 1B. There was a neat bias for the presence of stbB-like genes in conjugative plasmids that carried an MPFT T4SS. In addition, an Stb-system was also found in some groups of mobilizable plasmids. In all cases, the stb genes were located at the 3′side of the nicked strand and are thus the first to enter the recipient cell during plasmid conjugation.
Alignment of StbB and homologs from each MOB group showed that they shared a deviant Walker A nucleotide triphosphate-binding motif (Figure S1A), also found in the ParA/Soj/MinD superfamily of ATPases [13]. Members of this superfamily include ParA, required for accurate chromosomal and plasmid DNA partitioning, MinD required for correct placement of the septa during cell division, and Soj, which plays a role in chromosome compaction required for nucleoid partition (reviewed in [3], [14], [15]). parA and soj genes are found in their respective operons adjacent to a second gene, parB and spo0J respectively, which encodes a DNA-binding protein. However, no homology of StbA to ParB-like partition proteins was detected. On the other hand, R388 StbA showed significant homology to the TraD protein of plasmid NAH7. Iterative PSI-BLAST also returned similarly located proteins in MOBP11, MOBP6 and MOBP13/P14 plasmids (including StbA_R46, TraK_RP4, MobC_pTC-FC2, and YciA_R721) after seven iterations. Alignment of R388 StbA and several similarly located proteins is shown in Figure S2.
According to the sequence of their StbB-like proteins, plasmids can be phylogenetically assorted in two large groups. The first group includes MOBP11, MOBP6 and MOBP13/14 plasmids, that encode StbB-like proteins (including TraL_RP4 or MobD_pTF-FC2) containing the classical motif KGGXXK[T/S] found in other ParA/Soj/MinD ATPases [13]. R388 plasmid belongs to the second group together with other MOBF11 and MOBP6 plasmids. These encode StbB-like proteins that contain the slightly divergent motif SGXXGK[T/S]. MOBP6 plasmids are pervasive in both groups, perhaps suggesting that they were the first in which the Stb-system was installed. Modeling the structure of StbB using the 3D structure of Soj (PDB ID: 2BEK) as template predicts that the dimerization role of the signature lysine K15 in Soj, [16] could be performed by other polar residue like serine S9 in R388-StbB (Figure S1B).
To investigate whether the stb operon was involved in R388 transfer, as suggested by its positional conservation relative to the MOB region, we first generated a derivative of R388 deleted of the entire stb operon, R388ΔstbABC. The method of Datsenko and Wanner [17] was used to replace the stb region by a DNA fragment conferring resistance to kanamycin. Mating experiments were carried out under appropriate conditions to avoid indirect effects due to plasmid instability (see below and Materials and Methods). As shown in Figure 2A, deletion of stb did not result in any noticeable effect on transfer frequencies compared to the wild-type plasmid R388 (R388). These results, which are consistent with previous data [18], could in principle suggest that the stb operon was not involved in R388 transfer. We nevertheless examined the effect of deleting each of the three genes of the stb operon independently. We constructed three R388 derivatives, R388ΔstbA, R388ΔstbB, and R388ΔstbC, which lack stbA, stbB, and stbC, respectively, and measured their transfer frequencies (Figure 2A). Surprisingly, R388ΔstbA was transferred at a frequency approximately 50-fold higher than R388, suggesting that StbA inhibits R388 conjugative transfer. In contrast, deletion of stbB resulted in a complete block of conjugation (transfer frequency <10E-9, Figure 2A), indicating that StbB is required for R388 conjugation. The transfer frequency of R388ΔstbC was comparable to that of R388, which indicated that StbC had no significant role in R388 conjugative transfer. Since deletion of the entire operon did not modify the conjugation frequencies, we concluded that StbB is required for conjugation only in the presence of StbA, which in turn, inhibits conjugation. StbA and StbB thus appear to have antagonistic effects in conjugation.
To further examine the interactions between the different functions of stb genes, we carried out a complementation analysis. We constructed plasmids carrying either the stbA or stbB genes, driven by the Plac promoter and controlled by a lacI q gene present on the same plasmid (pStbA and pStbB, respectively, Table S1). pStbA and pStbB were introduced in donor cells by transformation, and mating experiments were carried out. Results are shown in Figure 2B. Supplying StbA in trans in donor cells harboring plasmid R388ΔstbA led to a 100-fold reduction in conjugation, which corresponded to a frequency comparable to that of R388. This result also indicated that the stbB gene was adequately expressed in plasmid R388ΔstbA (that is, the stbA deletion did not cause a polar effect). Supplying StbB in trans in donor cells containing plasmid R388ΔstbB led to restoration of the transfer frequencies to wt level.
Providing StbA in trans in donor cells harboring plasmid R388ΔstbABC abolished transfer (Figure 2B), further demonstrating that StbB is required for conjugative transfer only when StbA is present, and that in turn StbA inhibits conjugation in the absence of StbB. Besides, supplying StbB in trans to plasmid R388ΔstbABC resulted in an increase of conjugation frequency of approximately 4-fold. This indicated that StbB stimulates conjugative transfer in the absence of StbA. We next introduced either pStbA or pStbB in donor cells containing plasmid R388 to analyse the effects of overexpressing the corresponding stb gene. Supplying StbA in trans had no effect, while supplying StbB in trans led to enhanced transfer frequencies to a level similar to that of plasmid R388ΔstbA (Figure 2B). This indicated that StbB stimulates conjugation also when StbA is present.
Taken together, the results shown in Figure 2 demonstrate that StbA and StbB, but not StbC, are involved in R388 conjugation in E. coli, and that their activities are functionally connected. StbB was strictly required for conjugation only when StbA was present. Besides, StbA prevented R388 conjugative transfer, whereas StbB stimulated R388 conjugative transfer, suggesting that StbA and StbB have balancing/compensatory effect to control conjugation.
Previous studies showed that the stb genes of the IncN plasmid R46 are required for stable plasmid inheritance in recombination-proficient but not in recA strains [19]. To examine whether R388 stb operon plays a role in plasmid R388 inheritance, plasmid R388ΔstbABC was subjected to stability studies in LN2666 (recA+) and in FC232 (LN2666 recA) strains. As we obtained similar results in both genetic backgrounds, only the results with LN2666 strain are presented in Figure 3. E.coli cells carrying R388ΔstbABC were grown in serial cultures in nonselective medium, and plasmid loss rates were measured by plating out every 20 generations to determine the proportion of cells retaining the plasmid (Materials and Methods).
Plasmid R388 was stably maintained in progeny cells and 100% of the cells retained the plasmid after 80 generations (Figure 3A). Deletion of the entire stb operon led to a significant stability defect with a rate of loss of 5% per generation. We next examined the effects of deletion of each stb gene on plasmid stability. R388ΔstbA showed a plasmid loss rate similar to that of R388ΔstbABC (5.1% loss per generation). In contrast, R388ΔstbB and R388ΔstbC loss rates were close to zero (10−2% and 8.10−3% per generation, respectively; Figure 3A). The stability defect of R388ΔstbABC was thus fundamentally due to the absence of stbA.
Complementation of R388ΔstbABC and R388ΔstbA with the StbA-producing plasmid pStbA decreased their loss frequency to 0.07% and 0.08%, respectively (Figure 3A). This result confirmed that StbA is required for the stable inheritance of R388 and showed that it acts in trans. We then addressed the question of whether the instability of plasmids R388ΔstbABC and R388ΔstbA was due to a decrease in their plasmid copy number per cell when compared to R388. However, the average number of copies per chromosome of both plasmids, as determined by real-time qPCR (Materials and Methods), were found to be similar to that of R388 (R388, 3.8±1.0, R388ΔstbABC, 4.5±1.2 and R388ΔstbA, 4.1±0.9). This result strongly suggested that the instability of R388ΔstbA and R388ΔstbABC was due to a defect in plasmid segregation.
The above results indicated that StbA is involved both in the stability and in conjugative transfer of plasmid R388, raising the question of how StbA interacts with R388 to perform such functions. The StbA homolog TraK_RP4 was shown to bind specifically to a DNA sequence containing the region between the nic site and the traK gene [20]. We thus explored the role of the upstream region of stb, which contains two sets of five direct repeats of a DNA consensus sequence 5′ TTGCATCAT (the stbDRs, Figure 1).
StbA protein was purified on a Ni-agarose resin as a StbA-His6-tagged protein (Materials and Methods). Incubation of a DNA fragment containing a complete set of stbDRs with increasing quantities of StbA and an excess of nonspecific competitor DNA gave rise to retarded species (Figure S3). Thus, StbA bound specifically stbDRs containing DNA in vitro.
To examine the role of the stbDRs in vivo, we constructed a derivative of plasmid R388ΔstbABC carrying a larger deletion which included the stbDRs (R388ΔstbDRs-stbABC). While providing StbA in trans to plasmid R388ΔstbABC abolished transfer (Figure 2B), StbA had no effect on conjugation of R388Δ (stbDRs-stbABC) (Figure 2B), showing that the role of StbA in R388 conjugation is mediated by its binding to the stbDRs: if StbA does not bind the DRs, StbB is not required for conjugation. Besides, providing StbB in trans to R388ΔstbABC, as well as to R388Δ (stbDRs-stbABC) results in 4-fold increase in transfer frequencies (Figure 2B), suggesting that the role of StbB does not depend on the presence of the stbDRs.
We next examined the role of the stbDRs in plasmid stability. R388Δ (stbDRs-stbABC) showed the same instability phenotype as R388ΔstbABC (Figure 3B). Providing StbA in trans, led to an almost complete reduction of R388ΔstbABC plasmid instability (rate of loss of 0.07% per generation). Furthermore, providing StbA in trans had only a partial stabilization effect on the R388Δ (stbDRs-stbABC) plasmid (rate of loss of 2.8% per generation). We concluded that, as for conjugation, the role of StbA in R388 stability is fundamentally mediated by its binding to the stbDRs.
Analysis of R388 genome showed that the 9-bp motif specific of the stbDRs is found in four other promoters within the establishment region of R388 (Figure 1; [12]). To examine whether these potential StbA binding sites sequences could have a role in R388 stability, we generated a series of plasmids carrying several deletions of this region. We found that, provided that the stb operon was preserved, the deletion of a DNA fragment ranging from orf14 to kfrA did not affect plasmid stability at all (R388Δ (orf14-kfrA), rate of loss of 0%, Figure 1A, Table S1). This indicated that this region does not contain genes required for R388 stability and demonstrated that within the establishment region, only the stb genes are required for R388 stability. As expected, a plasmid lacking the entire establishment region (i.e. from stbA to kfrA gene, see Figure 1, R388Δ (stbABC-kfrA)) was not stably maintained (rate of loss of 5%, Figure 3B). Supplying StbA in trans led to the stabilization of such a plasmid carrying the stbDRs region (R388Δ (stbABC-kfrA), Figure 3B). As expected, this stabilisation effect was dependent on the stbDRs since the equivalent plasmid lacking the stbDRs, R388Δ (stbDRs-kfrA), remained highly unstable upon StbA production (3.7% of loss per generation). We concluded that the potential StbA binding sites located outside the stbDRs are not sufficient to support StbA-dependent plasmid stabilization.
To get a deeper insight into R388 segregation, we undertook a cellular localization study of plasmid R388 using the parS/GFP-ParB system [21]. The parS site of bacteriophage P1 was inserted into R388 plasmid DNA as a parS-chloramphenicol (parS-Cm) cassette. This allowed visualization of the cellular localization of the resulting plasmid R388parS (R388, Table S1) in live E. coli cells expressing the green fluorescent marker protein GFP-Δ30ParB from a second plasmid (pALA2705; [21]; Materials and Methods). We used two different R388parS plasmids (R388parS1 and R388parS2), in which the P1 parS site had been inserted in two different intergenic regions R388 and obtained comparable results (Materials and Methods; data not shown). Figure 4A shows representative fluorescence images of cells containing R388parS1 plasmid. Discrete foci could be visualized without IPTG induction, at the basal level of expression of the fluorescent GFP-Δ30ParB from the lac promoter [22]. Under these conditions, neither the parS insertions into R388, nor the expression of the GFP-Δ30ParB protein had a noticeable effect on R388parS plasmid stability and conjugative transfer (plasmid loss rate <0.01%; conjugative transfer frequency 2.10−1/donor).
In the conditions used, more than 98% of the cells analyzed contained GFP-foci, showing that the efficiency of focus detection was high. There were 4 to 10 foci per cell and the average number of foci per cell increased with cell size (data not shown). A majority of the cells (64%) had 4 to 6 foci, and about 34% had 7 to 10 foci (Figure 5, R388). The population average was approximately 6 foci per cell, with most small cells harboring 4 foci and longest ones harboring 8 to 10 foci. The number of foci per cell thus roughly corresponded to the copy number of R388 as calculated by qPCR, suggesting that most observed foci contained a single copy of the plasmid.
To determine the subcellular position of R388 plasmids, the distance from one cell pole to each focus was measured and plotted as a function of the cell length. For the sake of clarity, only distributions of foci in cells with 5 and 6 foci are presented in Figure 6A. Foci were broadly located and we observed no evidence for preferential positioning at the cell centre or at the ¼ and ¾ cell length positions. However, foci distribution was not random. Indeed, the proportions of cells containing at least one focus in each cell quarter were significantely different from those expected for a random distribution (Observed/expected for a random distribution: 45%/9,4% for 4-foci cells; 86%/23,4% for 5-foci cells; 89%/38,1% for 6-foci cells; p-values <10−4 in all cases using the χ2 test). Thus, foci appeared to be evenly distributed along the cells, suggesting a mechanism of active distribution of R388 plasmid copies.
To obtain a global view of foci assortment, we counted the number of foci located within five fractions of half-cells length (Figure 6E). The majority of foci were positioned equally within the four central slices of cells (from 0.1 to 0.5 fractional cell length). However, 7% of foci were located within the most polar region (from 0 to 0.1 fractional cell length, Figure 6A and 6E), showing that the broad distribution of foci extended to the cell poles.
We next observed the subcellular localization of the unstable R388ΔstbA plasmid, using the same parS/ParB-GFP system and conditions described above. Under these conditions, neither the parS insertions into R388, nor the expression of the GFP-Δ30ParB protein had a noticeable effect on R388parS plasmid stability and conjugative transfer (plasmid loss rate <0.01%; conjugative transfer frequency 2.10−1/donor). Representative images are shown in Figure 4B. About 11% of the cells were devoid of fluorescent focus (Figure 5), a value consistent with the degree of instability of the plasmid. A majority of cells showed a number of foci in the range of 1 to 3 (85%, Figure 5). The population average was 2 foci per cell, which is approximately 3-fold lower than R388. As mentioned above, the copy number of R388ΔstbA was found to be similar to that of R388, implying that most R388ΔstbA foci contained 2 or 3 plasmid copies. Thus, the decrease in number of foci is most likely due to plasmid clustering.
The subcellular distribution of foci in cells harboring R388ΔstbA is shown in Figure 6B and 6E. In cells with only one focus, the single focus was primarily in the polar region (91% of foci located from 0 to 0.2 fractional cell length). All cells having two or three foci had at least one polar focus. In cells having two foci, the other focus was localized mainly at the opposite polar region (48%) or in the cell center (38%). In cells having 3 foci, a majority had one focus at each pole and the other at the cell center (55%) or one focus at a pole and the two others at the cell center (43%). All in all, approximately 33% of the focus-containing cells contained all foci in one side of the cell and no focus close to the center, i.e., they are cells that would give rise to plasmid-free cells if the cell divided at the mid-position without any change in plasmid number or position. The location of plasmid foci relative to the nucleoid was determined by visualizing cells stained with DAPI (Figure S4). Fluorescence foci were mainly localized in nucleoid-free areas that were not occupied by the chromosomal DNA, either at the cell poles or at the cell center in the cytosol space between two nucleoids in dividing cells. Therefore, subcellular distribution of the unstable R388ΔstbA plasmid is markedly different from that of the stable R388 plasmid carrying stbA, and correlated with its instability in E. coli (see above).
To assay whether StbB had a role in R388 localization, we observed R388ΔstbB-harbouring cells (Figure 4C and Figure S4). The distribution of foci was almost identical to that of R388. More than 97% of the cells contained GFP-foci, and had focus numbers in the range of 4 to 10 (Figure 5), with a population average of 5.6 foci per cell. The mean number of R388ΔstbB molecules per cell was found to be 4.0±1.1, showing that, similarly to what was observed with R388, most foci contain a single molecule of the plasmid.
Figure 6C shows the subcellular positioning of R388ΔstbB plasmid molecules in cells containing 5 and 6 foci. R388 deleted of stbB was evenly distributed within the cell with the exception of the most polar region (from 0 to 0,1 fractional cell), which appeared to contain less foci than R388. Comparison of R388ΔstbB and R388 foci distributions within the five fractions of half-cells length (Figure 6E) using the χ2 test revealed that they were indeed significantly different (χ2 = 61.1 corresponding to a p-value <10−4). This difference can be attributed to a different polar localization of R388 and R388ΔstbB since the distribution of these plasmids within the three central slices of the half-cell (from 0.2 to 0.5) were not significantly different (p-value = 0.18). We concluded that StbB is required for the localization of a fraction of R388 copies towards the cell poles. Without StbB, R388 molecules are excluded from the poles.
To further investigate the role of StbB in intracellular positioning, we compared the localization of the R388ΔstbA and R388ΔstbABC plasmids. As in the case of R388ΔstbA (see above), R388ΔstbABC formed 1 to 5 fluorescent foci per cell, with a majority (78%) of cells containing 1 to 3 foci (Figure 4D and Figure 5), and showing a strong bias for location at the center and cell poles (Figure 6D). However, R388ΔstbABC and R388ΔstbA foci distributions were found to be significantly different (p-value <10−4). As illustrated in Figure 6E, this difference mainly relied on the proportion of foci within the most polar region (from 0 to 0.1 fractional cell length; R388ΔstbABC, 12%, compared to R388ΔstbA, 24%) and at the center (R388ΔstbABC, 33%, compared to R388ΔstbA, 22%). Thus, StbB is required for the localization of R388 at the pole and midcell positions in both the presence and absence of StbA.
In this study, we show that protein StbA is strictly required for stability and intracellular positioning of plasmid R388 in E. coli. We found that fluorescent foci of R388, most of which contain a single copy of the plasmid, are evenly distributed along the cell, with no evidence for preferential localization. This is in contrast with other low-copy number plasmids, such as mini-F, mini-P1, R27 and RK2 plasmids, which were reported to localize as clusters at the ¼–¾ or midcell positions [21], [23]–[26]. In these cases, duplication of the central focus is presumed to represent active partition of plasmid copies. However, it has been recently shown for mini-P1 plasmid that more than two foci are present in most conditions, and that the behavior of foci is more dynamic than previously reported [27]. This model of segregation of mini-P1 is more consistent with the even distribution of R388 copies that we have observed. We thus hypothesize that, as proposed for mini-P1, R388 copies segregate as single units and distribute into a dynamic and evenly spaced pattern along the cell to ensure a proper distribution of the plasmid copies at cell division. Moreover, in contrast to non-conjugative mini-F and mini-P1 plasmids, which were reported to be contained within the nucleoid region [21], [23], [25], [27], a significant fraction of R388 foci are found at the extreme cell ends. This observation may reflect plasmid R388 ability to undertake conjugative transfer. In agreement with this, we have recently shown that R388 coupling protein TrwB localizes to the cell poles (data not shown). Besides, it has been reported that the T4SS apparatus of Agrobacterium tumefaciens and of plasmid pCW3 from Clostridium perfringens, assemble at the cell poles [28]–[31]. In contrast to the even distribution of R388, R388ΔstbA plasmid foci clustered at the cell poles or at the cell center, in nucleoid-free areas. Mislocalized plasmid clusters appear to be the main cause of instability, as they are not adequately distributed in cellular spaces corresponding to future daughter cells (Figure 7). Mislocalized plasmid clusters were reported in derivatives of plasmids mini-F, mini-P1, R27, and R1 in which their Par regions were inactivated [25], [32]–[34]. In these cases, plasmid foci appeared distributed randomly in nucleoid-free spaces. This similarity suggests that StbA acts as a partition system. Indeed, the stb operon shares many characteristics with Par systems implicated in plasmid and chromosome partitioning. StbA is a DNA-binding protein which binds a cis-acting sequence (it is thus a ParB-like protein), the stbDRs, and StbB is a putative motor protein harboring Walker-type ATPase motifs (thus a ParA-like protein). However, and in contrast to ParA-like counterparts, StbB is not required for R388 stability, suggesting that the StbAB system does not constitute a typical ParAB system. StbB may either have no role in R388 maintenance or may be replaced by an equivalent cellular function when inactivated. Alternatively, R388 segregation may not need to involve an active motor. In this view, the StbA-stbDRs complex may be used to pair plasmid molecules with the host chromosome, ensuring an even distribution of R388 copies along the nucleoid length by an unknown mechanism.
Besides, we demonstrate that, although deletion of the entire stb operon does not affect conjugation, StbA and StbB, but not StbC are involved in R388 conjugation. Deletion of stbA results in an enhanced frequency of conjugation, while deletion of stbB leads to a conjugation defect, indicating that StbA and StbB have opposite but connected effects. As explained above, our results further suggest that these conjugation defects are a consequence of variations in the intracellular positioning of plasmid DNA. Whatever the way StbA promotes R388 segregation, the associated localization is certainly not convenient for maximal conjugation frequency, since StbA inactivation, associated with plasmid localization at the cell poles, strongly enhances conjugation. In addition, R388ΔstbB conjugation defect correlates with the absence plasmid foci at the cell polar membrane (Figure 7). We thus assume that the role of StbB is either to counteract StbA by locating some plasmid copies at the polar conjugal transport site, thereby allowing conjugation to occur, or StbA may modulate the activity of StbB and/or delocalize the plasmid copies.
The way the relaxosomal complex is transferred to conjugative pores remains unknown. It was previously suggested that A. tumefaciens VirC1 protein, which belongs to the ParA and Soj/MinD ATPases family, spatially positions the relaxosome at the cell pole to coordinate substrate-T4SS docking [35]. StbB also shares features with ParA/Soj/MinD ATPases (Figure 2B). These proteins are though to employ a principle of dynamic oscillation between specific surfaces such as membrane or bacterial chromosome to explore and mark the cellular environment [4]. Several models for the action of the Walker partition ATPases have been proposed. Formation of dynamically unstable filaments in a nucleotide-dependent manner was suggested following the example of actin-type partition ATPases [4]. Such cytomotive filaments could achieve partitioning by pushing plasmids attached to growing filaments, or by pulling plasmids attached to retracting filaments and there has been some evidence for both modes of action [36], [37]. Alternatively, a diffusion ratchet model was proposed (Vecchiareli et al., 2010). In this case, the motive force for plasmid positionning does not rely on the ParA ATPase polymerization, but instead is directed toward regions of high ParA concentration. These models are all consistent with our current data and such dynamic modes of action constitute appealing mechanisms for how StbB might recruit R388 molecules from a cytosolic pool to the membrane for conjugative transfer. Since inactivation of either the coupling protein (TrwB) or the relaxase (TrwC) did not affect plasmid R388 cellular localization (data not shown), the StbB-dependent mechanism of transport of R388 molecules to the cell membrane is neither associated with relaxosome formation, nor it requires either the coupling protein or cleavage at oriT. StbB may interact either with plasmid DNA or with StbA bound to DNA to form nucleoprotein complexes analogous to the ParA/ParB/parS partitioning complex, but linked to conjugative DNA processing. We are presently investigating the detailed molecular mechanisms by which StbB interacts with R388 molecules to recruit them to the cell membrane prior to transfer.
Our comparative genomics studies showed the conservation of synteny of three genes, of which the second gene is the most conserved, at the leading region of conjugative plasmids of mobility groups MOBF11, MOBP11, MOBP6, indicating that the stbABC operon is widespread among plasmids. Moreover, the stb operon is apparently linked to MPFT T4SS systems, although not exclusively, as it is also carried by mobilizable plasmids of the MOBP13/P14 group. It remains to be explored if these plasmids require a MPFT-type T4SS for conjugative transfer. Synteny conservation may reflect a requirement for stb in plasmids carrying such conjugation machinery under natural conditions. This is supported by our observation that the presence of R388 transfer region leads to instability of a pBR322-derivative plasmid containing it (data not shown).
In summary, we present experimental evidence that the StbAB system constitutes an atypical plasmid stabilization system intimately linked to conjugative transfer. On one hand, StbA is strictly required for plasmid R388 stability. Its inactivation results in mislocalization of R388 copies towards the cell poles, which is correlated with a significant increase in transfer frequencies (Figure 7). On the other hand, StbB is necessary for conjugative transfer only in the presence of StbA. Its inactivation leads to conjugation defect, which is associated with the absence of plasmid molecules at the extreme cell poles (Figure 7). Our results thus suggest that the StbAB system may act as a molecular scale between two possible transmission modes of plasmid R388: vertical transmission by faithful segregation to daughter cells and horizontal transmission by conjugative transfer to a different cell (Figure 7). It would seem that an active conjugation system provokes plasmid instability per se, which could only be counterbalanced by Stb or an analogous stability system. In this case, it remains to be investigated if other plasmids use functionally analogous but phylogenetically unrelated systems to balance their propagation and vegetative modes. This is, to our knowledge, the first report of a system involved in the reconciliation of these two cellular processes.
For bacterial strains, plasmid constructions and oligonucleotides, see Text S1 and Tables S1 and S2.
For mating experiments, donor (LN2666; [38]) and recipient (BW27783; [39]) strains were grown overnight from single colonies in LB medium at 37°C with appropriate antibiotics. After washing, 50 µl of donor cells were mixed with 800 µl of recipient cells, the mixture centrifuged for 1 min, resuspended in 10 µl LB medium and cells placed onto a GS Millipore Filter (0.22 µm pore size) on a LB-agar plate at 37°C for 20 min, which corresponds to a period shorter than the generation time to limit indirect effects due to plasmid instability. Bacteria were then washed from the filter, diluted in 2 ml LB medium and serial dilutions plated on selective media. Conjugation frequencies were expressed as the number of transconjugants per donor cell. When providing plasmids pStbA or pStbB (Table S1), the amount of protein StbA or StbB produced was found to be sufficient to restore the wt phenotype without the need to induce their expression with IPTG.
Single colonies of LN2666 or DH5α strains (Table S1) containing R388 or a derivative of it were used to inoculate LB containing selective antibiotics, and the cultures were incubated overnight at 37°C. For each strain, stability experiments were performed at least four times, starting from separate colonies. 48.8 µl of a 10-fold dilution of overnight cultures were transferred to 5 ml LB, containing streptomycin (300 µg/ml) but lacking the antibiotic selective for the plasmid, and grown for 12 h, i.e. for 10 generations. These freshly inoculated cultures constituted time point zero. From then on, 48.8 µl of a 10-fold dilution of the full-grown cultures was transferred every 12 h to fresh 5 ml LB and incubated at 37°C to reach a total of 80 generation times. Each 12 h, the cultures were also diluted and plated onto LB plates. Determining the fraction of plasmid free cells in the population was done by replica-picking 100 randomly chosen colonies per culture from the LB plates onto LB plates containing the appropriate selective antibiotics, and scoring the proportion of colonies with a given resistance. The percentage of plasmid loss per generation was calculated as described in Yates et al., 1999 [40].
A derivative of strain LN2666 (recA+) containing plasmid pALA2705 (Ap; Table S1), which produces the fluorescent GFP-Δ30ParB protein [21], was transformed with DNA of plasmid R388::parS-Cm or one of its derivatives. Neither the parS insertions into R388 (or stb derivatives listed in Table S1), nor the expression of the GFP-Δ30ParB protein had a noticeable effect on the corresponding R388parS plasmid stability and conjugative transfer (data not shown). Single-colony isolates were grown overnight in M9 medium supplemented with 0.2% casamino acids, 0.4% glucose, 2.0 µg/ml thiamine, 20 µg/ml leucine and 20 µg/ml thymine, (suppl. with Ap, Cm) at 30°C. Cultures were then diluted 1/100 in the same medium and grown at 30°C to OD600 of 0.8. With these constructions, foci could be adequately visualized without the need to induce expression of the fluorescent GFP-Δ30ParB protein [27]. When needed, DAPI stain (1 µg/ml; Molecular Probes) was added to the culture for 20 min to label DNA. Suspensions of growing cells were directly deposited on glass slides covered by a layer of 1% agarose containing the same growth medium and examined by phase-contrast and fluorescence microscopy. Images were captured with an inverted Olympus X81 microscope equipped with a 100x oil-immersion Olympus lens (N.A. of 1.3) and a Roper Coolsnap CCD camera, using Metamorph software. Cell length and focus position was measured manually using ImageJ. Each strain was examined in at least four independent experiments with similar results. At least 200 cells were inspected for each experimental observation.
|
10.1371/journal.pgen.1001350 | REVEILLE8 and PSEUDO-REPONSE REGULATOR5 Form a Negative Feedback Loop within the Arabidopsis Circadian Clock | Circadian rhythms provide organisms with an adaptive advantage, allowing them to regulate physiological and developmental events so that they occur at the most appropriate time of day. In plants, as in other eukaryotes, multiple transcriptional feedback loops are central to clock function. In one such feedback loop, the Myb-like transcription factors CCA1 and LHY directly repress expression of the pseudoresponse regulator TOC1 by binding to an evening element (EE) in the TOC1 promoter. Another key regulatory circuit involves CCA1 and LHY and the TOC1 homologs PRR5, PRR7, and PRR9. Purification of EE–binding proteins from plant extracts followed by mass spectrometry led to the identification of RVE8, a homolog of CCA1 and LHY. Similar to these well-known clock genes, expression of RVE8 is circadian-regulated with a dawn phase of expression, and RVE8 binds specifically to the EE. However, whereas cca1 and lhy mutants have short period phenotypes and overexpression of either gene causes arrhythmia, rve8 mutants have long-period and RVE8-OX plants have short-period phenotypes. Light input to the clock is normal in rve8, but temperature compensation (a hallmark of circadian rhythms) is perturbed. RVE8 binds to the promoters of both TOC1 and PRR5 in the subjective afternoon, but surprisingly only PRR5 expression is perturbed by overexpression of RVE8. Together, our data indicate that RVE8 promotes expression of a subset of EE–containing clock genes towards the end of the subjective day and forms a negative feedback loop with PRR5. Thus RVE8 and its homologs CCA1 and LHY function close to the circadian oscillator but act via distinct molecular mechanisms.
| Circadian clocks help organize 24-hour rhythms in physiology and behavior so that critical organismal functions are optimally timed relative to highly predictable daily changes in the environment. Circadian clocks run at approximately the same pace across a wide range of temperatures, ensuring accurate timekeeping in all seasons. Although molecular components of the circadian clock are not conserved across higher taxa, eukaryotic circadian clocks are composed of analogous interlocked transcriptional feedback loops. In this study, we report the isolation and characterization of a new component of the plant circadian system, REVEILLE 8 (RVE8). RVE8 is a clock-regulated Myb-like transcription factor that binds with high affinity to the evening element (EE) promoter motif and helps to set the pace of the clock in a light- and temperature-dependent manner. RVE8 promotes expression of the clock component PSEUDO-RESPONSE REGULATOR 5 (PRR5), likely via direct action at the PRR5 promoter. RVE8 expression is in turn repressed by PRR5. Thus, RVE8 is a new component of the plant circadian oscillator that takes part in a novel transcriptional feedback loop.
| We live on a world with prominent and predictable daily changes in the environment. To better anticipate and mitigate these changes, most organisms possess circadian clocks, internal timers that generate roughly 24-hour rhythms in physiology even in the absence of environmental cues. Processes influenced by the circadian clock include once-in-a-lifetime developmental events such as the eclosion of insects from their pupal cases or the transition of plants from vegetative to reproductive growth, to daily events such as changes in activity levels in animals or the opening and closing of flowers [1], [2]. Given this diversity of clock outputs, it is not surprising that the circadian system influences expression of a substantial fraction of the genome, with 30% of plant genes and 10% of mammalian genes estimated to be circadian regulated [3]–[6].
Although the molecular components of the circadian clock are not conserved across higher taxa, basic features of the circadian system are shared. In all organisms that have been investigated, circadian clocks are cell autonomous and can be reset by environmental cues such as changes in light or temperature [7]–[9]. However, circadian clocks are strongly temperature compensated; that is, they run at a similar pace across the physiologically relevant range of temperatures [10], [11]. This allows them to keep accurate time in all seasons. In eukaryotes, transcriptional feedback loops play a crucial role in the circadian oscillator, although post-transcriptional events such as regulated protein degradation are also essential for robust clock function [12].
In recent years, rapid progress has been made in uncovering the mechanisms underlying clock function in many model organisms. For example, molecular genetic and genomic studies in Arabidopsis thaliana (Arabidopsis) have led to the identification of interlocked transcriptional feedback loops that act at the heart of the plant clock [1]. Two Myb-like transcription factors, CIRCADIAN CLOCK ASSOCIATED 1 (CCA1) and LATE ELONGATED HYPOCOTYL (LHY), inhibit expression of TIMING OF CAB EXPRESSION 1 (TOC1), which encodes a nuclear-localized protein that indirectly promotes expression of CCA1 and LHY, forming the first transcriptional feedback loop [13], [14]. A second negative feedback loop is formed between CCA1 and LHY and two TOC1 homologs, PSEUDO-RESPONSE REGULATOR 7 and 9 (PRR7 and PRR9). CCA1 and LHY promote expression of PRR7 and PRR9, which in turn negatively regulate CCA1 and LHY via direct binding to their promoters [15], [16]. Another TOC1 homolog, PRR5, has recently been shown to also negatively regulate CCA1 and LHY expression [17], although it is not currently clear how expression of PRR5 itself is regulated. A third negative feedback loop has been proposed, based primarily on mathematical modeling, in which an unknown component forms a negative feedback loop with TOC1 [18].
In addition to the transcriptional feedback loops, post-transcriptional control mechanisms are also indispensible for normal clock function. ZEITLUPE (ZTL), a blue light photoreceptor with an F-box domain, is involved in the regulated degradation of TOC1 [19]. ZTL stability is in turn regulated by its light-dependent interaction with GIGANTEA (GI) [20]. Regulation of clock protein phosphorylation and intracellular localization are also important control mechanisms [21]–[23].
Analysis of circadian regulated genes has led to the identification of several promoter motifs implicated in the phase-specific regulation of gene expression [4], [24]–[26]. One such motif is the evening element (EE, AAAATATCT), which when multimerized is both necessary and sufficient to confer evening-phased expression on a reporter gene [27]. A number of clock genes, including TOC1, GI, and PRR5, have EE sequences in their promoters. CCA1 and LHY bind to the EE-containing region of the TOC1 promoter to directly repress TOC1 expression [13]. We have recently shown that a protein related to CCA1 and LHY, REVEILLE 1 (RVE1), also binds specifically to the EE. However, rather than affecting central clock function, RVE1 controls daily rhythms of auxin production, serving as a node connecting the circadian and auxin signaling networks [28]. Two RVE1 homologs, CIRCADIAN 1 (CIR1)/RVE2 and EARLY-PHYTOCHROME-RESPONSIVE 1 (EPR1)/RVE7, also seem to function primarily as clock outputs via undefined mechanisms [29], [30].
Although a framework describing the plant circadian oscillator is now in place, there are still considerable gaps in our understanding. The interactions between known components are not completely understood, some clock genes have been identified but not yet placed into the clock model, and some components have been predicted but not yet identified on a molecular level [1]. We have therefore taken a biochemical approach to identify new factors that act within the circadian system. Previously, we found a specific EE-binding activity in extracts made from both wild-type and cca1 lhy plants [27]. We now report on the characterization of RVE8, an EE-binding protein identified by liquid chromatography tandem mass spectrometry (LC-MS/MS) analysis. RVE8 is a member of a previously uncharacterized clade in the CCA1/LHY/RVE transcription factor family. RVE8 acts not only in setting the pace of the clock, but also plays roles in temperature compensation and light signaling. Although the circadian transcriptional profiles of CCA1, LHY, and RVE8 are very similar, the protein accumulation patterns are quite different: RVE8 protein peaks in the subjective afternoon while CCA1 and LHY proteins peak near subjective dawn. Our data suggest that RVE8 directly promotes expression of PRR5. Microarray data indicate that PRR5, in turn, represses expression of RVE8. Thus RVE8 and PRR5 comprise a negative transcriptional feedback loop that acts within the plant circadian network.
In previous experiments, we identified an afternoon-phased activity in plant extracts that bound specifically to the EE [27]. To identify the factor(s) responsible, we used wild-type and mutant versions of the EE bound to magnetic beads to isolate DNA binding proteins from plants harvested eight hours after dawn. Purified proteins were subjected to LC-MS/MS and peptide sequences compared to the Arabidopsis proteome. At a peptide false discovery rate of 0.1%, a number of transcription factors were identified, including three Myb-like, two B3-domain, two trihelix, two WHIRLY, one WRKY, and one basic-leucine zipper transcription factors (Table 1). Several of these factors were identified specifically in wild-type but not in mutant EE samples. Only the Myb-like factor RVE8 (At3g09600) was identified in each of three independent experiments; intriguingly, its close homolog RVE4 (At5g02840) was also identified in one of the replicates. In addition to the peptides that could be assigned uniquely to either RVE8 or RVE4 (indicated in Table 1), additional peptides that could have been derived from either protein were identified in each of the three experiments (Table S1). Using more liberal filtering parameters to identify proteins with lower sequence coverage, peptides uniquely derived from RVE5 and RVE6 could also be identified (Table S1). Complete protein and peptide identification data can be found at ProteomeCommons.org and in Table S2.
The RVE genes are part of a small family of transcription factors that includes the well-studied clock genes CCA1 and LHY. Many RVE proteins have previously been reported to bind the EE in vitro [31]. One clade consists of RVE1, RVE2/CIR1, and RVE7/EPR1, which likely functional primarily as clock outputs [28]–[30]. RVE3, RVE4, RVE5, RVE6, and RVE8 belong to a separate clade not previously functionally characterized. Expression of RVE3, RVE4 and RVE8, but not RVE5 or RVE6, was clock-regulated in seedlings, with peak message levels occurring near subjective dawn (Figure 1) [4], [32]. RVE8 expression in particular was very similar in amplitude and phase to that of the known EE-binding proteins CCA1, LHY, and RVE1 (Figure 1) [28], [33], [34].
Since RVE8 was the only transcription factor identified as a sequence-specific EE binding protein in all three of our LC-MS/MS experiments, we further characterized its ability to bind to the EE. First, we used yeast one-hybrid experiments to assay RVE8 binding in a heterologous system. RVE8 interacted with wild-type but not mutant EE sequences, showing similar EE binding activity as CCA1 and RVE1 (Figure 2A, 2B). Electrophoretic mobility shift assays (EMSAs) carried out with recombinant RVE8 also demonstrated that RVE8 bound the EE specifically and with similar affinity as RVE1 (Figure 2C–2D).
We next investigated whether RVE8 might contribute to the EE binding activity we found in plant extracts, first generating plants overexpressing HA-tagged RVE8 (Figure S1A). A similar amount of EE binding activity was detected in 25 µg of extract made from RVE8-OX plants as in 100 g of wild-type extract (Figure 2E, compare lanes 3 and 8). However, the mobility of DNA/protein complexes was slightly different depending upon whether extracts were made from wild-type (arrow) or RVE8-OX (arrowhead) plants, suggesting that the composition of EE binding complexes was somewhat different in these genotypes.
The increased EE binding activity in RVE8-OX plant indicated that RVE8 directly or indirectly contributed to EE binding activity in these plants. To help distinguish between these possibilities, we added anti-HA antibody to our EMSA reactions. The addition of anti-HA antibody caused a reduction in the amount of shifted probe in the RVE8-HA lanes (Figure 2E; compare lanes 2 and 5, and lanes 3 and 6) but no reduction in binding activity in the wild-type control lanes (Figure 2E, compare lanes 10 and 11). These data suggested that RVE8 can bind directly to the EE in plants, at least when overexpressed, and encouraged us to further examine its role in the circadian system.
Plants with circadian clock defects frequently also show changes in light-mediated inhibition of hypocotyl elongation and photoperiodic control of flowering time [35], [36]. Indeed, overexpression or mutation of CCA1, LHY, and several other RVE-family genes affects these processes [29], [30], [33], [34], [37]. We therefore examined light regulation of hypocotyl growth and flowering time responses in both RVE8-OX and SALK-053482, a rve8 loss-of-function mutant obtained from the SALK T-DNA collection [38]. This mutant, designated rve8-1, did not express detectable levels of RVE8 message (Figure S1B, S1C) and is thus likely a null allele.
Appropriate photoperiodic control of flowering time requires a functional circadian clock [36]. We therefore examined time to flowering in plants grown in short day (SD; 8 hours light:16 hours dark) or long day (LD; 16 hours light:8 hours dark) conditions. rve8-1 plants flowered slightly earlier than wild type in both LD and SD (2.6 days earlier in LD and 5.8 days earlier in SD) (Figure 3A, 3B). The effects of RVE8-OX overexpression were also relatively modest, but acted to delay rather than promote flowering (RVE8-OX plants flowered 3.8 days later than wild type in LD and 17 days later in SD) (Figure 3A, 3B). Although these effects were small, they were reproducible and highly statistically significant. These moderate differences in time to flowering are quite different from the strong flowering time phenotypes seen in CCA1 and LHY mutant and overexpressing plants [33], [34], [37], but are reminiscent of the relatively subtle effects on flowering time reported in rve2, RVE2-OX, and RVE7-OX plants [29], [30].
We next examined the hypocotyl length of seedlings grown in different light conditions. Hypocotyl lengths of etiolated RVE8-OX and rve8-1 seedlings were very similar to those of wild-type seedlings (Figure 3C). However, when grown in low or medium fluence-rate constant white light, rve8-1 seedlings had longer hypocotyls and RVE8-OX had shorter hypocotyls than wild type (Figure 3C). These differences in phenotype diminished at higher fluence rates of constant white light. When grown in SD conditions, rve8-1 and RVE8-OX displayed long and short hypocotyls, respectively, over a range of fluence rates, although the differences were slightly reduced at the highest light intensity tested (Figure 3D). The rve8-1 long-hypocotyl phenotype was rescued by transformation of these plants with a wild-type genomic copy of the RVE8 locus (Figure 3E), demonstrating that loss of RVE8 expression was indeed responsible for this trait.
Given the dawn phase of expression of RVE8, it was somewhat surprising that we identified peptides uniquely derived from this protein in extracts from plants harvested eight hours after dawn. To investigate the temporal pattern of RVE8 protein accumulation, we performed immunoblots on samples made from a rve8-1 line rescued with a RVE8::RVE8-HA construct (Figure 3E). We found that peak levels of RVE8-HA protein occurred three to six hours after subjective dawn (ZT27 – ZT30; Figure 4A, 4B). In contrast, CCA1 and LHY protein levels closely track their respective transcript levels [33], [39]. However, post-transcriptional regulation of clock genes is not uncommon; for example, accumulation of many of the pseudo-response regulator proteins that function in the plant clock is also significantly delayed relative to their transcript profiles [21], [40]–[42], as is accumulation of the clock protein PER in animals [43], [44]. This delay helps explain why RVE8 was identified as an afternoon-phased EE-binding protein. The mechanism underlying the sizable delay in RVE8 protein accumulation relative to its transcript is not clear and will be an interesting topic for further research.
Loss-of-function alleles of CCA1 and LHY cause period shortening whereas overexpression of these genes causes arrhythmicity [33], [34], [37], [45], [46]. Similarly, overexpression of RVE1, RVE2, or RVE7, all members of the same clade, reduces rhythmic amplitude [28]–[30]. However, rve1, rve2, and rve1 rve2 rve7 mutants have no circadian phenotypes [28], [30] suggesting that they normally primarily act as output genes rather than clock components. In contrast, RVE8 overexpression caused an approximately one hour shortening of free-running rhythms of CCR2::LUC activity when plants were maintained in constant red or red + blue light (Figure 5A and 5E). In constant blue light, the average period of RVE8-OX plants was only 0.3 hours shorter than wild type, a difference that was not statistically significant in this experiment (Figure 5C). In contrast, the average free-running period of rve8-1 plants was approximately 1 hour longer than wild type in all three light conditions (Figure 5A, 5C and 5E; see figure legend for period estimates). The robustness of rhythms, as measured by the relative amplitude error (RAE), was not appreciably different in rve8-1 or RVE8-OX compared to Col controls (Figure 5B, 5D, and 5F).
Changes in free-running period are often accompanied by changes in the expression levels of core clock genes. In particular, CCA1, LHY, and other members of the RVE family have previously been reported to regulate each other's expression levels [28], [30], [33], [34], [37]. We therefore used quantitative reverse-transcriptase polymerase chain reaction (qRT-PCR) assays to monitor expression of CCA1, LHY, and TOC1 in rve8-1 and RVE8-OX plants. No changes in expression of LHY or CCA1 were observed immediately upon release into constant conditions in either rve8-1 or RVE8-OX; however, the times of peak expression were clearly altered after five days in free running conditions (Figure 6A, 6B). In rve8-1 plants, the time of peak expression was delayed whereas in RVE8-OX the time of peak expression was advanced, consistent with the respective long- and short-period CCR2::LUC phenotypes seen in these genotypes (Figure 5). In contrast, the phase of TOC1 expression was appreciably advanced in RVE8-OX in the first day of free run, although a delay in the peak phase of TOC1 expression in rve8-1 was not seen until the plants had been in free run for five days (Figure 6C). The late phase of expression of all three clock genes was rescued by introduction of a RVE8 transgene expressed under the control of the native promoter (Figure 6D–6F), indicating that the rve8-1 mutation was indeed responsible for the observed long-period phenotype. However, there were no reproducible changes in overall expression levels of CCA1, LHY, or TOC1 in either RVE8-OX or rve8-1. This suggests these core clock genes are not significant targets for RVE8.
Light signaling pathways affect the circadian clock, both setting clock phase and influencing the pace at which the clock runs. When monitored under constant light conditions, Arabidopsis plants display a shorter free-running period as fluence rates are increased [47]. We speculated that the altered free-running periods in rve8-1 and in RVE8-OX might be due to altered sensitivity to light in these genotypes. To test this, we examined the effects of different fluence rates of continuous red or blue light on CCR2::LUC rhythms in Col, rve8-1, and RVE8-OX. After entrainment in white light/dark cycles, seedlings were moved to free run in continuous red or blue light of different fluence rates. In red light, both the rve8-1 long-period phenotype and the RVE8-OX short-period phenotype were observed across a wide range of fluence rates (Figure 7A). The responsiveness of rve8-1 mutants to red light was very similar to that of wild type, whereas the short-period phenotype of RVE8-OX seedlings was slightly enhanced at higher fluence rates of red light. Two-way ANOVA analysis revealed that only about 3% of the variance could be attributed to a genotype by fluence rate interaction. In contrast, the genotypes and fluence rates each accounted for approximately 30% of the total variance. This indicates that the red light sensitivity of the circadian clock in Col, rve8-1, and RVE8-OX plants is fundamentally similar.
We next examined the effects of different fluence rates of blue light on Col, rve8-1, and RVE8-OX. As is shown in Figure 7B, RVE8-OX plants had a period approximately 0.6 hours shorter than wild type at most fluence rates tested, a weaker effect than seen in constant red light. The long-period phenotype of rve8-1 plants was observed at all but the lowest fluence rate (0.2 mol m−2 s−1) tested; however, it is possible that the relatively poor rhythmicity of all three genotypes in this low-light condition may have masked a period phenotype in rve8-1. Two-way ANOVA analysis indicated there was not a significant genotype by fluence rate interaction. Therefore clock sensitivity to continuous blue light is not altered in plants with perturbed RVE8 expression.
Since mutation of genes involved in light input to the clock can affect resetting of clock phase in response to light [48], we examined how rapidly Col, rve8-1, and RVE8-OX plants re-entrained to new light conditions. Wild-type and mutant plants containing a CCR2::LUC reporter gene were entrained in 12 hrs light: 12 hrs dark and luciferase rhythms were initially monitored under the same photocycles (Figure 7C). Even in light/dark cycles, the early-phase phenotype of RVE8-OX and the late-phase phenotype of rve8-1 plants were evident. We next subjected these three genotypes to an extended night of 24 hours before resuming light/dark cycles. All three genotypes recovered from this 12 hour ‘jet lag’ treatment at approximately the same rate, regaining appropriately evening-phased CCR2::LUC activity within two days of the perturbation. Together, the fluence rate response curve and phase-resetting data strongly suggest that the period phenotypes in rve8-1 and RVE8-OX are not caused by changes in light input to the circadian clock.
To assess the importance of light for the rve8-1 and RVE8-OX period phenotypes, we examined free-running rhythms in constant darkness (DD) after entrainment in light/dark cycles. RVE8-OX plants had a period approximately one hour shorter than that of the Col control, similar to the RVE8-OX phenotype in constant red or red + blue light. In contrast, the rve8-1 free running period was slightly shorter than but not significantly different from wild type (Figure 7D). This is in marked contrast to the long-period phenotype of rve8-1 plants seen in all light conditions tested (Figure 5, Figure 6, and Figure 7A and 7B). Therefore the rve8-1 circadian phenotype is light-dependent, even though light input to the clock is not appreciably altered in this mutant.
Circadian clocks are typically robustly temperature compensated; that is, they maintain approximately the same free-running period over a physiologically relevant range of temperatures [1]. Since mutation of clock genes can specifically affect this process, we examined periodicity of CCR2::LUC activity in Col, rve8-1, and RVE8-OX plants at different ambient temperatures. rve8-1 seedlings had significantly longer free-running periods when assayed at 22°C or 27°C (Figure 7E), but essentially wild-type periodicity at 12°C or 17°C. RVE8-OX plants had significantly shorter free-running rhythms than Col at 17°C, 22°C, and 27°C but had the same period as wild-type controls at 12°C (Figure 7E). Therefore temperature compensation is disrupted in both rve8-1 and RVE8-OX plants, with loss- and gain-of-function RVE8 alleles showing normal rhythmicity at the lowest temperature assayed. This suggests that RVE8 normally functions only at the high end of the range of physiologically relevant temperatures.
Since RVE8 binds to EE sequences in vitro, we next investigated whether it binds to promoters containing these motifs in vivo. TOC1 is perhaps the best-known clock gene that contains an EE within its regulatory region [13]. However, other genes that function within the circadian network also contain EE sequences in their promoter regions. One, PRR5, has a similar phase of expression as TOC1 and has recently been shown to regulate nuclear accumulation of TOC1 protein and to repress CCA1 and LHY expression [17], [23]. We therefore investigated whether RVE8 could bind to the EE-containing regions of the TOC1 and PRR5 promoters using chromatin immunoprecipitation (ChIP) assays followed by qRT-PCR. We used plants entrained in white light/dark cycles and then transferred to constant red light since the phenotype of RVE8-OX plants was strongest in this condition (Figure 7A, 7B). Wild-type, rve8-1 + RVE8::RVE8-HA, and rve8-1 + 35S::RVE8-HA plants were harvested at subjective dawn (Figure 8A) and in the subjective afternoon (Figure 8B). ChIPs were carried out using anti-HA antibodies (experimental samples) and anti-GST antibodies (negative controls) and the ratio of genomic DNA isolated in each type of IP was determined. In plants harvested at subjective dawn, PRR5 and TOC1 sequences were slightly enriched (∼2 fold) in extracts made from RVE8::RVE8-HA plants compared to wild-type controls. A greater enrichment for these promoter sequences (∼6 fold) was found in RVE8::RVE8-HA plants harvested in the subjective afternoon (Figure 8B). In contrast, a similar strong enrichment for TOC1 and PRR5 promoter sequences was found in extracts made from plants overexpressing RVE8 harvested at either subjective dawn or subjective afternoon (Figure 8A, 8B). These data indicate that RVE8 binds to EE-containing promoter sequences in vivo. Since RVE8 protein levels are higher in the afternoon than in the morning (Figure 4) it is not surprising to find that RVE8 expressed under its endogenous promoter has more EE binding activity at that time.
In addition to binding to EE sequences, CCA1 and LHY are thought to regulate clock gene expression by binding to a related motif termed the CCA1 binding site (CBS) [14], [49]; for example, CCA1 and LHY are thought to promote expression of PRR7 by binding to a CBS-containing region of the PRR7 promoter [16]. We therefore examined the ability of RVE8 to bind to this region of the PRR7 promoter by ChIP. However, at both subjective dawn and afternoon there was not appreciable enrichment of the PRR7 promoter region in plants expressing RVE8 under its native promoter, and only a very modest enrichment was seen in plants overexpressing RVE8 (Figure 8A, 8B). These data suggest that RVE8 does not bind to the CBS-containing portion of the PRR7 promoter in vivo.
We next examined expression levels of PRR5, PRR7, and TOC1 in plants grown in the same conditions used for the ChIP assays. PRR5 levels were very similar in Col and rve8-1 two and three days after transfer to constant environmental conditions. However, trough levels of PRR5 were significantly elevated in RVE8-OX plants; this was especially apparent during day three of free run (Figure 8C). In contrast, peak and trough expression levels of TOC1 and PRR7 were very similar in Col, rve8-1, and RVE8-OX (Figure 8D, 8E).
Since RVE8 binds to the PRR5 promoter and PRR5 transcript levels are elevated in plants overexpressing RVE8, we examined the temporal relationship between amounts of RVE8 protein and PRR5 transcript (Figure 8F). Their levels are highly correlated, with both peaking in the subjective afternoon and having low levels during the subjective night. Together with the direct binding of RVE8 to the PRR5 promoter and the increase in PRR5 expression in RVE8-OX plants, this suggests that RVE8 directly and positively regulates PRR5 expression. In contrast, TOC1 levels are not appreciably altered in RVE8 loss- or gain-of-function alleles (Figure 6C and Figure 8D), suggesting that although RVE8 binds to the TOC1 promoter in vivo it may not regulate TOC1 expression.
PRR5, PRR7, and PRR9 have recently been shown to be direct and potent repressors of CCA1 and LHY expression [17]. Since RVE8 has a very similar pattern of gene expression to these homologs (Figure 1), we examined whether RVE8 expression was altered in prr5 prr7 prr9 triple mutants and in PRR5-overexpressing plants. Using publicly accessible microarray data [50], we found that RVE8 levels were significantly increased in prr5 prr7 prr9 triple mutants and decreased in PRR5-overexpressing plants (Figure 8G, 8H). This indicates that PRR5 represses RVE8 expression, either directly or indirectly. RVE8 and PRR5 therefore comprise a novel negative feedback loop within the plant circadian network.
We isolated RVE8 from plant extracts based upon its ability to bind to the EE. Further investigation in yeast and in vitro (Figure 2) revealed its EE-binding affinity is similar to that of both CCA1 and RVE1, which is not surprising since these proteins all contain a similar Myb-like/SANT DNA binding domain. The Myb-like domain of RVE8 shares 64% identity and 86% similarity with that of CCA1; its relatedness to LHY and RVE1 is similar. Sequence similarity between these proteins is also seen in short basic regions immediately N-terminal and short proline-rich regions immediately C-terminal to the Myb-like domains. These three regions are conserved among all 11 of the RVE-related proteins [28]. A search of the Eukaryotic Linear Motif resource database [51] suggests a portion of the conserved proline-rich region (PRPKRKAA in RVE8) may act as a nuclear localization signal while part of the conserved basic region (RKPYTIT in RVE8) may be a binding site for 14-3-3- proteins. 14-3-3 proteins often bind to their ligands in a phosphorylation-dependent manner, and this binding may affect client protein activity or intracellular localization [52]. It will be very interesting to determine whether the RVE-related proteins are regulated in this manner.
It is notable that despite the similar affinity of RVE8, RVE1, and CCA1 for the EE (Figure 2B) we identified peptides derived from RVE4, RVE5, RVE6 and RVE8 but not from other RVE-related proteins in our affinity purifications. This may be explained by our finding that RVE8 protein levels are high in the subjective afternoon (Figure 4) whereas CCA1 and LHY proteins are difficult to detect at that time [33], [39]. It will be of interest to determine whether RVE4, RVE5, and RVE6 protein levels are also highest in the subjective afternoon.
Mutations in clock genes often affect temperature compensation, the ability of the circadian system to run at a similar pace across a wide range of temperatures. For example, mutation of casein kinase 2 in Neurospora, casein kinase I epsilon in hamster, or PERIOD in Drosophila, affects clock pace and disrupts temperature compensation [53]–[55]. Similarly, mutation of CCA1 or LHY differentially affects circadian period at different temperatures in Arabidopsis. CCA1 function is more important at lower temperatures while LHY function is more important at higher temperatures [56]. We show that RVE8 is also involved in temperature compensation, but in a different manner from its homologs: the period phenotypes of the RVE8 mutant and overexpressing plants completely dissipated at the lowest temperatures tested (Figure 7E). It has recently been reported that prr7 prr9 mutants have a similar temperature compensation phenotype as rve8 mutants, with normal free-running rhythms at low temperature and longer rhythms at higher temperatures [57]. It will be very interesting to determine whether other RVE genes also exhibit temperature-dependent phenotypes and to investigate possible temperature-dependent regulatory relationships between the RVEs and the PRRs.
Light is a potent regulator of most aspects of plant growth and development, not only directly affecting many processes that are also modulated by the circadian clock [35], [58] but also influencing clock function directly. Illustrating the intertwined natures of the clock and light signaling networks, many genes that act close to the plant circadian oscillator, such as TOC1, ZTL, PRR5 and PRR7, also function in light signaling pathways [59]–[62]. It can therefore be difficult to determine whether clock mutant phenotypes are due to alterations in light signaling, the circadian oscillator, or both. To address this point, we examined the light-dependence of developmental and clock phenotypes in rve8-1 and RVE8-OX plants.
Using light inhibition of hypocotyl elongation to investigate the role of RVE8 in light-regulated development, we found that rve8-1 and RVE8-OX seedlings had long and short hypocotyls, respectively, at low fluence rates of continuous white light. Thus RVE8 differs from its characterized homologs in that it represses rather than promotes hypocotyl elongation. Interestingly, the RVE8 phenotypes were less obvious at fluence rates of 8 mol m−2 s−1 or higher, and almost absent at a fluence rate of 85 mol m−2 s−1 (Figure 3C). This type of light-dependent phenotype is reminiscent of mutants in the phyA signaling pathway such as fhy1 [63], and suggests that RVE8 may be a positive mediator of the very low fluence response. None of the other characterized RVE-related genes shows a similar low-light dependent hypocotyl phenotype, indicating they affect hypocotyl growth in a fundamentally different manner from RVE8 [28]–[30], [33], [34], [59].
Unlike the hypocotyl phenotypes, rve8-1 and RVE8-OX displayed similar circadian period phenotypes across a wide range of fluence rates (Figure 7A, 7B). This discrepancy strongly suggests that the primary mechanism by which RVE8 affects the pace of the clock is not via a light input pathway. In support of this conclusion, rve8-1 and RVE8-OX showed similar kinetics of clock re-setting in a jet lag experiment (Figure 7C). Given these results, it was somewhat surprising to discover that rve8-1 mutants did not have a long-period phenotype when assayed in constant darkness (Figure 7D), demonstrating that RVE8 function is light-dependent. However, this is not a unique observation since other clock-associated genes including EARLY FLOWERING 3 (ELF3), JUMONJI DOMAIN-CONTAINING PROTEIN 5 (JMJD5), PRR5, and PRR9 have been reported to have light-dependent period phenotypes [64]–[66]. Together, these data indicate that RVE8 functions in multiple signaling pathways, affecting light signaling and clock pace via different mechanisms.
What is causing the period phenotypes in rve8-1 and RVE8-OX plants? By analogy with CCA1 and LHY, one obvious role for RVE8 might be the direct regulation of TOC1. Indeed, we found that RVE8 binds directly to the TOC1 promoter in vivo (Figure 8A, 8B). However, the overall levels of TOC1 expression were not appreciably altered in either rve8-1 or RVE8-OX (Figure 6C, 6F, and Figure 8D). This suggests that despite the ability of RVE8 to bind the TOC1 promoter, TOC1 is not an important target for RVE8 function. It is not uncommon for transcription factors to bind to the promoters of genes in vivo but not affect their expression levels [67], [68]. This discrepancy may be explained by the requirement at some promoters for specific combinations of transcription factors to activate gene expression [69].
Whereas we found no changes in TOC1 expression levels in plants misexpressing RVE8, trough levels of PRR5 transcript were increased in RVE8-OX plants (Figure 8C). Combined with the ability of RVE8 to bind the PRR5 promoter and the correlation between RVE8 protein and PRR5 message levels (Figure 8A, 8B, 8F), this suggests that PRR5 is a direct target of RVE8. However, PRR5 levels are not affected in rve8 mutants. Given that we isolated not only RVE8 but also its homologs RVE4, RVE5, and RVE6 bound to the EE (Table S1), it may be that these proteins function in a partially redundant manner. The lack of apparent changes in PRR5 levels in rve8-1 (Figure 8C) might be caused by such redundancy. Similar findings have been previously reported; for example, cca1 and lhy single mutants have no apparent change in TOC1 levels but cca1 lhy double mutants have greatly increased levels of TOC1 [37]. CCA1 transcript levels are only slightly altered in prr5, prr7, or prr9 single mutants [16], [62], [66], [70], [71] but are greatly increased in prr7 prr9 double and prr5 prr7 prr9 triple mutants [15], [16]. We are currently generating higher-order rve mutants to investigate whether PRR5 levels are altered in these plants. It seems likely that RVE8 has many targets and that the long-period rve8 phenotype is due to subtle alterations in expression levels of multiple clock genes.
Members of the three separate clades of the CCA1/LHY/RVE family of transcription factors have now been found to fall into separate functional categories. CCA1 and LHY are the best-studied and affect clock pace by repressing TOC1 expression and promoting expression of PRR7 and PRR9 [13], [16]. They also profoundly influence control of hypocotyl elongation and flowering time by regulating expression of the PHYTOCHROME INTERACTING FACTOR4 (PIF4) and PIF5 transcription factors and genes in the photoperiodic pathway, respectively [36], [72]. RVE1 acts primarily as a clock output gene, regulating daily rhythms in auxin production but not playing an important role in clock function [28].
We now show that RVE8 plays a distinct role in the light signaling and circadian networks. RVE8 affects light inhibition of hypocotyl elongation at low light intensities, suggesting it may affect the phytochrome-mediated very low fluence response, a novel phenotype for a clock mutant. RVE8 also acts in temperature compensation, suggesting that its activity is important for fine-tuning clock function in different environmental conditions.
Our findings also provide insight into the architecture of the transcriptional networks that make up the plant circadian clock. PRR5, PRR7, and PRR9 act sequentially to repress CCA1 and LHY expression throughout the day [17]. While CCA1 and LHY promote expression of the morning-phased genes PRR7 and PRR9 [16], the mode of regulation of the afternoon-phased gene PRR5 has not previously been reported. We now provide evidence that RVE8 promotes PRR5 expression and that PRR5 represses RVE8 expression (Figure 8), the same type of regulatory interactions previously reported for their homologs [16], [17]. However, the delay in accumulation of RVE8 protein relative to its transcript likely accounts for the delayed phase of PRR5 accumulation compared to PRR7 and PRR9. The post-transcriptional mechanisms controlling RVE8 protein accumulation may therefore be key to proper functioning of the clock network.
All of the CCA1/LHY/RVE transcription factors that have been characterized to date have similar in vitro binding affinity for the EE [13], [27], [28], leaving open the question as to why their mutant phenotypes are so different. Inspection of the AtGenExpress developmental data set shows that individual family members are expressed throughout plant development and in most organs. In general, expression levels are lower in roots than in aerial tissues and expression is highest in germinating seedlings, flowers, and developing siliques [73]. Since there are no obvious differences in spatial and developmental expression patterns, their contrasting functions in the circadian system are likely due primarily to post-transcriptional differences. Our data suggest at least two such mechanisms are crucial: differential regulation of protein accumulation and differential association of co-regulators at target promoters. The study of these regulatory mechanisms will now be a high priority as we begin to resolve the disparate circadian functions of this fascinating group of transcription factors.
Whole-cell extracts were generated from Arabidopsis plants as previously described [27]. 5′-biotinylated oligonucleotides containing four tandem repeats of the EE_wt sequence (AAAATATCT) or EE_mt sequence (AAAATcgag) were purchased from Sigma (St. Louis, MO) and annealed together to obtain double stranded DNA (dsDNA). 18 mg of DYNAL®M-280 Streptavidin (Invitrogen) beads (1.8 ml) were washed twice with wash buffer [10 mM Tris-HCl (pH 7.5), 1 mM EDTA and 2.0 M NaCl] and then incubated with 50 pmoles of biotinylated dsDNA/mg DYNAL beads for 30 min at room temperature in incubation buffer [5 mM Tris-HCl (pH 7.5), 0.5 mM EDTA and 1.0 M NaCl]. Strepatividin beads bound to biotinylated DNA were washed twice with incubation buffer and stored at 4°C until further use. The beads were washed twice with reaction buffer [20 mM HEPES, pH 7.2, 80 mM KCl, 10% glycerol, 0.1 mM EDTA, 8 ng/µl of poly (dI-dC)] and incubated with whole cell extracts (∼100 mg total protein at 20 mg/ml, which was extracted from ∼250 g of Arabidopsis tissue) for 30 min at 4°C with gentle shaking. Beads were then washed twice with reaction buffer containing 0.5 mg/ml of salmon sperm DNA (ssDNA), followed by three washes with 100 mM ammonium bicarbonate and one wash with 50 mM ammonium bicarbonate solution. The beads were stored at 4°C in 50 mM ammonium bicarbonate solution until further use. Protein bound to the beads was digested by addition of 10 µl of diluted sequencing-grade Promega trypsin (13 ng/ul) (Promega, Madison, WI) and incubation at 37°C for 6 hr. Supernatant containing the digested peptides were removed and acidified with trifluoroacetic acid to a final concentration of 0.1%. The samples were stored at −80°C and were analyzed by LC-MS/MS.
Nano LC-MS/MS analysis was performed on a LTQ linear ion trap mass spectrometer (Thermo-Scientific, San Jose, CA), equipped with a Picoview nanospray source (New Objective, Woburn, MA), and an Eksigent nano 2d HPLC and autosampler (Eksigent, Dublin, CA). The tryptic peptide mixture was separated on a 75 µm ID PicoFrit column packed in-house with Magic C18AQ (Michrom BioResources, Auburn, CA) to a length of 15 cm with a 100% MeOH slurry of C18 reversed-phase material (100 Å pore size, 3 µm particle size) using a high-pressure cell pressurized with helium. The column was pre-equilibrated for 10 min at 2% solvent B [0.1% (v/v) formic acid in acetonitrile] and 98% solvent A [0.1% (v/v) formic acid in water] at a flow rate of 300 nL/min. Separation was achieved using a linear gradient from 2 to 60% solvent B in 45 min at a flow rate of 300 nL/min. The LTQ mass spectrometer was operated in the data dependent acquisition mode using a standard TOP10 method: 1 full-scan MS acquired was followed by 10 MS/MS scans.
Tandem mass spectra were extracted by Bioworks version 3.2. Charge state deconvolution and deisotoping were not performed. All MS/MS samples were analyzed using X! Tandem (The GPM, thegpm.org; version TORNADO (2010.01.01.4)). X! Tandem was set up to search the Uniprot Knowledge Base Arabidopsis thaliana complete proteome set database (Release 2010_08 July 12 2010, 31975 entries) and the cRAP common laboratory artifacts database (release 1.0, 112 entries) plus an equal number of reverse decoy sequences assuming the digestion enzyme trypsin. X! Tandem was searched with a fragment ion mass tolerance of 0.40 Da and a parent ion tolerance of 1.8 Da. Deamidation of asparagine and glutamine, oxidation of methionine and tryptophan, sulphone of methionine, tryptophan oxidation to formylkynurenin of tryptophan and acetylation of the n-terminus were specified in X! Tandem as variable modifications. Next, Scaffold (version Scaffold_3_00_03, Proteome Software Inc., Portland, OR) was used to validate MS/MS based peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 80.0% probability as specified by the Peptide Prophet algorithm [74]. Protein identifications were accepted if they could be established at greater than 95.0% probability and contained at least one identified peptide. Protein probabilities were assigned by the Protein Prophet algorithm [75]. Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. Using the above criteria the false discovery rate (FDR, Decoy/Target) was calculated as 0.1% on the peptide level and 2.4% on the protein level using a target-decoy (reverse) search strategy [76].
T-DNA insertion mutant rve8-1 (SALK_053482) was obtained from ABRC; PCR with RP 5′-AGTTTGCTGCTGATTTCTGAG-3′ and LP 5′-TTCAGCAAAATCAGGAACACC-3′ generates an approximately 1.1 kb band in Col but not in the mutant. RP with LBb1, 5′-GCGTGGACCGCTTGCTGCAACT-3′ gives a band of 700 bp in the mutant. Binary vector containing CCR2:: LUC+ [77] was transformed into rve8-1 and wild type (Col) by Agrobacterium mediated transformation [78]. Primary transformants were screened on MS medium containing 6.5 mg/ml gentamycin (EMD Chemicals) to select transformants. TOC1::LUC reporter [77] was introgressed into rve8-1 by crossing. For overexpression and rescue experiments, binary vectors containing either CaMV35S::HA-RVE8 or RVE8:: RVE8-HA were transformed by vacuum infiltration [78] into Col reporter lines also containing CCR2::LUC+. Binary vector containing PRR5::LUC2+ reporter was also transformed into Col and rve8-1 by Agrobacterium mediated transformation [78]. Primary transformants were selected on MS medium containing 150 µM basta (Chem Services, West Chester, PA).
To generate the overexpression binary vector containing CaMV35S::HA-RVE8, a 0.9 Kb RVE8 cDNA was PCR-amplified with Pfu ultra high fidelity enzyme (Stratagene) using primers F: 5′- CACCAGCTCGTCGCCGTCAAGAAATCC -3′ and R: 5′-TTGTTATGCTGATTTGTCGCTTGTTGAG -3′ and the plasmid U19901 obtained from Arabidopsis Biological Resource Center (ABRC) as a template. The PCR amplified product was cloned into pENTR/D-TOPO (Invitrogen) followed by LR recombination with pEarleyGate201. To generate the RVE8 rescue construct, a 2.5Kb genomic fragment containing ∼0.7Kb RVE8 upstream sequence was amplified from Arabidopsis genomic DNA using primers F: 5′- CACCTGTTTCGTAAGATTTGAATACAAAACCG -3′ and R: 5′- TGCTGATTTGTCGCTTGTTGAGTTC -3′ and Pfu ultra high fidelity enzyme (Stratagene) according to the manufacturer's protocol. The PCR amplified product was cloned into pENTR/D-TOPO (Invitrogen) followed by LR recombination with pEarleyGate301 to generate a binary vector containing RVE8::RVE8-HA. The 3kb region upstream of PRR5 was PCR-amplified using primers F 5′-CACCAGATTTTGTCACGCATCATTTTT-3′and R 5′-CAGCAAAATACTGTATACGAGACAAA-3′and using Col DNA as template. This fragment was cloned into pENTR/D-TOPO followed by LR recombination with pEARLYGATE 301-LUC2 [28] to generate a binary vector containing PRR5::LUC2. All clones were sequenced for any PCR-generated errors before being moved into Agrobacterium strain GV3101.
The full length cDNA of RVE8 was moved from the pENTR/D-TOPO vector to pDEST22 containing GAL4AD by LR recombination. Yeast one-hybrid assays were performed as previously described [28].
The His-RVE8 vector was generated by LR recombination of full length RVE8 cDNA in pENTR/D-TOPO with pDEST 15 (Invitrogen). Gel shift assays were performed as previously described [28].
Luciferase imaging was performed as previously described [28]. Seeds were entrained in 12 hour white light (50 mol m−2 sec−1, provided by cool white fluorescent bulbs)/12 hour dark cycles at 22°C for 6 days before being released into the indicated conditions for luciferase activity analysis, using either an ORCA II ER (Hamamatsu) or a DU434-BV (Andor Technology) CCD camera. Illumination was provided by red and/or blue LED Snap-Lites (Quantum Devices). Neutral density filters (RoscoLux no. 98 and no. 398) were used to obtain specific fluence levels for the fluence rate response curves. Images were analyzed using MetaMorph (Molecular Devices) and rhythms were estimated by Fourier Fast Transform-Non-Linear Least Squares [79].
For hypocotyl measurements, seeds were stratified at 4°C in the dark for 48 hours and sown on MS medium (0.8% agar and 3% sucrose) in petri plates. The plates were treated with white light (55 µmol m-2s-1) for 6 hours, and then either kept either in darkness or grown either under constant white light or in 8 hour light/16 hours dark cycles using neutral density filters (RoscoLux no. 98and no. 398) to obtain the indicated fluence rates. After 6 days of growth, seedlings were transferred to transparencies and scanned. Individual hypocotyl was measured using the application ImageJ (http://rsb.info.nih.gov/ij). For flowering time analysis, seeds were soaked in 0.1% agar in the dark at 4°C for 3 days and then sown in soil. Plants were grown either in short day (8 hours light/16 hours dark) or long day (16 hrs light/8 hours dark) conditions at 22°C, and monitored daily for bolting.
Plants were grown, RNA isolated, and qRT-PCR performed as previously described [80], with the following modifications. Seedlings were entrained in 12 hour white light (50 mol m−2 sec−1, provided by cool white fluorescent bulbs)/12 hour dark cycles at 22°C for 6 days before being released into either constant white light (50 mol m−2 sec−1) provided by cool white fluorescent bulbs or constant red light (35 mol m−2 sec−1) provided by red LED Snap-Lites (Quantum Devices). Samples were run in triplicate using iQ5 (Bio-Rad), and starting quantity was estimated from critical thresholds using the standard curve method. Data for each sample were normalized to the respective PROTEIN PHOSPHATASE 2a (PP2a) expression level. Primer sets used for amplification of CCA1, LHY, TOC1, PRR5, PRR7, and PP2a have been described [80], [81]. RVE8 mRNA was amplified using primers F 5′- GGGAAGCTCAAGCCGAACAGTATC-3′ and R 5′- GGCCTCTCGTTTCAGGATCAAAGA-3′, which flank the T-DNA insertion site in rve8-1.
For each time point, approximately forty RVE8::RVE8-HA or 35S::RVE8-HA seedlings were collected, frozen in liquid nitrogen and stored at −80°C until analysis. Plant tissue was ground in homogenization buffer (25 mM MOPS (pH 7.8), 0.25 M sucrose, 0.1 mM MgCl2, Complete EDTA-free protease-inhibitor cocktail (Roche) at 4°C. Protein concentrations of total cell extracts were then determined by Bradford assay (Bio-rad). 50 µg of each sample was analyzed by immunoblotting, using anti-HA-antibody conjugated to peroxidase (Roche, 3F10), or anti-UGPase antibody (AS05086, AgriSera, Vännäs, Sweden) followed by a secondary antibody, goat anti-rabbit IgG-HRP (1858415, Pierce). ECL Plus reagent (GE Healthcare) was used to generate chemiluminescence which was then detected with BioMax Light Film (Kodak). Data was quantified using ImageQuant software (GE Healthcare).
ChIP on wild type or plants expressing 35S::HA-RVE8 or RVE8::RVE8-HA was carried out as previously described [82]. Plants were entrained in 12 hour white light (50 mol m−2 sec−1, provided by cool white fluorescent bulbs)/12 hour dark cycles at 22°C for 6 days before being released into constant red light (35 mol m−2 sec−1) provided by red LED Snap-Lites (Quantum Devices). Seedlings were harvested at ZT48 (subjective dawn) or ZT32 (subjective afternoon). Immunoprecipitation was carried out using either an anti-HA antibody (Sigma, catalog #SAB4300603) or an anti-GST antibody (Santa Cruz Biotechnology, catalog #sc-459) as a negative control. Primers used to amplify the region -468/-345 basepairs upstream of the predicted PRR5 translational start site (containing one EE) were: (F) 5′-TGCAAACCTATGTACCAAACAGA-3′ and (R) 5′-TCCCACTCGTGACTTT-3′. Primers used to amplify the region -881/-701 basepairs upstream of the predicted TOC1 translational start site (containing one EE) were: (F): 5′-TGGTTTGGTCTGATCTGGTCAT-3′ and (R): 5′-AGGCCACGTCATCTTGGAGAAA-3′. Primers used to amplify the region -915/-765 basepairs upstream of the predicted PRR7 translational start site (containing three CCA1 binding sites) were: (F): 5′-CACGTGTAATGGTGGGTAAGG-3′ and (R): 5′-TGGGTTAAAATCTTTTTGAATGG-3′. The primer set used to amplify the UBQ10 locus as a negative control has been previously described [14].
The mass spectrometry data associated with this manuscript may be downloaded from the ProteomeCommons.org Tranche network using the following hash:
4fTHRVBPxyFD+GzvduEyt/sCUPO+bIIc4aNGUl1EUNGIaTr0jgZdcpdX5Ivu19clTLIQiHcaUDNICymEkeEGuhyPP+YAAAAAAAAL8A = =
The hash may be used to prove exactly what files were published as part of this manuscript's dataset, and the hash may also be used to check that the data has not changed since publication.
|
10.1371/journal.pcbi.1005198 | Control of Gene Expression by RNA Binding Protein Action on Alternative Translation Initiation Sites | Transcript levels do not faithfully predict protein levels, due to post-transcriptional regulation of gene expression mediated by RNA binding proteins (RBPs) and non-coding RNAs. We developed a multivariate linear regression model integrating RBP levels and predicted RBP-mRNA regulatory interactions from matched transcript and protein datasets. RBPs significantly improved the accuracy in predicting protein abundance of a portion of the total modeled mRNAs in three panels of tissues and cells and for different methods employed in the detection of mRNA and protein. The presence of upstream translation initiation sites (uTISs) at the mRNA 5’ untranslated regions was strongly associated with improvement in predictive accuracy. On the basis of these observations, we propose that the recently discovered widespread uTISs in the human genome can be a previously unappreciated substrate of translational control mediated by RBPs.
| Gene expression is a dynamic program by which the information stored in the genome is rendered functional by production and degradation of two types of macromolecules, RNAs and proteins. mRNAs are templates for proteins; therefore we expect correspondence between quantities of mRNAs and proteins. Genome-wide studies instead indicate a marked discrepancy between them, when considering their steady-state levels or their variations across different conditions. We employed linear regression approaches with paired mRNA/protein datasets in order to develop a model predicting the protein level of a gene from both the mRNA level and the protein levels of RBPs inferred to bind the mRNA untranslated regions. The results of our analyses restricted the utility of RBPs to improve accuracy of predicted protein abundance to a small fraction of the total modelled genes, and identified a novel association of the improvement induced by RBPs with the presence of upstream translation sites. This finding suggests a new avenue of experimental studies aimed at exploring the hypothesis that RBPs could influence protein abundance by changing the preference for certain translation initiation sites.
| High throughput technologies such as RNA-sequencing (RNA-seq) and mass-spectrometry-based protein analyses provide transcriptomic and proteomic profiles, which are the basis to draft a comprehensive picture of gene expression regulation [1],[2],[3].
Several studies have reported a lack of concordance between transcriptome and the proteome profiles [3],[4],[5],[6],[7], both at the steady state [8],[9],[10] and dynamically [11],[12],[13]. Even though this phenomenon is partially accounted for by technical factors such as noise [14], biased detection [15] and limited and variable coverage of mRNA and protein measurements [16], the discrepancy is so considerable that undoubtedly it implies an unresolved complexity in the regulation of gene expression downstream of transcription. Several studies have sought to examine the extent to which specific levels of regulation contribute to determine protein abundance at the steady state [17],[13],[8]. It was initially estimated that in mouse fibroblasts transcription explains 34% of variance in protein abundance, mRNA degradation 6%, translation 55% and protein degradation 5% [8]. Employing additional statistical efforts to account for the influence of measurement error on mRNA/protein correlation, recent studies proposed a correction of the initial estimates and brought back the role of translation to 30% [18]. Several studies highlighted the strong influence of translation on differential protein abundance during dynamic responses [19], [20], [21], [22].
The regulatory mechanisms by which the various post-transcriptional processes exert their effects on protein abundance are not well understood. Regulatory features associated with these processes have been identified not only in the coding regions but also in the 5’ and 3’ untranslated regions (UTRs) of mRNAs in multiple species [23],[24]. After their synthesis, processing, and export to the cytoplasm, mRNAs are broadly engaged in two activities: they may serve as templates for translation or as substrates for degradation pathways. Translational control, principally involving the initiation stage, can occur on a global basis by changes in the amounts and activation state of components of the translational machinery: translation factors [25], tRNAs [26] and ribosomes [27],[28]. Transcript-specific control of translation is less understood. The mechanisms of selective translation through recognition of target mRNAs by trans-acting factors, such as non-coding RNAs [28],[29],[30] and RNA-binding proteins (RBPs) [31],[32], are still subject of investigation [33],[34],[35],[36],[37], and are known only in a limited number of cases [38],[39],[40],[27],[41],[42].
Here, we developed a model of post-transcriptional control of gene expression by using multivariate linear regression to estimate protein levels from transcript levels. The model is empirically developed from two types of primary data: quantitative transcriptome assays matched with proteome assays, and post-transcriptional regulatory annotations of mRNA untranslated regions (UTRs) obtained by scanning for occurrences of in vitro experimentally determined RBP binding sites [31]. Including RBP levels and binding sites resulted in a statistically significant improvement of accuracy in protein abundance estimates of a fraction of the total modeled mRNAs in three panels of tissues and cells. We showed this improvement to be associated with the presence of upstream translation initiation sites (uTISs). This observation suggests the possibility that RBP could influence protein abundance by modulating alternative translation initiation, a mechanism of translational control still not experimentally described.
To devise a model of protein levels from transcript levels including a quantitative description of the contribution of RBP-mediated post-transcriptional control, we selected three data panels consisting of matched transcript and protein profiles: twelve normal human tissues [43], 59 cancer cell lines (the NCI-60 panel) [44], and 87 colorectal cancer tissues (the CPTAC CRC panel) [45]. The normal tissue panel contains the widest physiological variability, therefore it was used for determining model predictiveness. The NCI-60 and CPTAC CRC panels were used to show repeatability of the major findings in independent panels, and to assess cross-panel transferability of protein abundance models.
The depth of proteome coverage in the normal tissue panel was substantially lower than of the transcriptome (Table A in S1 Text), confirming previous reports [46]. We avoided genes whose transcripts and proteins were not reliably measured in a substantial number of samples in each panel (S1 Fig). This filtering resulted in the selection of more highly abundant genes than the overall pool at either the mRNA or protein level (S2 Fig). This effect was expected, considering the low frequency at which lowly abundant peptides could be selected for peptide sequence analysis and subsequent protein quantification. Filtering for adequately measured proteins introduced a bias in the genes we were able to study, highlighted by depletion and enrichment of several Gene Ontology (GO) categories (S3 Fig). The NCI-60 and CPTAC CRC panels also showed partial proteome coverages (Table A in S1 Text, S1 Fig), and consequent biases (S2 Fig, S3 Fig).
When measuring gene expression, multiple biological and technical factors can interact to produce the variability in average mRNA/protein levels, which we observed across the samples of each panel (S4 Fig, S5 Fig). To eliminate the possibility that average protein levels could help in predicting protein abundance (S6 Fig), mRNA and protein data were mean-centred per sample in each panel (Supporting Information). No sample turned out to be systematically associated with outlier measurements in any of the three panels (Supporting Information, S7 Fig).
We used RNA-binding motifs in linear regression modeling to infer models of RBP post-transcriptional regulation for all genes where transcripts, proteins, and RBPs were measured in a sufficient number of samples in a panel. The compendium of RNA-binding motifs was derived for 85 human RBPs by RNAcompete [29], an in vitro method for rapid and systematic analysis of RNA sequence preferences of RBPs shown to be predictive of in vivo binding [47]. We scanned the 5’ and 3’ UTRs of the mRNAs to identify sequences matching to the RNA-binding motifs, and detected RBP binding sites for 50 RBPs within the 5’ and 3’ UTRs of the 1,109 genes modeled in the normal tissue panel (q < 0.20). For genes modeled in the NCI-60 panel we identified binding sites for 40 RBPs on 1,327 mRNAs; in the CPTAC CRC panel for 66 RBPs on 1,825 mRNAs. The inferred RBP-mRNA interactions confirmed the previously reported tendency of multiple mRNAs to be regulated by multiple RBPs [47],[48],[49], with the number of RBPs per mRNA ranging from 1 to 38 based on inferred RBP binding sites in mRNA UTRs. This observation was independent of the stringency in statistical significance used for predicting RBP binding sites (S8 Fig).
We assessed the accuracy of the RBP-inclusive models to predict the protein abundance of modelled mRNAs by cross-validation and cross-panel validation. Finally, the relevance of RBPs in transcript/protein coupling was tested for association with regulatory features of the modelled mRNAs.
We then explored the features associated with the improvement in accuracy of protein abundance prediction achieved by the RBPplus model over the RNAonly one, as quantified by the difference in their R2 values (R2RBPplus–R2RNAonly). For this purpose, we analysed the association of the improvement in predictive accuracy with the major gene-specific sequence and structure annotations of the genes modelled in the normal tissue panel. We considered annotations which have been associated with post-transcriptional regulation of protein abundance [17],[10],[51], and which can be loosely classified by their demonstrated impact mostly on transcript stability and/or translation efficiency (Table 1). Spearman’s correlation with most of the tested characteristics was very low (Table 1). Interestingly, the only statistically significant correlation was observed between the improvement in accuracy of predicted protein abundance and the number of upstream Translation Initiation Sites (uTISs), as shown in Fig 4A.
Our results indicate that the presence of uTISs is a common feature of those mRNAs where RBPs included in the RBPplus model improved predictive accuracy compared to the RNAonly model. Even if this association does not mean a biological link between RBPs and uTISs, it suggests that translational regulation of the main ORF could be exerted by some of the considered RBPs through an uTISs. A potential, direct mechanism for this regulation could be steric control of uTIS elements by local RBP binding. We adopted this hypothesis to attempt an initial prioritization of RBPs. In case of steric control, RBP binding sites need to be in the proximity of a uTIS. No demonstrated example of such a control is present, at the best of our knowledge, in the literature. A functional proximity between uTISs and RBP binding sites has been reported only in one study involving the Drosophila SXL protein, but in this case the uTIS defines a uORF [61]. We selected the closest RBP binding site to each uTIS identified in a gene where the RBPs in the RBPplus model improved the accuracy in predicted protein abundance relative to the RNAonly model (p < 0.05 by randomization of proteins). We then ordered the RBPs according to the proportion of genes where they were inferred to recognize the binding sites located nearest to the uTISs. This analysis led us to prioritize the 15 RBPs inferred to bind the identified mRNAs (Fig 5). Of them, PCBP2 has been previously implicated in translational control by an internal ribosomal entry site (IRES) [62].
Although transcriptomic and proteomic assays are rarely integrated in large-scale studies, such integration provides a still unexploited instrument to study post-transcriptional control in a large-scale perspective. We performed an integrative analysis of matched RNAseq-based transcript and MS-based protein profiles to assess potential interaction between RBPs and mRNAs to determine protein abundance, beyond the contribution of transcript abundance. The pool of adequately measured proteins, as expected, was a fraction of the transcriptome coverage and was functionally biased for certain GO themes. RNAonly and RBPplus model were fitted for each mRNA/protein pair employing linear regression. To define the extent to which the RBPplus model improves the accuracy in predicted protein abundance over the RNAonly model, we harmonized our regression approaches for the RNAonly and RBPplus models, so that if RBPs are useless covariates, the RBPplus model is expected to converge to RNAonly one. We carefully checked the extent to which the effect produced by the RBPs, which were inferred to bind the modelled mRNAs, can be recapitulated by randomly sampled predictors, assessing statistical significance of improvement in predictive accuracy by empirical randomization tests. Our analysis suggested a large room for improvement over the RNAonly models, but the improvement in accuracy of predicted protein abundance achieved by the RBPs included in the RBPplus models could be reconstructed by randomly sampled proteins in the largest majority of the genes that we could model. Indeed, gene-level randomization tests identified a small fraction of genes where the impact of inferred RBP-mRNA interaction on improved predictive accuracy was statistically significant. Measuring the association of the improvement in accuracy of predicted protein abundance with mRNA features led to identify uTISs as a common feature of the genes where RBPs were shown to be informative.
Recently, allele-specific translational efficiency in an F1 hybrid mouse was determined by transcriptome and polysome profiling, and an analysis of sequence features of mouse genes with biased allelic translation revealed that out-of-frame uTISs could affect translational efficiency [63]. The impact of RBPs on the improvement in accuracy of predicted protein abundance was limited to a fraction of mRNAs, and it was dependent on the number of uTISs present in mRNA 5’ UTRs but not on the strength of the downstream aTISs. Our analysis cannot provide for a potential mechanism or decide for a direct versus an indirect effect, but given these features one of the possibilities is that some of the informative RBPs could modulate translation initiation of the downstream ORFs by simply either repressing or promoting alternative, uTIS-based, translation initiation.
Regulation of translation initiation in mammalian cells by interaction of RBPs with mRNA 5’ UTRs has been rarely documented, with a few examples involving the interaction between RBPs and internal ribosomal entry sites (IRESs) of specific stress-related mRNAs (reviewed in [64]), or the interaction between the IRP-1 RBP and the iron-responsive element (IRE) of the ferritin mRNA [35]. But no uTIS-dependent effect has been found in these well-studied cases. The presence of uORFs is known to regulate translation of primary downstream ORFs by operating via decay, re-initiation, or peptide-mediated ribosomal stalling during uORF translation [55],[65],[25]. Although uORFs can regulate protein levels without involving RBPs [65], an already cited previous study in Drosophila offers an example where the SXL RBP promotes translation initiation at the uORF of the msl-2 and Irr47 transcripts [61], which thus results in translational repression. More recently, the DENR-MCT-1 complex has been identified as a regulator of eukaryotic uORF-dependent translation re-initiation of a specific group of mRNAs [34]. But to our knowledge no RBP-induced, non uORF-mediated translational control mechanism in uTIS-endowed loci has yet been identified.
Based on the hypothesis of a direct mechanism of RBP control of uTISs, such mechanism could be sensitive to changes in the position and spacing between RBP binding sites within the mRNA 5’ UTR. We therefore used the criterion of spatial proximity to prioritize the RBPs which were shown to help in predicting protein abundance. Of course, we cannot exclude that this control could be due to 3’ UTR binding, considering also that 3’ UTR-acted and RBP-mediated translational initiation controls are an established model. Yet, this model has been proposed [25] on the basis of few notable cases in Drosophila [66],[67],[68],[69] and Xenopus [70] translational control always during development and differentiation. In these examples, the RBP ensures specificity to the regulation of translation by binding sites within the 3’ UTR of the mRNA and contributes to the formation of a closed loop which precludes formation of the initiation complex eIF4F, therefore exerting an inhibitory effect on translation. It is worth considering that, again, in this “classical” model no role is attributed to uTISs.
With the limitations highlighted in mind, the study presented here allowed us to estimate the impact of RBP-mRNA interactions on quantitative relationships between mRNA and protein abundances. RBPs were shown to help in predicting protein abundance relative to an RNAonly model, but not relative to randomly selected proteins, in the majority of considered mRNAs. Nonetheless, our analysis identified genes for which inferred RBP-mRNA interactions were informative. The association between the improvement in accuracy of predicted protein abundance and uTISs suggests that RBPs could modulate the expression of these genes by mediating alternative translation regulation. The usefulness of RBPplus models need to be further tested as soon as suitable datasets are produced by RNAseq and MS-based technologies. The pervasive presence of conserved uTISs in the human transcriptome, which has been recently revealed by ribosome profiling and related approaches [52],[62],[63], awaits a clarification of their functional role.
Matched transcriptome and proteome profiles were downloaded in the processed form provided by three independent datasets: 1) a panel of twelve human normal tissues [43], 2) the 59 samples from the US National Cancer Institute (NCI)-60 dataset [44], and 3) 87 colorectal cancer (CRC) samples profiled by The Cancer Genome Atlas (TCGA) in combination with the Clinical Proteomic Tumour Analysis Consortium (CPTAC) [45].
Processed data derived from gene expression analysis in the normal tissue panel were downloaded from the online Supplementary Information of the study [43]. Normalized transcriptome data for NCI-60 cell lines were obtained from the Gene Expression Omnibus (series accession number GSE32474), while processed proteome data were downloaded from http://wzw.tum.de/proteomics/nci60. Processed proteome data for TCGA colorectal cancer samples were downloaded from the online Supplementary Information of the study [45], while processed transcriptome data were downloaded from TCGA (http://cancergenome.nih.gov/).
In the normal tissue panel and CRC panel, transcript abundance data were obtained by RNA sequencing (RNA-seq) and expressed as Fragments Per Kilobase per Million, log-base-10 FPKM. NCI-60 transcriptome profiles were obtained by microarray. Normal tissue proteome profiles were obtained by the intensity-based Absolute protein Quantification method, and expressed as log-base-10 iBAQ. NCI-60 and CPTAC CRC proteome profiles were based on liquid chromatography-tandem mass spectrometry (LC-MS/MS)-based shotgun proteomic analysis. Intensity- and spectral count-based label-free quantifications were used to obtain protein abundance in the NCI-60 cell lines and in the CPTAC CRC specimens, respectively.
In the normal tissue panel we excluded genes and proteins below the detection limit in more than three out of twelve tissues at either transcriptome or proteome level; in the NCI-60 and CRC panels we excluded genes below the detection limit in more than five out the total number of specimens at either the transcriptome or proteome level. Genes below the detection limit were assigned zero values or Not Available (NA) labels in the files processed data were acquired from.
Within each panel, we applied inter-sample normalization by mRNA and protein mean-centring per sample.
We scanned non-redundant 5’ and 3’ untranslated region (UTR) sequences of the genes profiled at both the transcript and protein levels with positional weight matrices (PWMs), which represent RNA sequence binding specificities of RBPs derived from RNAcompete [31] data and which are available through the cisBP-RNA database (http://cisbp-rna.ccbr.utoronto.ca/). In each panel of matched transcriptome/proteome datasets, the inference of RBP binding sites in mRNA UTRs was restricted to the subset of RBPs which were detected both at the transcript and at the protein level. For each considered RBP, RBP binding sites as well as corresponding q-values were obtained using the FIMO algorithm [71] of the MEME toolkit (http://meme-suite.org/) and retained at the false discovery rate of 20%.
We built two models for each considered gene: a basic (RNAonly) model, where the abundance of protein j in sample i (PROTij) was predicted by the corresponding mRNA level only in a simple linear regression model:
RNAonly:PROTij=β0j+βmRNA,jmRNAij+εij;εij∼iidN(0,σi)
where β0j is the intercept term, βmRNA,j is the regression coefficient for the mRNA predictor and the error term εij is an independent and identically distributed (iid) random variable following a normal distribution of mean 0 and standard deviation σ. This model was fit for each mRNA/protein pair. An RBP-inclusive (RBPplus) multiple linear regression model was also fitted for each mRNA/protein pair:
RBPplus:PROTij=β0j+βmRNA,jmRNAij+βRBP,jkRBPijk+εij;εij∼iidN(0,σi)
where βRBP,jk is the regression coefficient for the kth RBP of mRNA j. This model was fitted by maximum penalized likelihood with Ridge or LASSO penalty applied to RBPs but not to mRNA measurements, using the pensim R package [72], which acts as a wrapper providing nested cross-validation to the penalized R package [73]. In the nested cross-validation scheme, test samples are held out for accuracy estimation in the outer layer of cross-validation, and penalty parameters are tuned in the inner layer of cross-validation within training samples only. In the outer layer of cross-validation, we used 5-fold for the three panels.
By not penalizing mRNA measurements, the model can be expected to converge to the RNAonly model in the absence of informative RBP protein measurements. Both Ridge and LASSO penalty help control of over-fitting of high-dimensional data; LASSO additionally provides feature selection by setting the coefficients of most covariates to exactly zero.
We fitted these two models, independently for each gene inferred to be bound by an RBP and in each tissue/cell panel where both transcript and corresponding protein met the missingness requirements described above.
The change in accuracy of predicted protein abundance obtained by the RBPplus model relative to the RNAonly model of each considered gene was quantified by the difference in the R2 coefficients between the RBPplus and RNAonly models. This analysis used the following R2 coefficient definition:
R2=1−∑i(yi−fi)2∑i(yi−〈y〉)2
where yi the i-th observation, <y> is the mean of the observations, and fi is the i-th prediction.
We evaluated the statistical significance of the improvement in accuracy of predicted protein abundance attained by the RBPplus model relative to the RNAonly model across the genes considered in each separate panel by Wilcoxon signed-rank test.
Empirical randomization tests were used to determine whether the accuracy in predicted protein abundance achieved by the RBPplus model of an individual gene was statistically significantly better relative than that expected for randomized RBPplus models of the gene. For each considered gene, we obtained 1000 randomized versions of the RBPplus model by 1) randomly sampling a number of protein predictors equal to the number of actual RBPs inferred to bind the mRNA UTRs, and 2) by permuting the protein levels of inferred RBPs across samples. As the actual RBPplus models, each randomized RBPplus model (by sample permutation or randomly sampling of protein predictors) was fitted by maximum penalized likelihood with Ridge penalty applied to RBPs but not to mRNA measurements in nested cross-validation scheme. Fig 3 illustrates the two randomization schemes.
The p-value of the R2 value observed for the actual RBPplus model of each considered gene was defined by the probability of sampling a R2 value from the null distribution of R2 values that is higher than the observed R2. The RBPplus model of a gene was deemed to improve the accuracy in prediction of protein abundance if the RBPplus model accuracy was higher than that of the RNAonly model and if the probability of attaining accuracy higher than that of the RBPplus model by randomly sampling protein predictors was < 0.05. Since our analysis involved multiple hypotheses testing, we reported false discovery rate by the Storey’s q-value method implemented in the qvalue R package [50].
We studied cross-panel model transferability of models trained using only the RBPs profiled in both of each pair of panels. For each considered mRNA, we developed the RNAonly and RBPplus models using all samples in a training panel, and tested them using all samples in the testing panel. The procedure was repeated for all possible combinations of training and test panels. We estimated model transferability computing Spearman’s correlation coefficient of protein predictive accuracies between the RBPplus models trained in a chosen panel and the RBPplus models trained in each of the other two panels.
Functional enrichment/depletion analysis was based on the Biological Process categories of the generic Gene Ontology (GO) slim, a cut-down version of the Gene Ontology annotations (http://geneontology.org/) and used hypergeometric test. Functional analysis was used 1) to assess over-/under-representation of GO themes in the genes which turned out to be adequately measured relative to the total of genes which were profiled at the mRNA/protein levels, and 2) to assess over-/under-representation of GO themes in the genes where RBPplus models were found to be informative relative to the total of modelled genes.
We surveyed appropriate data sources to gather several gene annotations relevant to post-transcriptional regulation of gene expression in mammalian cells (Table 1). We quantified the selected features in the mRNAs modelled in the normal tissue panel as follows. Normalized lengths of the coding sequence as well as of the 5’ and 3’ UTRs were calculated for each mRNA according to the sequence annotations (hg38 assembly) available at the UCSC Genome Browser (https://genome.ucsc.edu/). Local folding energy was computed within a window of 30 nucleotides upstream and downstream of the annotated translation initiation site of the modelled mRNAs using the RNAfold algorithm of the Vienna RNA package (www.tbi.univie.ac.at/RNA/). Transcript half-life measures were acquired by two distinct studies which relied, respectively, on biosynthetic labelling of newly transcribed RNA and estimation of newly/total RNA ratio in human B cells [74], and on transcription blocking in HepG2 and Bud8 cell lines [75]. A measure of efficiency of start codon recognition of primary ORFs was derived from a quantitative analysis of translation initiation sites by FACS-seq, Fluorescence-Activated Cell Sorting and high-throughput DNA sequencing [52]. The tRNA adaptation index (tAI), an estimate of the translational optimality of a coding sequence to cellular tRNA pools was computed by the codonR software [76]. Annotation of upstream translation initiation sites (uTISs) was derived by Global Translation Initiation sequencing (GTI-seq) in HEK293 cells and downloaded from the TISdb database [77]. We included an additional mapping of upstream translation initiation sites which was obtained by Quantitative Translation Initiation sequencing (QTI-seq) in HEK293 cells [58].
Upstream Open Reading Frames (uORFs) were defined by: (i) an uTIS out-of-frame at the 5’ UTR, with a stop-codon, in the same frame, downstream of it, and with a minimal length of nine nucleotides, (ii) an uTIS in-frame at the 5’UTR with a stop codon in frame after the main stop codon or before the main start codon.
We used Spearman’s correlation coefficient to estimate the correlation of the change in accuracy of predicted protein abundance with each aforementioned feature. Furthermore, we used Fisher’s test to assess the enrichment of the genes where the RBPplus model was found to be informative in uTIS-containing genes as well as in uORF-containing genes. Testing was performed for uTISs identified by GTI-seq and QTI-seq technologies and for each panel of paired mRNA/protein datasets.
RBPs were prioritized by an analysis of the frequency at which the binding sites of an RBP occur in the proximity of uTISs of mRNAs. We identified the closest RBP binding site to each uTIS present in the 5’ UTR of each mRNA. We then quantified the frequency of the binding sites of each RBP in the binding sites situated nearest to the uTISs overall mRNAs. RBPs were ordered according to the number of genes where they were found to recognize the binding sites closest to the uTISs.
In the contexts where multiple tests were performed, raw P-values were adjusted by the Benjamini-Hochberg method for controlling false discovery rate at 5%.
|
10.1371/journal.pcbi.1001138 | deFuse: An Algorithm for Gene Fusion Discovery in Tumor RNA-Seq Data | Gene fusions created by somatic genomic rearrangements are known to play an important role in the onset and development of some cancers, such as lymphomas and sarcomas. RNA-Seq (whole transcriptome shotgun sequencing) is proving to be a useful tool for the discovery of novel gene fusions in cancer transcriptomes. However, algorithmic methods for the discovery of gene fusions using RNA-Seq data remain underdeveloped. We have developed deFuse, a novel computational method for fusion discovery in tumor RNA-Seq data. Unlike existing methods that use only unique best-hit alignments and consider only fusion boundaries at the ends of known exons, deFuse considers all alignments and all possible locations for fusion boundaries. As a result, deFuse is able to identify fusion sequences with demonstrably better sensitivity than previous approaches. To increase the specificity of our approach, we curated a list of 60 true positive and 61 true negative fusion sequences (as confirmed by RT-PCR), and have trained an adaboost classifier on 11 novel features of the sequence data. The resulting classifier has an estimated value of 0.91 for the area under the ROC curve. We have used deFuse to discover gene fusions in 40 ovarian tumor samples, one ovarian cancer cell line, and three sarcoma samples. We report herein the first gene fusions discovered in ovarian cancer. We conclude that gene fusions are not infrequent events in ovarian cancer and that these events have the potential to substantially alter the expression patterns of the genes involved; gene fusions should therefore be considered in efforts to comprehensively characterize the mutational profiles of ovarian cancer transcriptomes.
| Genome rearrangements and associated gene fusions are known to be important oncogenic events in some cancers. We have developed a novel computational method called deFuse for detecting gene fusions in RNA-Seq data and have applied it to the discovery of novel gene fusions in sarcoma and ovarian tumors. We assessed the accuracy of our method and found that deFuse produces substantially better sensitivity and specificity than two other published methods. We have also developed a set of 60 positive and 61 negative examples that will be useful for accurate identification of gene fusions in future RNA-Seq datasets. We have trained a classifier on 11 novel features of the 121 examples, and show that the classifier is able to accurately identify real gene fusions. The 45 gene fusions reported in this study represent the first ovarian cancer fusions reported, as well as novel sarcoma fusions. By examining the expression patterns of the affected genes, we find that many fusions are predicted to have functional consequences and thus merit experimental followup to determine their clinical relevance.
| Gene fusions are known to play an important role in the development of haematalogical disorders and childhood sarcomas, while the recent discovery of ETS gene fusions in prostate cancer [1] has also prompted renewed interest in gene fusions in solid tumors. ETS gene fusions are present in 80% of malignancies of the male genital organs, and as a result these fusions alone are associated with 16% of all cancer morbidity [2]. The discovery of the EML4-ALK fusion in non-small-cell lung cancer and the ETV6-NTRK3 fusion in human secretory breast carcinoma suggest that gene fusions are also recurrent at low levels in other solid tumor types [3], [4]. The discovery of such rare but recurrent gene fusions may be of significant clinical benefit where they provide the potential for targeted therapy.
Gene fusions are thought to arise predominantly from double stranded DNA breakages followed by a DNA repair error [2], [5]. Promoter exchanges are one class of gene fusions, characterized by the replacement of an oncogene's regulatory regions with those of another gene, resulting in deregulation of transcription of the oncogene. For ETS gene fusions in prostate cancer, the androgen-responsive regulatory elements of TMPRSS2 drive the expression of the ETS family member to which TMPRSS2 is fused [1]. Another class of gene fusions leads to the creation of a chimeric protein with biological function distinct from either of the partner genes from which it originated. A classic example is BCR-ABL1, a chimeric protein that is the defining lesion in chronic myelogenous leukaemia (CML), and which induces growth factor independence and the inhibition of apoptosis [6].
Large scale, genome-wide efforts to comprehensively identify and characterize genomic rearrangements that lead to gene fusions in human cancers have recently been made possible through next generation sequencing technologies. These technologies provide a deeper level of sequencing than is possible by cytogenetic and Sanger sequencing methods and are poised to reveal a more detailed understanding of the extent and nature of genomic rearrangements in cancer. For example, using low-coverage paired end whole genome (gDNA) shotgun sequencing, Stephens et al. [7] reported that the genomes of breast cancer cells harbour many more rearrangements than previously thought, and suggested that this class of somatic mutation needs to be carefully considered when interpreting breast cancer genomes. Using similar experimental and analytical techniques, Campbell et al. [8] profiled tumor evolution in pancreatic cancer patient samples by profiling the pattern of somatic rearrangements found in primary tumors and distant metastases extracted from the same patient.
Next generation sequencing of cDNA (RNA-Seq or whole transcriptome shotgun sequencing) provides an ideal experimental platform for expressed gene fusion discovery. Analogous to genome sequencing, RNA-Seq enables an unbiased and relatively comprehensive view into tumor transcriptomes, and can provide information about the rarest of transcripts. RNA-Seq targets only expressed sequences from protein coding genes and is thus more focused than whole genome sequencing. Maher et al. [9] demonstrated the capacity of RNA-Seq to find gene fusions in prostate cancer samples. They identified potentially fused gene pairs using discordantly aligned paired end reads, and also identified potential fusion splices by mining end sequences for alignments to all possible pairings of exons of the potentially fused gene pairs. A study of the melanoma transcriptome by Berger et al. (2010) used many of the same principles. Another recently developed method called FusionSeq identifies gene fusions from discordant alignments, and uses a variety of novel filters and quality metrics to discriminate real fusions from sequencing and alignment artifacts [10]. FusionSeq has been used to identify fusions in prostate tumor samples and cell lines [10], [11]. While the methods used for these studies are capable of identifying genuine gene fusions, many challenges and limitations remain in the analysis of RNA-Seq data. For example, the aforementioned studies only considered reads that align uniquely to the genome. However, errors in next generation sequencing together with homologous and repetitive sequences shared between genes often produce ambiguous alignments of the short reads generated in RNA-Seq experiments. While resolving the ‘correct’ placement of these reads is often not possible, we propose that ambiguously-aligning reads provide important evidence of real gene fusions, and therefore should be leveraged by analysis methods.
Sequence reads that align across a gene fusion boundary (so-called split reads) are a strong source of evidence for gene fusions in paired-end RNA-Seq data. Hu et al. [12] propose a strategy centered on the ability to identify split reads called PERAlign: the method uses split read aligner MapSplice [13] to identify single end reads split by fusion boundaries, and then verifies those fusion boundaries using a probabilistic model to infer the alignment of discordantly aligning pairs. As described below, our method also combines the complementary sources of split reads and discordant reads, but we show using real patient data that discordant read analysis followed by split read analysis is considerably more sensitive for gene fusion discovery than the reverse procedure described by Hu et al.
With the goal of resolving the limitations described above and therefore providing a more accurate method for detecting gene fusions from RNA-Seq, we developed a novel algorithm called deFuse. The central idea behind deFuse is to guide a dynamic programming-based split read analysis with discordant paired end alignments. This is in contrast to PERAlign, which uses discordant paired end alignments to verify the results of a split read analysis. Furthermore, unlike previous approaches, we do not discard paired end reads that align ambiguously, but instead consider all alignments for each read, and attempt to resolve the most likely alignment position for each read. We show that using ambiguously-aligning reads results in an increased amount of evidence for predicted gene fusions and an increase in the number of relevant gene fusions predicted. In addition, our method is not limited to finding gene fusions with boundaries between known exons, and therefore can identify fusion boundaries in the middle of exons or involving intronic or intergenic sequences. Finally, the method attempts to provide a number of confidence measures to estimate the validity of each prediction.
We obtained three sarcomas and 40 ovarian carcinomas from the OvCaRe (Ovarian Cancer Research) frozen tumor bank. Patients provided written informed consent for research using these tumor samples before undergoing surgery, and the consent form acknowledged that a loss of confidentiality could occur through the use of samples for research. Separate approval from the hospital's institutional review board was obtained to permit the use of these samples for RNA-sequencing experiments.
In this section, we describe the deFuse algorithm. We begin by defining essential terms. We define a fragment as a size selected cDNA sequence (usually approximately 250 bp) during RNA-Seq library construction. We define a read as a sequenced end of a fragment (usually approximately 50 bp). We define paired ends as the pair of reads sequenced from the ends of the same fragment. The insert sequence is the portion in the middle of the fragment that is not sequenced. A fusion boundary is the precise, nucleotide-level genomic coordinate that defines the breakpoint on either side of the gene fusion. We define spanning reads as paired ends that harbour a fusion boundary in the insert sequence, whereas a split read harbours a fusion boundary in the read itself. A discordant alignment is produced by spanning reads of a fragment with each end aligning to a different gene, whereas a split read will often produce a single end anchored alignment for which one end aligns to one gene and the other end does not align.
With these definitions in hand we will now describe how deFuse predicts gene fusions by searching RNA-Seq data for fragments that harbour fusion boundaries. As mentioned previously, the problem of identifying the true genomic origin of a set of RNA-Seq reads is confounded by several factors, and as a result, a proportion of the RNA-Seq reads will have ambiguous alignments to the genome. The deFuse method, outlined schematically in Figure 1, combines an approach for resolving the actual alignment location of ambiguously aligning spanning reads with a dynamic programming based split read analysis for resolving the nucleotide level fusion boundary with high sensitivity. A novel confidence measure is provided based on the degree of corroboration of evidence supporting the prediction.
The method consists of four main steps. The first step is alignment of paired end reads to a reference comprised of the sequences that are expected to exist in the sample, with all relevant alignments considered. We use spliced and unspliced gene sequences as a reference because we have found that fusion genes often produce splice variants that express intronic sequences, and that some of those splice variants are biologically relevant (unpublished data). We define two necessary conditions for considering discordant alignments to have originated from reads spanning the same fusion boundary and use these conditions to cluster discordant alignments representing the same fusion event. The second step resolves ambiguous discordant alignments by selecting the most likely set of fusion events, and the most likely assignment of spanning reads to those events (Figure 1a). The third step is a targeted search for split reads using a dynamic programming based solution to resolve the nucleotide level fusion boundary of each event (Figure 1b). The forth step involves a test for the corroboration of the spanning and split read evidence. For each spanning read, we calculate the putative length of the fragment that generated that paired end read given the fusion boundary predicted by the split reads. The resulting set of fragment lengths is used to test the hypothesis that those fragments were generated by the inferred fragment length distribution (Figure 1c). Finally, we compute a set of quantitative features, and use an adaboost classifier to discriminate between real gene fusions and artifacts of the sequencing and alignment process.
To classify paired end reads as concordant we aligned reads to spliced genes, the genome, and UniGene sequences using bowtie [19] in paired end mode. We also aligned reads to spliced and unspliced genes in single end mode with parameters −k 100 −m 100. We classified any paired end read as concordant if both ends aligned to the same gene, regardless of the location of the alignment in that gene. Paired end reads aligning with one or both ends to ribosomal RNA sequences were removed from the analysis as has been done previously [10].
Paired end reads not classified as concordant were classified as discordant. Single end mode alignments of discordant paired end reads were then classified as fully aligned or single end anchored. Fully aligned discordant paired end reads were clustered with and the maximum parsimony solution was found using the algorithm given above. Split alignments were generated using , , , (see Supplementary Methods, Text S1). Finally, a predicted fusion sequence was assembled that included the regions in each gene to which spanning reads aligned, joined together at the fusion boundary predicted by the split reads.
Predicted fusion sequences were annotated as open reading frame preserving, 5′ or 3′ UTR exchanges, interchromosomal, inversion, eversion, and between adjacent genes. The translational phase for each coding nucleotide was calculated using the frame column for each exon in the ensembl GTF file. Given nucleotide of 5′ gene with phase spliced to nucleotide of 3′ gene with phase , if , the fusion – is annotated as open reading frame preserving. Note this method would not detect open reading frames of novel proteins, only those of chimeric proteins that are combination of the protein sequences of the fused genes. Results were also annotated for their position (UTR, exonic, intronic, coding, upstream, downstream) within each gene.
We computed a set of features to better characterize our predicted fusions. The features were calculated for each fusion prediction with the aim of discriminating between true and false positives. We initially lacked a set of positive and negative controls that would have been necessary for a principled machine learning based classification method. Thus initial validation candidates were identified by thresholding these features at levels we suspected would enrich for real fusions (see Results). Validation was also attempted for suboptimal predictions and 40 randomly chosen predictions in order to establish a set of negative controls.
Once we had performed a significant number of validations, these validations became the training set for a classifier. We calculated the following 11 features for the examples in our training set (detailed descriptions in Supplementary Methods, Text S1):
Spanning read coverage Normalized spanning read coverage.
Split position p-value P-Value for the hypothesis that the split position statistic was calculated from split reads that are evenly distributed across the fusion boundary.
Minimum split anchor p-value P-Value for the hypothesis that the minimum split anchor statistic was calculated from split reads that are evenly distributed across the fusion boundary.
Corroboration p-value P-Value for the hypothesis that the lengths of reads spanning the fusion boundary were drawn from the fragment length distribution.
Concordant ratio Proportion of spanning reads supporting a fusion that have a concordant alignment using blat with default parameters.
Fusion boundary di-nucleotide entropy Di-nucleotide entropy calculated 40 nt upstream and downstream of the fusion boundary for the predicted sequence, taking the minimum of both values.
Fusion boundary homology Number of homologous nucleotides in each gene at the predicted fusion boundary.
cDNA adjusted percent identity Maximum adjusted percent identity for the alignments of the predicted sequence to any cDNA.
Genome adjusted percent identity Maximum adjusted percent identity for the alignments of the predicted sequence to the genome.
EST adjusted percent identity Maximum adjusted percent identity for the alignments of the predicted sequence to any EST.
EST island adjusted percent identity Maximum adjusted percent identity for the alignments of the predicted sequence to any EST island.
We then used the ada (2.0–2) package in R (2.11.0) to train an adaboost model using the stochastic gradient boosting algorithm with exponential loss, discrete boosting, and decision stumps as the base classifier [20]. We used conservative regularization (shrinkage parameter ) and permitted the algorithm 200 iterations. Adaboost was selected because it would enable us to leverage the weak predictive power of individual features, and would provide a straightforward way of evaluating the predictive power of each feature. Finally, the classifier was used to classify all predictions for our ovarian and sarcoma datasets.
deFuse is implemented in C++, perl and R. A typical library of 120,000,000 paired end reads completes in approximately 6 hours using a cluster of 100 compute nodes. The human genome (NCBI36) and gene models in GTF format (ensembl 54) were downloaded from Ensembl [21]. EST sequences and spliced EST alignments were downloaded from UCSC [22]. UniGene sequences were downloaded from NCBI [23]. This study focused on the 21295 genes annotated as protein_coding, processed_transcript, IG_C_gene, IG_D_gene, IG_J_gene, and IG_V_gene in the ensembl GTF file (see Table S10 for ensembl IDs for each gene). An unspliced gene sequence is composed of the genomic sequence starting 2 kb upstream of the most upstream exonic nucleotide of that gene's splice variants, to the genomic position 2 kb downstream of the most downstream exonic nucleotide of that gene's splice variants. A spliced gene sequence is composed of the concatenated sequences of each exon of a single splice variant of a gene. Our reference sequences were comprised of 21295 unspliced gene sequences and 46662 spliced gene sequences. All data files are available as part of the deFuse software package at http://compbio.bccrc.ca.
Fusion sequence predictions were obtained for the 44 datasets as detailed in Methods. This study only considered sequence predictions supported by five or more spanning reads and one or more split reads, though theoretically the limit on the number of spanning reads could be lowered for smaller datasets as was done for the melanoma datasets. The total number of unfiltered predictions at this stage numbered 20,327.
We evaluated the ability of deFuse to rediscover known gene fusions in publicly available RNA-Seq data. Using deFuse, we searched for the TMPRSS2-ERG fusion in the NCI-H660 prostate cell line dataset, the three fusions previously identified in the CML dataset (SRA accession: SRR018269) and the 11 fusions in melanoma libraries identified by Berger et al. [16]. Since seven of the fusions in the melanoma datasets are supported by fewer than five spanning reads, we altered the configuration of deFuse for the melanoma libraries such that only two spanning reads and one split read were required for deFuse to attempt assembly of a fusion boundary sequence. For all 15 fusions, deFuse was able to assemble the correct fusion boundary sequence. We evaluated the performance of deFuse using heuristic filters (deFuse-Thresholds), and deFuse using the adaboost classifier (deFuse-Classifier), when applied to the prostate, CML and melanoma datasets (Table 5). Since the training set for deFuse-Classifier includes the prostate, CML and melanoma fusions, we used a leave one out method classify each fusion.
deFuse-Thresholds identifies 7 of the 15 known fusions, whereas deFuse-Classifier identifies 10 of the 15 fusions. Notably, TMPRSS2-ERG is not included in the deFuse-thresholds or the deFuse-Classifier results, primarily because the TMPRSS2-ERG prediction is an outlier on both the Spanning read coverage feature and the Fusion boundary di-nucleotide entropy feature. deFuse-Classifier assigns a probability of 0.48 to the TMPRSS2-ERG prediction. A probability threshold of 0.48 (instead of 0.81 as calculated in “Classification of ovarian and sarcoma predictions”) would result in a true positive rate of 93%, and false positive rate of 14%. These numbers suggest that our initially selected probability threshold of 0.81 may have been overly conservative given that we could have increased our true positive rate by 11% at the expense of only a 4% increase in false positive rate. Using a threshold of 0.48, deFuse-Classifier would recover 13 of the 15 events including TMPRSS2-ERG.
We sought to understand each fusion's impact on the expression patterns of the fused genes. For a given fusion boundary , let be the expression of exons on the preserved side of , normalized by the length of those exons. Also let be the length normalized expression of the remaining exons, not predicted to be part of the fusion gene. We define the interrupted expression index as the ratio of the expression of preserved versus remaining exons, analogous to the splicing index [28]. For each PCR-validated fusion boundary predicted for an ovarian dataset we calculated for all ovarian datasets and compared for the dataset with the predicted fusion to for the datasets without the fusion using a Wilcoxon test [28], resulting in 22 fusion events with at least one partner predicted as interrupted (p-values 0.05, see Table 2 and Table S2).
Promoter exchanges are characterized by overexpression of the 3′ exons of a gene resulting from the replacement of 5′ regulatory regions [1]. For each PCR validated fusion we calculated whether the 3′ partner was expressed significantly higher in the dataset harbouring the fusion compared to other ovarian datasets (p-values 0.1, see Table S2, Table S6 and Table S7). We then overlapped the overexpression results with the interrupted expression results to find seven fusions representing potential promoter exchanges (Table 2). The remaining 15 expression-interrupting fusions represent either biallelic inactivations (for example, HNF1A described below) or dominant expression of the fusion allele (for example, RREB1-TFE3 described below).
We sought to rule out genomic amplification as a mechanism of overexpression for the seven putative promoter exchanges. Analysis of Affy SNP6.0 genome data indicates that two of the 3′ partners, SH2D1A and UMOD, are in regions of genomic amplification (Table S3). Given that UMOD is not expressed in any other ovarian library (Table S9), a genomic amplification alone cannot explain UMOD expression in HGS4. For the FRYL-SH2D1A fusion in HGS3, a marked coincidence between the fusion boundary and an expression changepoint implies that only the fused copy of SH2D1A is expressed (Figure 6). FISH evidence for FRYL-SH2D1A indicates that at most one copy of the FRYL-SH2D1A fusion exists in the genome of each tumor cell (Figure 6), suggesting that amplification of the SH2D1A region is not the underlying cause of SH2D1A overexpression. FRYL expression is on average 670 fold higher than SH2D1A expression in the non-HGS3 ovarian libraries (Table S6), implying that the FRYL promoter would overexpress SH2D1A, were it fused to SH2D1A. In HGS3, SH2D1A expression is on average 36 fold higher than in other ovarian libraries, supporting the theory that the FRYL promoter is driving SH2D1A expression. The FRYL-SH2D1A fusion does not preserve the open reading frame of SH2D1A. Investigation of the functional impact of FRYL-SH2D1A and the other six promoter exchanges is ongoing.
We sought to identify previously described rearrangements in our sarcoma and ovarian carcinoma data. Although generally considered a breast cancer rearrangement, amplification of ERBB2 has also been shown to occur in mucinous ovarian tumors [29]. In our ovarian cases, one mucinous tumor, MUC1, harbours a fusion between ERBB2 and adjacent PERLD1 caused by an underlying genomic inversion. CNV analysis of Affy SNP6.0 genome data predicted ERBB2 to be highly amplified in the genome of MUC1 (Table S3), and ERBB2 expression is approximately 10 fold higher in MUC1 than in any other ovarian library (Table S6). Since amplification of ERBB2 requires replication of ERBB2 across the genome, a reasonable explanation for the ERBB2-PERLD1 fusion is that it is a secondary effect of the process of ERBB2 amplification.
Analysis of the two epithelioid sarcomas and one intermediate grade myofibroblastic sarcoma produced five fusion predictions between non-adjacent genes, three involving genes previously described as translocated in cancer. The CMKLR1-HNF1A fusion is predicted to significantly interrupt expression of HNF1A (Table 2). In fact, there is no evidence of wild-type HNF1A expression in SARC1, indicating the possibility that the CMKLR1-HNF1A fusion transcript is evidence of a biallelic inactivation of HNF1A in SARC1 (Figure 7a). Biallelic inactivation of HNF1A has been previously reported to lead to aberrant activation of signalling pathways involved in tumorigenesis in human hepatocellular adenomas [26].
The RREB1-TFE3 gene fusion found in the intermediate grade myofibroblastic sarcoma SARC3 fuses the first eight exons of RREB1 to the last nine exons of TFE3, preserving the open reading frame of both RREB1 and TFE3. The fusion is predicted to interrupt expression of RREB1 (Table 2), indicating that RREB1-TFE3 is the dominantly expressed RREB1 allele. The underlying translocation leaves intact the DNA binding domain and N-terminal activation domain of TFE3 (Figure 7b). TFE3 is a known fusion partner in papillary renal cell carcinoma [30] and alveolar soft part sarcoma [31].
Finally, the SMARCB1-WASF2 gene fusion found in SARC2 is predicted to produce a transcript that preserves the reading frame of both SMARCB1 and WASF2. The predicted fusion protein would be composed of amino acids 1–209 of SMARCB1, which would preserve a DNA binding domain at 106–183 but interrupt a MYC binding domain at 186–245 [32], suggesting that the SMARCB1-WASF2 fusion protein would retain only partial SMARCB1 function. SMARCB1 has been shown to be frequently inactivated in epithelioid sarcomas [27].
We have developed a new algorithmic method called deFuse for gene fusion discovery in RNA-Seq data. We evaluated deFuse on 40 ovarian cancer patient samples, one ovarian cancer cell line and three sarcoma patient samples. Using these data, we demonstrate with RT-PCR validated fusions how deFuse exhibits substantially better accuracy than two competing methods and that deFuse is able to discover gene fusions that are not discoverable by more simplistic methods. deFuse computes a set of 11 quantitative features used to characterize its predicted fusions. In our initial analysis we used heuristic, intuitively chosen thresholds to eliminate false positives and nominated expected true positives and false positive predictions for RT-PCR validation. This yielded a set of benchmark fusion predictions: 60 true positives and 61 true negatives that we in turn leveraged to train an adaboost classifier to more robustly and objectively identify real gene fusions from the features. The classifier yielded an AUC accuracy of 0.91. Importantly, the validated fusions in ovarian cancer represent the first reported gene fusions in that tumor type.
The lack of a sufficient number of positive and negative controls for a particular type of event, such as gene fusions, represents a major challenge when evaluating novel algorithms designed for discovery of those events. This challenge is exacerbated when the prediction set contains a much larger proportion of negatives than positives. We attempted to select candidates to enrich for positive examples to provide a balanced set of ground truth events with which to train our classifier. While this has inherent biases, only one in 40 randomly chosen predictions validated indicating that a completely unbiased selection would have yielded too few positives to robustly fit a classifier. We attempted to mitigate the acknowledged biases by using other software to find additional positives and also included the very limited set of published examples from the literature.
The main limitation of deFuse is the requirement of at least five discordant read pairs to nominate a gene fusion to the adaboost classifier. This will certainly miss fusions that have very low expression and may result in insensitivity to fusions from RNA-Seq datasets with minimal sequence generation. This is suggested by the results in “Rediscovery of known gene fusions”. However, sequencing platforms are increasing throughput at exponential rates and it will soon be rare for an RNA-Seq library to under-sample a transcriptome. Another potential limitation of deFuse is its reliance on an annotated set of genes. As such, it will not be able to discover fusions that involve loci that are not annotated as genes. Finally, deFuse relies on alignment to a reference as its primary analytical step. Thus deFuse would miss gene fusions involving completely novel sequences that may exist in a transcriptome library but are not represented in the reference used by the aligner. In such situations, de novo assembly based methods such as Trans-ABySS [33] may outperform deFuse.
Full characterization of the mutational composition of cancer genomes will provide the opportunity to discover drivers of oncogenesis and will aid the development of biomarkers and drug targets for targeted therapy. As production of RNA-Seq data derived from tumor transcriptomes becomes routine, sophisticated techniques such as those used by deFuse will be required to identify the gene fusions that are part of each tumor's mutational landscape. As a first step in this process, we have identified gene fusions as a new class of features of the mutational landscape of ovarian tumor transcriptomes, in addition to discovering novel gene fusions in three sarcoma tumors.
|
10.1371/journal.pcbi.1001005 | Network Analysis of Global Influenza Spread | Although vaccines pose the best means of preventing influenza infection, strain selection and optimal implementation remain difficult due to antigenic drift and a lack of understanding global spread. Detecting viral movement by sequence analysis is complicated by skewed geographic and seasonal distributions in viral isolates. We propose a probabilistic method that accounts for sampling bias through spatiotemporal clustering and modeling regional and seasonal transmission as a binomial process. Analysis of H3N2 not only confirmed East-Southeast Asia as a source of new seasonal variants, but also increased the resolution of observed transmission to a country level. H1N1 data revealed similar viral spread from the tropics. Network analysis suggested China and Hong Kong as the origins of new seasonal H3N2 strains and the United States as a region where increased vaccination would maximally disrupt global spread of the virus. These techniques provide a promising methodology for the analysis of any seasonal virus, as well as for the continued surveillance of influenza.
| As evidenced by several historic vaccine failures, the design and implementation of the influenza vaccine remains an imperfect science. The virus's rapid rate of evolution makes the selection of representative strains for vaccine composition a difficult process. From a global health viewpoint, how to optimally implement a limited stockpile of vaccines is another fundamental question that remains unanswered. An understanding of how influenza spreads around the world would greatly aid the design and implementation process, but regional and seasonal bias in collected virus samples hampers epidemiologic analysis. Here, we show that it is possible to counter this data bias through probabilistic modeling and represent the global viral spread as a network of seeding events between different regions of the world. On a local scale, our technique can output the most likely origins of a virus circulating in a given location. On a global scale, we can pinpoint regions of the world that would maximally disrupt viral transmission with an increase in vaccine implementation. We demonstrate our method on seasonal H3N2 and H1N1 and foresee similar application to other seasonal viruses, including swine-origin H1N1, once more seasonal data is collected.
| Influenza, a negative-sense RNA orthomyxovirus, is one of the few diseases that is truly global in scale. It is responsible for approximately three to five million cases of severe acute respiratory illness and 250,000 to 500,000 deaths each year throughout the world [1]. In 2009, the swift isolation of swine-origin H1N1 strain (S-OIV) from all continents within several weeks of onset reinforced the idea that influenza is a highly infectious agent circulating worldwide [2], [3].
Although vaccination remains one of the most powerful ways of combating influenza, choosing a representative strain for vaccine composition poses a challenging problem. Due to the virus's high evolutionary rate, significant resources must be spent to update vaccines each year in order to match the dominant epitope of the season. Even with annual strain selection, major antigenic reassortment can obviate otherwise promising vaccine candidates, as occurred with the ‘Fujian/411/2002’-like H3N2 strain in 2003 [4], [5]. To prevent such vaccine failures, a solid understanding of the global spread of influenza must inform the design process. If reservoirs for new viral strains can be identified, surveillance in these areas can better optimize prediction of seasonal variants in seeded regions.
Previous papers investigating the global circulation of H3N2, the major seasonal influenza subtype prior to pandemic H1N1, focused on transmission within and between climate zones. Important motivating factors for such analysis include increased aerosol transmission in cold and dry conditions, as well as increased indoor crowding and decreased host immunity in cold and wet conditions [6], [7]. In the temperate zones, influenza exhibits distinct seasonality with flu-related cases spiking in the winter. However, several papers have confirmed the presence of viral diversity even between these epidemic peaks [8], [9], [10], suggesting two possible scenarios during the inter-epidemic period: either viral infections locally persist at a low level only to reemerge as the dominant strains of the epidemic season, or an outside source introduces new genetic diversity into temperate populations each year. Although a degree of local persistence may occur, phylogenetic analysis supports the latter scenario, with few direct links between strains of the same region but successive seasons [8], [9], [10].
For a given temperate zone, these conclusions suggest the tropics or the opposite temperate zone as plausible external seeding regions. At first blush, northern-southern temperate oscillations seem credible. Each year, northern and southern temperate climates have alternating seasonal influenza epidemics, lasting from November to April, and May to September respectively [11]. A possible mechanism of viral spread could involve transmission from the seasonal peak of one temperate zone into the season ebb of the other. On the other hand, specific epidemiological characteristics suggest a tropical origin for influenza. For example, although both climates share a similar yearly burden of mortality from influenza, the tropics do not possess the same consistent seasonal peaks during the winter months [9], [12], [13]. With a constant, low-level circulation of viruses year-round, the tropics represent an ideal epicenter for the extended transmission of new viruses to the rest of the world [14], [15], [16].
Several papers tracking H3N2 across continents have asserted that this tropical reservoir of influenza strains lies within East-Southeast Asia [12], [14], [17]. Russell, et al. analyzed H3N2 data to identify regions of the world that are antigenically and genetically leading or trailing. They found that newly emerging strains appeared in E-SE Asia roughly 6–9 months earlier than in other parts of the world, while South America experienced delayed transmission of roughly 6–9 months following other parts of the world [8].
However, such studies have been limited by several drawbacks. Most papers focus on H3N2 as a single entity, when in reality, it co-circulates with several other subtypes, the most important of which is seasonal H1N1 [11]. Although they possess different surface antigens, H3N2 and H1N1 share enough genetic similarity to display cross-immunity. As a result, seasonal H1N1 may demonstrate transmission patterns distinct from H3N2's [18], [19]. Such codependence between different subtypes is exemplified by the pandemic years of 1957 and 1968, when H2N2 replaced preexisting H1N1 and H3N2 replaced preexisting H2N2, respectively [20], [21]. Similarly, the antigenically different pandemic H1N1 strain of 2009 has largely overtaken previously circulating H1N1 and H3N2 [22]. During the years our dataset took place, evidence that H3N2 and H1N1 rarely co-dominate in a season further supports the idea of codependent dynamics [7].
A second shortcoming stems from biases in the number of sequences from different regions and different seasons [8]. Most isolates of H3N2 and H1N1 were sampled from North America, whereas Africa and South America have been largely neglected [23]. Many sequences were obtained within the last 15 years, making reliable tracking over long periods of time problematic. On the level of climate zones, the number of temperate isolates far outstrips the tropics. Although hemagglutinin (HA), the HA1 domain, and neuraminidase (NA) have the most globally representative distributions of sequences, even these remain skewed (Figure S1, Figure S2).
In this paper, we present a novel probabilistic model for tracking the spread of influenza that employs two strategies to eliminate regional and seasonal data bias. The first involves clustering isolates of high sequence similarity by region and season. Since we would expect highly similar sequences from the same time and location to be related, we considered seeding events between clusters to be of greater significance. Consideration of clusters rather than individual sequences nullifies the over-representation of a high number of isolates from a single region and season (Figure 1). As a second strategy for eliminating bias, we determined statistical significance of inter-cluster seeding events by modeling transmission as a binomial distribution with prior probabilities based on the proportion of sequences isolated before a given time point. To illustrate our methodology, Figure 2 depicts the 2003–2004 flu season, which was marked by failure to predict the dominant, tropically-derived Fujian/411/2002-like H3N2 strain. We identified a strong seeding pattern from the tropics to all three climate zones, supporting the effectiveness of our methodology.
We applied this model to the H3N2 and H1N1 coding regions of HA and NA, the most antigenic proteins of the eight viral segments. Clustering H3N2 sequences confirmed previous findings that this strain originates in the tropics, specifically E-SE Asia, and seeds South America by way of North America last. Clustering H1N1 NA also revealed a similar pattern of circulation beginning in the tropics. However, similar H1N1 analysis by continent and country was not possible due to the absence of a larger number of countries in the dataset.
Applying the same methodology to the H3N2 HA1 domain increased the geographic diversity enough to enable reconstruction of the global influenza network prior to the 2009 pandemic strain at a country level. Our results suggest a possible flu seeding hierarchy beginning in China and spreading throughout a highly interconnected E-SE Asian subnetwork. From there, viruses transmit to an Oceanic subnetwork dominated by interchange between Australia and New Zealand. Both subnetworks seed into the USA, which in turn seeds many countries, particularly in South America.
Expanding upon the sink-source hypothesis of global influenza dynamics proposed by Rambaut, et al. [15], we applied techniques of graph theory to identify important source and sink regions in the global flu network. These techniques better describe the dynamic nature of influenza movement across the globe, as well as suggest different vaccination strategies to disrupt maximally viral flow around the world.
Spatiotemporally clustering the complete H3N2 and H1N1 coding sequences for HA and NA allowed the determination of multiple statistically significant seeding seasons between 1988 and 2009. For our initial analysis, we clustered sequences into three climate zones—northern temperate, tropical, and southern temperate. To determine seasonal boundaries, we defined the northern temperate season to last from 1st July to the 30th June of the following year and the southern temperate season to last from 1st January to the 31st December of the same year [11]. Although the tropics do not have a well-defined seasonal pattern, we determined a consensus tropical flu season from 1st October to 30th September of the next year (Text S1, Table S1).
Results for H3N2 showed that the overwhelming majority of statistically significant seeding seasons came from the tropics, confirming previous findings (Figure 3A, Figure S3A). Clustering H3N2 by the six major continents rendered an even more detailed picture. For HA, Asia was the primary seeder of Asia, North America, and Oceania. Prominent transmission from North America to Europe and South America was also observed (Figure S3B). Interestingly, this hierarchical seeding structure reflects the findings of Russell, et al., which identified Asia and South America as antigenically advanced and lagging continents respectively [8]. This network of hierarchical seeding can be visualized as a directed graph plotted against the world map (Figure 4A). Analysis of NA produced similar findings with the exception of North America being its own primary seeder (Figure 3B). No complete HA and NA isolates existed in the NCBI Influenza Virus Resource database [24] for Africa.
The complete dataset of HA and NA represented only 17 and 21 countries respectively. Despite the sparse number of countries for analysis, both HA (Figure S3C) and NA (Figure 3C) consistently identified Hong Kong (considered a country by NCBI sequence annotation) as the primary external seeder of USA and New Zealand among others, and New Zealand as the primary external seeder of Australia.
Due to fewer available sequences, clustering H1N1 did not yield as many significant seeding events as H3N2; however, our tests suggest that H1N1 adopts a similar seeding pattern with the tropics as a source. Of the two segments, NA sequences display a broader geographical profile than HA. In particular, our HA dataset for H1N1 contained no sequences from Hong Kong and only 1 (0.091%) China sequence, while NA contained 9 (0.69%) Hong Kong and 3 (0.23%) China sequences. Consequently, we considered NA to be more suitable for comparison between H3N2 and H1N1 and HA to be a background signal to assess the effect of Hong Kong and China on global influenza transmission. Even so, the number of these H1N1 Hong Kong and China sequences remained vastly disproportionate to the 361 (7.42%) Hong Kong and 133 (2.73%) China sequences of H3N2.
Clustering H1N1 NA by climate zone supported the theory of global viral spread from the tropics (Figure 5B). Unlike H3N2, H1N1 analysis by continent and country was inconclusive due to low (typically fewer than 3 seeding events), homogeneous counts. Although inconclusive, the fact that a tropical signal could be detected at all from such few tropical countries, including Hong Kong and China, suggests that H1N1 adopts a similar seeding pattern out of the tropics. Due to insufficient sampling, however, a more detailed transmission pattern could not be discerned.
Although using the complete HA and NA coding genomes facilitated differentiation of isolates by Hamming distance, the absence of data from certain countries limited the information gained from clustering at this geographic detail, a problem that has plagued previous studies [8]. To increase the amount of data from different geographical regions, we clustered H3N2 sequences of the HA1 epitope, expanding the number of isolates in the dataset from 2,251 to 4,864, and the number of countries from 17 to 81. A necessary consequence of expanding geographic coverage was an increase in the number of non-unique solutions (Text S1).
Importantly, clustering HA1 by climate and continent was corroborated by findings from the complete HA and NA sequences, lending credence to the validity of the dataset. Due to the inclusion of isolates from Africa, which was hitherto not present in our datasets, H3N2 HA1 analysis also revealed Europe and North America tied for being the primary seeders of Africa.
Country clustering of the HA1 data produced a highly detailed global network of influenza variants. USA, Hong Kong, Australia, and China were identified as the four most prominent seeding countries in that order (Figure 3D, Table S2). From the data, an inferred seeding hierarchy would begin with China at the epicenter of an E-SE Asian influenza subnetwork. Our analysis supports China as the most predictive seeder of many Asian countries, including Hong Kong. Both China and Hong Kong then serve as a launching pad for the dispersal of new seasonal variants to the rest of the world [14], [17], in particular USA and an Oceanic subnetwork dominated by interchange between Australia and New Zealand. Viruses from USA, the largest seeder of the entire world, then spread to a number of South American, European, and African countries. Interestingly, Australia and Hong Kong are equally probable seeders of the USA (Figure 3D). Detailed transmission events are enumerated in Table S2. An inset of the Asian subnetwork is depicted in Figure 4C, a demonstration of this study's high geographic resolution.
As can be seen with the world map plots (Figure 4A,B), a natural representation of the global influenza network is a directed graph with each node representing a clustered region (climate, continent, and country) and each edge representing a seeding event with a weight equal to the number of significant seeding seasons. To quantify observed patterns, we employed principles of graph theory to measure the importance of nodes using four different metrics.
By counting the number of indegrees and outdegrees of each node for H3N2, we identified that the tropics and the northern temperate zone (Figure S4A), specifically Asia and North America (Figure Figure S5A), transmit and receive the most seeding events to and from the rest of the world, respectively. In a similar manner, we identified USA, Hong Kong, Australia, and China as the greatest seeders, and USA, Japan, Australia, and Hong Kong as the most seeded (Figure 6A).
In this analysis, we differentiated between internal (self-seeding) and external (seeding between nodes) transmission events. Importantly, we can accurately detect internal events in temperate countries since their flu seasons are discrete. On the other hand, the specificity for internal events in the tropics is much lower due to unpronounced seasonal peaks. To minimize the number of local false positives, we demarcated seasons within the tropics on a per country basis. We found that for all climate zones except the tropics (Figure S4A) and all continents except Asia (Figure S5A), the number of internal seeding events paled in comparison to the proportion of external seeding events,. The more numerous internal events in the tropics and Asia indicate a high level of circulation between tropical countries and between Asian countries. This pattern is supported by the highly interconnected E-SE Asian subnetwork depicted in Figure 4C. The small proportion of internal events for countries supports the notion that local persistence often plays only a minor role in influenza transmission [8], [9], [10] (Figure 6A).
Beyond the absolute number of seeding events, a region's influence on global viral spread is also dependent on the topological structure of the graph itself. As an analogy, consider the influenza network as a system of connected train stations each representing a single region seeding influenza. In such systems, trains begin and end their routes at terminal stations. Similarly, influenza commuters begin their journeys at terminal sources and end at terminal sinks in each season. These start and end terminals can represent regions where new influenza variants respectively originate and ultimately spread to. To quantify the terminal characteristic, we calculated the outdegree minus the indegree of each node, which we term “degree flow.” Positive degree flow indicates terminal sources, while negative indicates terminal sinks. Countries were also ranked by calculating the proportion of nodes in a 1,000 randomized networks with a greater, or lesser, degree flow (Text S1).
For analysis by climate zone, the tropics was identified as the only terminal source, suggesting that flu spreads from the tropical belt outward to both temperate zones (Figure S4B). As for continental clustering, Asia was the only terminal source, indicating that global circulation begins in Asia and ends in terminal sink continents, of which North America was the most prominent (Figure S5B). On a country level, Hong Kong and China were the greatest terminal sources, corroborating our observations (Figure 3D). Australia was also a conspicuous terminal source, especially within the Oceanic subnetwork where it seeded the greatest terminal sink, New Zealand. Several South American countries, including Chile and Argentina, figure as terminal sinks too, correlating with such countries as antigenically delayed [8] (Figure 6B).
Trains also stop at waypoint stations, which can be the junction of a large number of routes. Correspondingly, certain regions act as waypoint sources: important intermediate launch pads to other destinations. Others act as waypoint sinks: important points of convergence for multiple routes. Eigenvector centrality can gauge this property on the principle that connections to high-scoring nodes contribute more to the score of the node in question than equivalent connections to low-scoring nodes. We used a method akin to PageRank, Google's method of assigning importance to web pages [25].
Using this method, the northern temperate zone was the most important waypoint source and sink (Figure S4C). Similarly, the predominantly northern temperate continents of North America and Europe were identified as prominent waypoint sources and sinks. Asia, however, was the greatest waypoint source but a poor waypoint sink, correlating with its role as a greater terminal source than North America or Europe (Figure S5C). Interestingly, USA was both the greatest waypoint source and sink (Figure 6C).
H1N1 NA clustering by climate zone produced results similar to that of H3N2 NA. The tropics consistently scored highest by seeding outdegree, positive degree flow, and PageRank source. In addition, the tropics possessed a large amount of internal seeding events. These results emphasize that similar to H3N2, H1N1 circulates within the tropics across seasons only to spread eventually to the temperate zones.
Betweenness measures the number of shortest paths between any two vertices in a network that lie on a given node. In the context of influenza, increasing vaccinations in regions of high betweenness would hypothetically have the greatest effect on diminishing the spread of infection worldwide. This novel strategy contrasts with previous studies simulating containment only at the source of influenza [26], [27]. For H3N2, this criteria highlighted Europe and North America as promising candidates for vaccination programs (Figure S5D). Clustering by country revealed USA, Japan, and Australia as sites in the influenza network vulnerable to disruption (Figure 6D).
Using statistical and network theory analysis, we analyzed H3N2 and H1N1 sequence data to determine the global spread of influenza. Our novel method employs two main strategies to eliminate geographic and seasonal bias: 1) Spatiotemporal clustering of sequence data to count seeding events between clusters and 2) Use of binomial prior probabilities based on the regional proportion of viral isolates to screen for significant seeding events.
Applying these techniques to coding HA and NA segments of H3N2 by climate zone and continent revealed a seeding pattern stemming from the tropics, particularly Asia. HA1 analysis produced a more detailed picture: each year, a wave of seasonal flu originates in China to feed an E-SE Asian subnetwork. From there, China and Hong Kong seed two major subnetworks, each dominated by Australia and USA.
Similar clustering of H1N1 NA sequences by climate zone reproduced tropical transmission to the rest of the world. However, due to inadequate geographic coverage, clustering H1N1 by continent and country proved inconclusive with few significant seeding events detected. One explanation for these results is that important seeding countries, such as China and Hong Kong, were too underrepresented in the dataset. Alternatively, global patterns may be weaker for H1N1 due to cross-reactivity between the two strains [18], [19], a conclusion reflected by the smaller number of seeding events for the strain.
In our analysis, the total number of seeding seasons for each region did not necessarily correspond to the total number of isolates from each region, indicating that our methodology counters data bias. However, certain confounders may affect results. First, selection bias in sampling remarkable variants, such as patients suffering severe rather than mild or non-symptomatic influenza, would poorly represent flu in the general population. Moreover, many sequences had to be excluded from our dataset due to poor annotation and lack of date information. Finally, although our probabilistic methodology accepts regional and temporal variability, it has low sensitivity for detecting anything but particularly significant seeding events for regions with very few sequences. This issue becomes important in analyses with regions that have no sequences whatsoever, as with near-absent sequences from Hong Kong and China for H1N1 HA. The persistence of such bias highlights the continuing need to sequence viruses in underrepresented areas, especially the tropics.
Each year, the current influenza vaccine is formulated separately for the Northern and Southern Hemisphere; one can surmise that two viral strains may not be enough to represent the entire pool of influenza strains around the world. Although there are many other economic and political concerns to consider, our methodology suggests several ways of guiding vaccine strain selection based on biological and epidemiological principles. Graph theory metrics—terminal and waypoint sinks and sources, as well as degree and betweenness centralities—pinpoint potential regions in which increased vaccinations could stem the transmission of influenza globally as well as locally. Increased analytical resolution could optimize vaccine design by choosing the dominant antigenic strain of a country's most predictive seeder. Vaccines could be catered to each country, rather than each hemisphere. At the very least, our analysis advises strain selection from the tropics, from which seasonal strains are dispersed each year. On the other hand, local strain selection within a country should prove comparatively ineffective, as few viruses persist in the inter-epidemic period to seed the following flu season.
Our analysis of terminal sources resonates with an old hypothesis that in southern China, zoonotic infection from live-animals markets [28] selling in particular duck—a natural host of influenza [29]—combined with a dense population for sustained viral circulation, could be the main ingredients for the creation of new seasonal influenza variants. In support, two major acute respiratory infections—SARS [30] and H5N1/97 [31], [32]—have been definitively traced back to southern China, with Hong Kong serving as an important sentinel post for the rest of the world. Other influenza pandemics, 1968 H3N2 (Hong Kong) [28] and even as early as 1889 pandemic influenza [33], have suspected origins in southern China.
It would be interesting to dissect the factors that govern waypoint sources and sinks. For example, air travel and other transportation may play a major role in the dispersal of virus worldwide [8], [19], [34], [35]. Many important hubs of the global flu network, including USA, Australia, Hong Kong, and China, have several of the world's busiest airports [36]. Understanding the reasons for these seeding patterns may offer other strategies for arresting the movement of flu.
The advent of 2009 pandemic S-OIV has largely depleted the number of seasonal H3N2 and H1N1 infections, most likely via cross-reactivity between novel and seasonal strains [22]. Consequently, the conclusions of this paper may not necessarily apply to current dynamics of seasonal H3N2 and H1N1. However, the fact that H1N1 shares a tropic-centric movement pattern with H3N2 despite cross-reactivity suggests that these patterns may still persist even in the presence of the cross-reactive S-OIV. Moreover, this paper demonstrates that when more sequence data is deposited in NCBI, a similar methodology can be applied to predict global circulation of S-OIV as well.
All sequence data used in this study was publicly available from the National Center for Biotechnology Information database (NCBI) [37]. For each segment, only protein coding regions were considered. Furthermore, we only used sequences with full date (year, month and day) and location information to build hierarchies. Geographical coordinates of each isolate were obtained using geolocation information from Google Maps. Sequences were then aligned using the ClustalW v. 1.83 multiple sequence alignment package using default parameters for H3N2 and H1N1, respectively. For each segment, sequences were aligned and those that were poorly aligned compared to the rest of the dataset were removed until all sequences aligned with a Hamming distance no greater than 0.15. Given estimated mutation rates of 6.7×10−3 nucleotide substitutions per site per year [12], [19], Hamming distances over the 20-year span of our dataset are expected to be no more than 0.15 of the sequence length. Outlying sequences were most likely incorrectly sequenced and were discarded from analysis.
Our methodology aimed to minimize data bias from geospatial and temporal variability in sequences from NCBI. First, we determined the most parsimonious evolutionary paths traversed by the flu virus. To this end, we sorted sequences from earliest to most recent viral isolates. Working backwards from newest to oldest, we calculated the sequence similarity of each virus to all earlier isolates regardless of geography. We defined a virus's most likely ancestor to be the sequence with minimum Hamming distance. From this data we built evolutionary paths for each virus. Related sequences were clustered (grouped) together by common geography and season to simplify the paths. For example, a chain of related viruses in the same region and season would be collapsed into a single umbrella node representing all of them. Our analysis was then based on looking at the transitions between clusters rather than individual viruses. We counted these “seeding events,” where the closest ancestor of a given cluster of sequences is from a different region or season [8] (Figure 1). When tallying seeding events, non-unique solutions were not considered where a given viral isolate possessed multiple closest ancestors from different geographical zones or seasons (Text S1, Figure S6).
The observed frequencies of seeding events between clusters were compared to expected frequencies based on the prior probability of randomly choosing a sequence from a given geographical zone in the past. Using the binomial distribution with the proportion of prior NCBI sequences as a binomial probability, a p-value was calculated for observing more seeding events than expected. The best predictor of a seeding region for each season had the greatest ratio of observed to expected seeding events with a p-value smaller than 0.05 (Figure 2).
|
10.1371/journal.pcbi.1000018 | Evolutionary Patterns in the Sequence and Structure of Transfer RNA: Early Origins of Archaea and Viruses | Transfer RNAs (tRNAs) are ancient molecules that are central to translation. Since they probably carry evolutionary signatures that were left behind when the living world diversified, we reconstructed phylogenies directly from the sequence and structure of tRNA using well-established phylogenetic methods. The trees placed tRNAs with long variable arms charging Sec, Tyr, Ser, and Leu consistently at the base of the rooted phylogenies, but failed to reveal groupings that would indicate clear evolutionary links to organismal origin or molecular functions. In order to uncover evolutionary patterns in the trees, we forced tRNAs into monophyletic groups using constraint analyses to generate timelines of organismal diversification and test competing evolutionary hypotheses. Remarkably, organismal timelines showed Archaea was the most ancestral superkingdom, followed by viruses, then superkingdoms Eukarya and Bacteria, in that order, supporting conclusions from recent phylogenomic studies of protein architecture. Strikingly, constraint analyses showed that the origin of viruses was not only ancient, but was linked to Archaea. Our findings have important implications. They support the notion that the archaeal lineage was very ancient, resulted in the first organismal divide, and predated diversification of tRNA function and specificity. Results are also consistent with the concept that viruses contributed to the development of the DNA replication machinery during the early diversification of the living world.
| The origins of the three major cellular lineages of life—Archaea, Bacteria, and Eukarya—and of viruses have been shrouded in mystery. In this study, we focus on transfer RNA, an ancient nucleic acid molecule that takes center stage in the process of protein biosynthesis and can be found everywhere in life. In a process that reconstructs history from molecular sequence and structure and at the same time forces molecules belonging to lineages into groups, we tested alternative hypotheses of origin and established when major organismal lineages appeared in evolution. Remarkably, timelines showed that Archaea was the most ancient lineage on earth and that viruses originated early in the archaeal lineage. Our findings unroot the universal tree of life, and, for the first time, provide evidence for an evolutionary origin of viruses.
| Transfer RNA (tRNA) molecules are central to the entire translation process. They interact with the ribosomal RNA (rRNA) subunits as they are being ratcheted through the center of the ribosome [1],[2]. Their acceptor arms charge specific amino acids through the activity of cognate aminoacyl-tRNA synthetases, while triplets of bases on their ‘anticodon’ arms recognize complementary ‘codon’ sequences in messenger RNA. These and many other molecular interactions define the identities and functions of these tRNA adaptors and establish a genetic code that translates nucleic acid into protein information in the cell. The structural make-up of tRNA is therefore fundamental to our understanding of how the biosynthetic machinery was set up into place in an emerging protein and organismal world. tRNAs are clearly ancient molecules [3] and they have been used profusely to study the evolution of ancient life [4]–[8]. The identity and function of tRNAs are fundamentally delimited by the structure of the molecules, and structure is more conserved than sequence. In fact, we recently showed that tRNA structure carries deep phylogenetic signal and can be used directly to reconstruct evolutionary history [9]. However, understanding phylogenetic trees is challenging because tRNA evolution embeds a history of recruitment in which structures gain or co-opt new identities and functions or takeover established ones.
The hierarchical branching patterns of the universal tree of life portray the natural history of the living world. The current accepted universal tree proposes a tripartite world ruled by three superkingdoms, Archaea, Bacteria, and Eukarya [10]. This view stems fundamentally from the study of the small subunit of rRNA, a molecule that is also ancient and central to translation. The rise of evolutionary genomics with an analysis of entire repertoires of nucleic acid and protein molecules supports this tripartite scheme [11],[12]. However, the root of the universal tree remains controversial and so is the nature of the universal ancestor of all life that this root defines [13],[14]. We recently embarked on a systematic and global study of evolution of domain structure and organization in proteins [15],[16] (Wang and Caetano-Anollés, submitted). Structures were assigned to protein sequences in hundreds of completely sequenced genomes and a structural census of protein domains used to generate phylogenomic trees of protein architectures. The evolutionary genomic analysis defined a universal ancestor that was eukaryotic-like and had a relatively complex proteome [16]. It also established that the archaeal lineage was the most ancient and originated from reductive evolutionary tendencies in the use of protein architectures.
In order to explore if similar phylogenetic signatures were present in tRNA, we apply a well-established cladistic method [17],[18] that embeds structure directly into phylogenetic analysis [19]. The method involves identifying features characteristic of the secondary structure of RNA molecules, coding these features as linearly ordered multi-state characters, and using them to build phylogenetic trees with optimal tree search methods. The strategy has been used to reconstruct a tripartite tree of life from rRNA structure [17], trace evolution of rRNA in ribosomes [18], study the evolution of closely related phytopathogenic fungi [17] or distantly related members of the grass family [20], and explore the structural origin and evolution of retrotransposons in eukaryotes [21]. We also used the approach to study the evolution of the major structural and functional components of tRNA, establishing that tRNA molecules originated in the acceptor arm and providing further support to the ‘genomic tag’ hypothesis [9]. Here we reconstruct global phylogenetic trees using information embedded in both the sequence and structure of tRNA molecules. As we have shown previously (Sun and Caetano-Anollés, submitted), the intrinsically rooted trees revealed that tRNA with long variable arms (known as class II or type II tRNA) coding for amino acids Sec, Ser, Tyr, and Leu were ancient. However, trees failed to show clear patterns related to tRNA function, an observation that underscores the importance of recruitment and phylogenetic constraint (factors that restrict the acquisition of phenotypic traits or functions in lineages) in tRNA evolution. In order to sort out these confounding processes we built trees while forcing monophyletic groupings of taxa (sets that share a common ancestor) to test alterative hypotheses or establish evolutionary timelines of structural, functional, or organismal diversification. This strategy (known as constraint analysis in phylogenetics) provided an unanticipated window into early evolution of life.
Phylogenetic analyses of the combined dataset of sequence and structure of 571 tRNAs produced most parsimonious trees that were 10,083 steps in length and were intrinsically rooted (Figure 1). The tRNA set was obtained from Part 2 of the Bayreuth tRNA Database and represented organisms in the three superkingdoms of life and viruses and covered all isoacceptor variants and amino acid specificities (Table S1). This molecular set is unique since it contains information of modified bases and structures derived by comparative analysis (see Materials and Methods). Bootstrap support (BS) values were generally low for most clades (<50%), but this was generally expected given the large number of taxa (molecules) analyzed. Class II tRNA molecules with long variable arms, including tRNASec and most tRNASer, tRNATyr, and tRNALeu isoacceptors, appeared at the base of the rooted trees (Figure 1). Besides this pattern, trees failed to reveal groupings that would indicate clear evolutionary links to organismal origin or molecular functions. The monophyly of tRNA belonging to each superkingdom (or viruses) or expressing different amino acid specificities was not revealed. Similarly, tRNAs with specificities for amino acids defined previously as being ancestral [22]–[27] did not form monophyletic groups. tRNA molecules sharing the first, second, or first and second bases in codons were not monophyletic either. These patterns were also observed in trees derived from partitioned matrices of superkingdoms or viruses (data not shown).
In order to uncover deep phylogenetic signals and test alternative evolutionary hypotheses we forced groups of tRNAs that shared a same organismal origin (molecules from each superkingdom of life or viruses) into monophyly using constraint analyses. We then recorded the length of the most parsimonious trees that were obtained and the number of additional steps (S) that were needed to force the constraint. This exercise was generally done with or without forcing class I and II tRNA molecules into separate groups, but overall results were congruent.
Constraints related to the diversification of the organismal world (Table 1) consistently showed Archaea as the ancestral group (i.e., forcing archaeal tRNAs into monophyly was always associated with low S), followed by viruses, Eukarya, and Bacteria (with S increasing in that order) (Figure 2). Hypotheses of relationship among superkingdoms clarified further the possible rooting of the universal tree. Constraining molecules from Eukarya and Bacteria into a monophyletic group [constraint (EB)] was the most parsimonious solution and was consistent with an early split of two ancient lineages, one of archaeal origin and the other of eukaryal-bacterial origin. When forcing molecules from two of the three superkingdoms individually and as a group into monophyly, the Eukarya and Archaea dichotomy [constraint ((E)(A))] was most parsimonious. This suggests the earliest two superkingdoms to diversify were Eukarya and Archaea. The S values for these constraints indicated that their diversification always preceded the onset of Bacteria. Finally, constraining molecules from the three superkingdoms into three separate groups in all possible 3-taxon statements showed that a polytomous arrangement was the most parsimonious. S values exceeded those indicating the onset of Bacteria as a group. These patterns maintained when tRNA structural categories were constrained and all phylogenetic statements were congruent (Figure 2).
We also explored the origins of viruses by constraining molecules from each individual superkingdom or viruses into monophyletic groups, together [e.g., (AV)] or separately [e.g., ((A)(V))] (Table 2). The most parsimonious scenario always linked the origins of viruses to the archaeal lineage, with S values matching those defining the organismal timeline (Figure 2). Dividing the viral sequences into two groups (i.e., viruses infecting Eukarya or Bacteria) maintained the linkage between the separated groups of viruses and Archaea for various competing hypotheses (Table 2).
Finally, we constrained trees according to isoacceptor group and then according to organismal group, or vice versa, with or without constraining tRNA categories (Table 3). A scenario in which organismal (superkingdom) diversification predated tRNA functional divergence was always more parsimonious (S = 2,338–2,481) than one where functional divergence predated organismal diversification (S = 2,415–2,534).
Since constraint analyses could be biased by unequal rates of evolution, we calculated average number of character change per branch in consensus trees generated from partitioned data matrices (Table 4). An analysis of variance (ANOVA) showed values were not significantly different in the three superkindoms of life and viruses (p>0.05). Similarly, we did not find differences when random trees were compared (not shown).
In order to uncover evolutionary patterns related to organismal diversification, we first generated rooted phylogenetic trees using information embedded in the structure and sequence of tRNA (Figure 1). As expected, class II tRNA molecules with long variable arms coding for Sec, Ser, Tyr, and Leu appeared at the base of the rooted trees and were ancient. We also observed a rather tight paraphyletic clustering of tRNAs of archaeal origin. However, we were unable to reveal any other pattern of significance in the trees; no monophyletic groupings could be established when tracing tRNA function, codon identity, or organismal origin (data not shown). In order to untangle the intricate history of tRNA, we forced trees to acquire pre-defined tree topologies representing competing (alternative) or non-competing phylogenetic hypotheses, constrained the exploration of tree space during phylogenetic searches, and produced sub-optimal tree reconstructions. Competing hypotheses were contrasted and those that imposed a minimum number of additional steps (S) on the optimal tree (i.e., more parsimonious) were not rejected. Using this approach, we tested for example competing chronologies or sister taxa relationships related to organismal diversification. In turn, non-competing hypotheses were ranked by the values of S according to some external evolutionary model. In this study, they were used to define timelines of first appearance of superkingdoms and viruses in evolution. Hypotheses of origin that were satisfied with fewer steps were considered less affected by the confounding effects of recruitment in lineages and more ancient than those that required more steps. In other words, it was easy to merge lineages in backwards time (a process known as coalescence) to fit the constraint. Plots mapping the correlation between S and number of nodes from a hypothetical tRNA ancestor in the trees confirmed the validity of this assumption of ‘polarization’ (data not shown). This type of analysis is not new. In cybernetics it is known as ‘constraint analysis’ and represents a formal method of decomposing a reconstructable system into its components by imposing natural or man-made limitations [28]. The method is widely used in cladistic and phylogenetic analyses to test for example hypotheses of monophyly [29], but to our knowledge, has never been used to dissect systematically patterns in a phylogenetic tree.
Two fundamental assumptions support the analysis. First, we assume tRNA structures acquired new identities and functions as the genetic code expanded, and that different structures were co-opted for the task in different lineages and different functional contexts. This assumption seems reasonable. Recruitment processes are common in evolution of macromolecules. In cellular metabolism, for example, enzymes are often recruited into different pathways to perform new enzymatic functions [16],[30],[31]. Moreover, structural diversification of tRNA appeared to have predated organismal diversification [32] (Sun and Caetano-Anollés, submitted) and the functions and identities attached to present-day tRNA structures probably developed in lineages and were shuffled by horizontal gene transfer. Second, we assume old tRNA structures developed or recruited new functions (co-options) more often than new tRNA structures acquired old functions (takeovers). This assumption is also reasonable and appears to be supported by our studies of enzyme recruitment in metabolism (Kim et al., ms. in preparation). Our trees show several instances of takeovers, in which modern class I structures lacking the long variable arms took over ancient amino acid charging functions associated with class II structures (Figure 1; Sun and Caetano-Anollés, submitted). However, old structures have more chances to succeed in a diversifying world, as they spread through lineages. Younger structures in turn are restricted to the lineage in which they originated, and can only spread further through horizontal transfer events. One implication of this assumption is that older functions will be less prone to co-options than younger functions, and that tRNA belonging to older lineages will be less affected by co-options than those in younger lineages. Consequently, ancient molecules sharing functions or belonging to selected lineages will be more easily constrained than younger variants in phylogenetic reconstruction.
We also assume phylogenies are free from systematic errors and the confounding effects of mutational saturation, long branch attraction artifacts, and unequal rates of evolution along branches of the trees [11]. However, most branching events in these phylogenies happened a relatively long time ago and phylogenetic analyses of ancient molecules carry all the problems of deep reconstruction [33]. While the impact of some of these effects diminishes when using multi-state characters in tRNA structure [34],[35], different rates of change could alter the coalescense of lineages and our results. For example, increased rates of change known to occur in rapidly evolving viral molecules could increase expected S values, making the viral lineage artificially younger. Nevertheless, an analysis of rates of change in consensus and random trees derived from partitioned data matrices showed that evolutionary rates of tRNAs in the three superkingdoms of life or viruses were not significantly different in our analysis (Table 4). The fact that evolutionary rates in the four lineages were similar decreases the impact of unequal rates of evolution and underscores the conserved nature of tRNA structure when compared to sequence. Similarly, problems of statistical consistency related to long branch attraction could bias the reconstruction of the tRNA tree. These artifacts, which are rather common in sequence analysis, result from unequal rates of variation in branches and the interplay of short and long branches in a tree [36]. They are however not so much related to branch length (which in our analyses do not vary considerably; Table 4) but to changes of a same character state occurring preferentially in long branches, forcing the tree-building method to join them artificially. However, the probability of these covarying homoplasies is known to decrease with increases in character states, as with the multi-state structural characters of this study [34], and when branches are separated by increased taxon sampling [37]–[39]. Consequently, large trees as the tRNA trees we have reconstructed from sequence and structure in this study should be considerably less prone to consistency problems [38],[39] than the four-taxon statements related to sequences originally used to define them [36], especially if they involve multiple character states depicting structure.
We constrained tRNA groups according to organismal origin using different schemes and tested possible competing and non-competing hypotheses describing timelines of organismal diversification and possible topologies of the universal tree of life (Figure 2). Constraining tRNAs belonging to individual superkingdoms or viruses showed Archaea as the most ancestral group, followed by viruses, Eukarya, and Bacteria, in that order. This timeline already suggests a very early split of the archaeal lineage in evolution. An analysis of the three possible two-superkingdoms single-group constraints showed that forcing molecules from Eukarya and Bacteria into a single monophyletic group [constraint (EB)] was most parsimonious and confirmed the early split of lineages and separation of Archaea. It also suggested an important lineage relationship between Eukarya and Bacteria and a relative time frame for their coalescence as a group. Interestingly, S values for the eukaryal-bacterial lineage always coincided with those for the viral group, suggesting viruses appeared at a time when this early lineage was coalescing (see below). Forcing molecules belonging to two superkingdoms into separate monophyletic groups once again confirmed the early split of Archaea and the late onset of Eukarya; the most parsimonious solution [constraint ((A)(E))] showed that the separate coalescence of the archaeal and eukaryal lineages followed the appearance of Eukarya as an organismal group [constraint (E)] and always preceded the appearance of Bacteria [constraint (B)]. Finally, constraining the three superkingdoms into separate monophyletic groups resulted as expected in higher S values, reflecting the coalescence of all lineages of a fully diversified organismal world. Out of all possible competing hypotheses of relationship several alternatives were most parsimonious, including an unresolved 3-taxon statement [constraint ((A)(B)(E))]. The confounding effects of recruitment were probably severe and were incapable of revealing the root of the universal tree at these high S values and late evolutionary stages.
The timeline of organismal diversification provides evidence that the lineage of Archaea segregated from an ancient community of ancestral organisms and established the first organismal divide. The scenario of organismal diversification described above is congruent with our recent phylogenomic analyses of protein structure [16] and domain organization (Wang and Caetano-Anollés, submitted) in hundreds of completely sequenced genomes. The result is also congruent with recent studies that have used tRNA paralog (alloacceptor) clustering as a measure of ancestry of tRNA genotypes [40] and multiple lines of evidence [41],[42] to suggest a Methanopyrus-proximal root of life. Although it is popularly accepted that the universal tree of life based on molecular phylogenies is rooted in the prokaryotes and that Archaea and Eukarya are sister groups, these recent results together with those presented in this paper offer compelling arguments in favor of an early appearance of the Archaea.
Our evolutionary timeline is also remarkable in that it identifies three epochs in the evolution of the organismal world that were analogous to those proposed earlier [16]: (1) an architectural diversification epoch in which tRNA molecules diversified their structural repertoires (light green areas in Figure 2), (2) a superkingdom specification epoch in which tRNA molecules sorted in emerging lineages that specified superkingdoms Archaea, Bacteria, and Eukarya (salmon areas), and (3) an organismal diversification epoch that started when all tRNA coalesced in each superkingdom (light yellow areas).
The evolutionary patterns observed in timelines appeared consistently in the absence or presence of class I or class II tRNA structural constraints (Figure 2). This suggests structural diversification predated organismal diversification during evolution of tRNA. Similarly, a scenario in which organismal diversification predated amino acid charging diversification was more parsimonious (Table 3), suggesting the discovery of both amino acid charging and associated codon function occurred in expanding lineages. These conclusions are supported by a recent study that compared sequence matches between tRNA halves and suggested the modern tRNA cloverleaf arose prior to the divergence of modern tRNA specificities and the three superkingdoms of life [32].
The organismal timeline inferred from tRNA sequence and structure showed Archaea was the most ancient superkingdom but established that viruses were also ancient. Viruses are relatively simple living entities and in many cases maintain a regular structure. They have long been considered fragments of cellular genomes and not living organisms and were generally excluded from consideration in evolutionary scenarios of the tripartite world, despite being important components of the biosphere. The importance of viruses and their potential roles in early cellular evolution were recently reevaluated [43]. A comparative analysis of structure and function, including virion assembly principles, suggested both RNA and DNA viruses may have been more ancient than previously thought, possibly even more ancient than the common ancestor of life [43]. However, they probably had a polyphyletic origin because structurally and functionally related viruses infect hosts in different lineages and even in different superkingdoms of the universal tree [44],[45]. It is therefore possible that viruses form lineages and share a common ancestor, and that these lineages extend from the root to all branches in the tree of life. For example, the overall similarity of viral structures, such as coat protein folds enclosing nucleoprotein filaments, suggests a common mechanism for their appearance. The construction of phylogenies addressing the questions of origin and evolution of viruses in the context of the three superkingdoms are now possible with the increasing number of sequenced genomes of viral origin. In fact, comparative genomic analyses suggested viruses could be the source of new proteins for cells [46]. Many DNA informational proteins encoded today in cellular genomes probably originated in the viral world and were later transferred into the three cellular superkingdoms. Forterre recently proposed that DNA itself appeared in ancestral viral lineages [47],[48]. He later on extended this proposal by suggesting that the DNA replication machineries of each superkingdom originated from three different ancestral viral lineages [49]. In his latest proposal, each cellular superkingdom originated independently from the fusion of an RNA-based cell and a large DNA virus [50].
In order to establish if the origin of the viruses was linked to one or more of the three superkingdoms of life we constrained viral and individual superkingdom tRNAs into competing monophyletic relationships (Table 3). Remarkably, most parsimonious constraints indicated viruses that associate with Eukarya and Bacteria had an origin in the archaeal lineage (Figure 2). The origin of viruses in Archaea is remarkable, especially if one considers the exceptional diversity and morphotype complexity of archaeal viruses [51]. Such an origin is compatible with the proposal by Forterre and colleagues that the transition from RNA to DNA genomes occurred in the viral world, and that cellular DNA and its replication machineries originated via transfers from DNA viruses to RNA cells. In fact, our phylogenomic analysis of structure [16] suggests a substantial portion of the replication machinery was developed during the architectural diversification phase immediately after reductive tendencies were already set in the archaeal lineage. This coincides with the relative time of emergence of viruses in the ancient world that was derived in this study. Since the appearance of a molecularly complex universal ancestor preceded the appearance of viruses, our results remain compatible with the accepted view that viruses originated from fragments of genetic material that escaped from the control of the cell and became parasitic (the escape theory) [52]–[55].
The origin of viruses is generally complex and may involve more than one mechanism [56]. Although several major classes of viruses are monophyletic, a common viral ancestry has not been evident [57]. Sequence analysis of viral genomes with various lengths (ranging from a few to hundreds of kilobases and containing several to hundreds of genes) and types (ranging from double-stranded DNA to single-stranded RNA) failed to reveal a common origin, suggesting instead polyphyletic (multiple) origins. However, a focus on sequence alone could be misleading. The viruses as a group contain more structural genomic diversity than cellular organisms such as plants, animals, or bacteria put together, and their sequences are fast evolving [58]. This could erase deep evolutionary history and confound analysis. Moreover, viruses also share many common features (e.g., genes coding for key proteins involved in viral replication and morphogenesis, parasitic nature of the replication mechanisms) not shared by any kind of cellular organisms [57], and these could be used to claim monophyly. This is especially true if the proposed ancient viral world existed [57]. This world harbored viral genes that retained their identity throughout the entire history of life. By this definition, the primordial pool of primitive genetic elements would be the ancestors of modern cellular and viral genes. This means that most, if not all, modern viruses were derived from elements that belonged to the primordial genetic pool, perhaps representing primitive form of self replicating DNA and precursor of life [59].
We end by noting that due to the small number of viral sequences sampled in our study, the conclusions drawn here should be taken with caution. However, a separate undergoing study analyzing a comprehensive dataset of tRNA sequences and structures but lacking information on base modifications support the evolutionary patterns presented in this study (Ospina, Sun, and Caetano-Anollés, unpublished).
Part 2 (compilation of tRNA sequences) of the Bayreuth tRNA Database (http://www.staff.uni-bayreuth.de/btc914/search/index.html; September 2004 edition; Table S1) contains a total of 571 tRNA sequences at RNA level with cloverleaf secondary structures. The structures were derived by comparative analysis using an alignment that is most compatible with tRNA phylogenies and known 3-dimensional models of structure [60],[61]. The composition of part 2 was not pruned in our analyses and represents the most complete tRNA dataset currently available that contains information about base modifications. A total of 42 structural characters describing geometrical features of tRNA molecules (Table S2) were scored, establishing character homology by the relative position of substructures in the cloverleaf [9] (Sun and Caetano-Anollés, submitted). The length (the total number of bases or base pairs) and number of the substructures were coded as character states and were defined in alphanumerical format with numbers from 0 to 9 and letters from A to F. The minimum state (0) was given to missing substructures. We followed the Bayreuth database to treat the modified bases as deviations from the cloverleaf model. They were not allowed to establish canonical Watson-Crick pairs. Each helical stem region was scored as two complementary sequences (5′ and 3′ sides). The dataset was then partitioned into four subsets categorized by molecules belonging to each of the three superkingdoms or viruses/bacteriophages. In this study, a “total evidence” approach [62],[63] (also called “simultaneous analysis” [64]) was invoked in phylogenetic analysis to combine both sequence and structure data of the complete (571 tRNAs) and partitioned matrices. The goal of this analysis was to provide stronger support for the phylogenetic groupings recovered from analyses of structural data.
We treated structural features in molecules as phylogenetic multi-state characters with character states transforming according to linearly ordered and reversible pathways. Character state transformations were polarized by assuming an evolutionary tendency towards molecular order. Characters were analyzed using maximum parsimony (MP), a popular phylogenetic optimization method that searches for solutions that require the least amount of change. It is appropriate to treat geometrical features as linearly ordered characters because RNA structures change in discrete manner by addition or removal of nucleotide units. This causes gradual extension or contraction of geometrical features. Although insertion and deletion are also possible, they are more costly. The validity of character argumentation has been discussed in detail elsewhere [9],[17],[18],[20]. A considerable body of evidence supports our polarization hypothesis depicting generalized trends applied to the structure of molecules: (i) the study of extant and randomized sequences shows that evolution enhances conformational order and diminishes conflicting molecular interactions over those intrinsically acquired by self-organization [20], [65]–[70], (ii) a molecular tendency towards order and stability has been experimentally verified using thermodynamic principles generalized to account for non-equilibrium conditions [71]; (iii) a large body of theoretical evidence supports the structural repertoire of evolving sequences from energetic and kinetic perspectives [72]–[74], with some important predictions confirmed experimentally [75], (iv) phylogenies generated using geometrical and statistical structural characters are congruent [9],[20],[21], and (v) the reconstructions of rooted trees generated from sequence, structure, and genomic rearrangements at different taxonomical levels are congruent [17], [18], [20], [21], [76]–[78]. Phylogenetic trees were polarized by distinguishing ancestral states as those thermodynamically more stable. This results in reversible character transformation sequences that are directional and show asymmetry between gains and losses. Maximum and minimum character states were defined as the ancestral states for structures that stabilize (stems, modified bases, and G:U base pairs) and destabilize tRNAs (bulges, hairpin loops, and other unpaired regions), respectively.
All data matrices were analyzed using equally weighted MP as the optimality criterion in PAUP* v. 4.0 [79]. Because MP may outperform maximum likelihood (ML) approaches [34],[35], the use of MP is particularly appropriate for our analysis. ML is precisely MP when character changes occur with equal probability but rates vary freely between characters in each branch and when using large multi-step character state spaces (decreasing the likelihood of revisiting a same character state on the underlying tree). This makes MP statistically consistent. Reconstructions of MP trees were sought using heuristic search strategies; 1,000 heuristic searches were initiated using random addition starting taxa, with tree bisection reconnection (TBR) branch swapping and the MulTrees option selected. One shortest tree was saved from each search. Hypothetical ancestors were included in the searches for the most parsimonious trees using the Ancstates command. BS values [80] were calculated from 105 replicate analyses using “fast” stepwise addition of taxa in PAUP*. The g1 statistic of skewed tree length distribution calculated from 104 random parsimony trees was used to assess the amount of nonrandom structure in the data [81].
Constraint analysis restricts the search of optimal trees to pre-specified tree topologies defining specific monophyletic groups, and was used here to test alternative or compare non-mutually exclusive hypotheses. The number of additional steps (S) required to force (constrain) particular taxa into a monophyletic group was examined using the “enforce topological constraint” option of PAUP*. The additional steps define an evolutionary distance that can be use to test alternative phylogenetic hypotheses or to compare hypotheses that are not mutually exclusive. The latter approach was used to construct evolutionary timelines, in which lower S values corresponded to ancient tRNAs, a trend that was derived from the rooted trees (and embedded assumptions of polarization). Constraint analyses were conducted based on amino acid specificity or grouping of molecules by organismal superkingdoms or viruses. |
10.1371/journal.pgen.1005806 | Quantitative Genetics Identifies Cryptic Genetic Variation Involved in the Paternal Regulation of Seed Development | Embryonic development requires a correct balancing of maternal and paternal genetic information. This balance is mediated by genomic imprinting, an epigenetic mechanism that leads to parent-of-origin-dependent gene expression. The parental conflict (or kinship) theory proposes that imprinting can evolve due to a conflict between maternal and paternal alleles over resource allocation during seed development. One assumption of this theory is that paternal alleles can regulate seed growth; however, paternal effects on seed size are often very low or non-existent. We demonstrate that there is a pool of cryptic genetic variation in the paternal control of Arabidopsis thaliana seed development. Such cryptic variation can be exposed in seeds that maternally inherit a medea mutation, suggesting that MEA acts as a maternal buffer of paternal effects. Genetic mapping using recombinant inbred lines, and a novel method for the mapping of parent-of-origin effects using whole-genome sequencing of segregant bulks, indicate that there are at least six loci with small, paternal effects on seed development. Together, our analyses reveal the existence of a pool of hidden genetic variation on the paternal control of seed development that is likely shaped by parental conflict.
| In plants and mammals, embryo development occurs under the protection and nourishment of maternal tissues. In polygamous species, this can lead to competition between siblings for privileged access to maternal nutrients. According to the parental conflict theory—a variation of the kinship theory—the asymmetric genetic relatedness between offspring from multiple fathers may lead to the evolution of parent-of-origin-dependent developmental regulation; paternally inherited alleles would benefit from maximizing embryo growth (at the expense of siblings), whereas maternally inherited alleles would benefit from restraining growth (to equalize sibling resource allocation). The kinship theory assumes that the paternal genome can actually influence embryo development; however, plant seed development is under strong maternal control. Here, we show that there is a hidden pool of variation in paternal effect loci that can be released upon loss of MEDEA, a major maternal regulator of seed development. Our results demonstrate that the maternal genome actively buffers the effects of paternal genomes on seed development, thereby providing strong functional support to the parental conflict theory.
| Post-fertilisation development is a complex process that involves dynamic interactions between maternally and paternally derived genomes. A correct balancing of parental genomes is essential for embryonic development, and disruptions of this balance (e.g. by crossing individuals with different ploidies) often lead to embryo inviability [1–6]. Genomic imprinting, an epigenetic mechanism that leads to differential expression of alleles in a parent-of-origin-dependent manner, is responsible for many parental asymmetries during embryo and seed development in mammals and flowering plants [7,8].
Transcriptome profiling of developing seeds has revealed the existence of hundreds of candidate imprinted genes in the embryo and/or endosperm, a biparental nourishing tissue that derives from a second fertilisation event (reviewed in [9–11]). However, the functional role of genomic imprinting is still a matter of considerable theoretical debate [12]. The parental conflict (or kinship) theory of genomic imprinting proposes that imprinting can evolve as the manifestation of a conflict of interests between maternal and paternal alleles over resource allocation during embryogenesis or seed development [13–15]. This conflict arises due to the asymmetric genetic relatedness between maternal and paternal alleles in polyandrous (multiple paternity) species, where maternal alleles are more likely to be shared between siblings than paternal alleles.
The parental conflict theory is supported not only by mutant phenotypes in mice [16–18] but also by the discovery of MEDEA (MEA), a major regulator of imprinting and the maternal control of seed development in Arabidopsis thaliana (L.) Heynh (referred to as Arabidopsis hereafter). Only the maternal MEA allele is expressed (before fertilization in the embryo sac that contains the female gametes and later in the embryo and endosperm derived from these gametes) [19,20], and seeds that maternally inherit a loss-of-function mea allele undergo excessive cell proliferation and eventually abort [21]. MEA encodes a SET-domain histone methyltransferase that catalyses the trimethylation of H3K27—a repressive epigenetic mark associated with gene silencing—as part of a seed-specific version of the Polycomb Repressive Complex 2 (FIS-PRC2) [22]. This suggests that an important function of MEA is to maternally restrain seed growth by negatively regulating the expression of genes that would otherwise promote embryo and endosperm growth.
While the parental conflict theory predicted the existence of maternal regulators of seed development such as MEDEA (MEA), the paternal genotype has no or very small effects on seed growth [23–30]. Here we show that there is a large hidden pool of natural variation in the paternal control of seed development that can be exposed using a maternal mutant mea background. Using a combination of classic quantitative trait analysis and a novel method for whole-genome sequencing of bulk segregants (Bulk-Seq), we determined that at least six loci contribute to the paternal rescue of mea seeds. Together, our results indicate that there is a large pool of natural variation in loci exerting paternal effects on seed development in Arabidopsis. These paternal effects are buffered by maternal MEA activity, suggesting that they were likely shaped by parental conflict.
When mea ovules are pollinated with wild-type pollen from the Landsberg erecta accession (hereafter referred to as Ler), seeds undergo excessive cell proliferation and abort before completing embryogenesis [21]. However, mea ovules pollinated with pollen from other Arabidopsis accessions (such as Cvi-0 or C24) can give rise to viable plump mea seeds (Fig 1A and S1 Fig). To dissect the relative paternal and maternal contributions to mea seed rescue, we introgressed mea-2 (originally in the Ler background) into Cvi-0 and C24. After six generations of backcrossing, we crossed three independent Cvi-0mea/MEA and C24mea/MEA lines with pollen from Ler, C24, and Cvi-0: all the pollinations made with Ler pollen resulted in high rates of seed abortion, whereas the pollinations made with Cvi-0 or C24 resulted in mostly viable plump seeds, independently of the genotype of the maternal plant used (Fig 1B). The magnitude of the Cvi-0 and C24 paternal rescue was modulated by the maternal genotype (e.g. in a Cvi-0mea/MEA maternal background the paternal effect of Cvi-0 was stronger and the effect of C24 was very weak). This suggests that the rescue of mea seeds is primarily a paternal-specific effect that can be partially modulated by the maternal genotype, indicating the existence of strong reciprocal interactions between the two parental genomes.
Since MEA is an imprinted gene (only its maternal allele is expressed) a potential explanation for mea seed rescue could be an activation of the paternal wild-type MEA allele. However, this hypothesis cannot easily be tested using allele-specific expression assays, because maternal mea mutations already induce low levels of paternal MEA expression [20,31–33]. To determine genetically if the paternal MEA allele is required for mea seed rescue, we examined the F2 progeny of crosses between mea-2 and different Arabidopsis accessions. If a paternal MEA allele was required for the rescue, we would not expect to recover viable homozygous mea/mea seeds. However, we recovered 9–20% of viable mea/mea plants in the F2 progeny of crosses with the accessions C24, Hs-0, and Lomm1-1 (S1 Table). This result clearly indicates that a paternal MEA allele is not required for mea seed rescue in these crosses. In the crosses with Cvi-0 we only recovered 3% of homozygous mea/mea seeds; when these different F2 homozygous Cvi-0mea/mea individuals were self-fertilized, however, we observed a range of 1–60% plump F3 seeds in their progeny (S2 Fig). This finding suggests that in mea crosses with Cvi-0, the paternal MEA allele (or a closely linked locus) can enhance but is not required for the rescue of mea seeds.
MEA encodes a subunit of the FIS-PRC2 complex, which also contains the zinc finger protein FERTILIZATION-INDEPENDENT SEED2 (FIS2) and the WD40 domain protein FERTILIZATION-INDEPENDENT ENDOSPERM (FIE) [22]. To test whether Cvi-0 and C24 can also rescue seed abortion caused by mutations in these genes, we crossed heterozygous fis2/FIS2 and fie/FIE plants with pollen from Ler, Cvi-0, and C24 (Fig 1C). While Cvi-0 could rescue fis2 seeds, there was no significant seed rescue using C24 pollen. fie seeds could not be rescued by pollen of either Cvi-0 or C24. These results indicate that the mea seed paternal rescue does not simply occur at the FIS-PRC2 level; rather it supports the hypotheses that the different FIS-PRC2 subunits play distinct roles [34] and that MEA participates in multiple protein complexes during seed development [32].
To determine the extent of species-wide variation on MEA-dependent parental interactions, we pollinated 164 Arabidopsis accessions with mea-2 to generate F1s that were allowed to self-fertilize to examine seed viability rates in the F2 generation. Each of the F2 populations segregates (1) MEA and mea, and (2) chromosomes from Ler and the respective parental accessions: therefore, we expected to obtain 50% viable seeds from accessions that do not modify the penetrance of mea (such as Ler), and up to 75% viable seeds from accessions with a strong paternal rescue effect (assuming no epistatic effects). Accordingly, we observed 52% plump seeds in the control Ler crosses, while in the Cvi-0 and C24 crosses there were 65% and 68% plump seeds, respectively (Fig 2A–2C). Roughly half of the accessions tested showed between 55% and 70% plump seeds, suggesting that alleles that modify the penetrance of mea seed abortion are widespread among natural Arabidopsis accessions.
Half of the strongest mea rescuers originate from latitudes more southern than 45° (Fig 2A), but we found no significant linear correlation between the geographical origin of the accessions and their effect on the penetrance of mea. We compared the mea rescue effect of these accessions with over 100 phenotypes reported for a large set of A. thaliana accessions [35]. The only statistically significant traits correlated with the rescue of mea were in planta magnesium and calcium concentrations (Pearson correlation, 5% false discovery rate) (S3 Fig). We also found evidence for an association between flowering time and mea rescue, as half of the strongest mea rescuers included very early flowering accessions under field or short day conditions (Cvi-0, C24, Se-0, Ts-1 and Co) (S3 Fig). We did not find a correlation between mea rescue and the size of self-fertilized or outcrossed Arabidopsis seeds [30,36].
We performed a genome-wide association study (GWAS) to identify regions in the genome whose species-level variation is linked to mea rescue. However, we were unable to detect clear statistically significant associations (Fig 2D), likely due to the weak power of GWAS to detect polygenic traits with low effect sizes [37]. Nevertheless, some of the most highly associated SNPs were in the vicinity of the regions identified with the Bulk-Seq analysis (see below).
We crossed homozygous mea/mea plants (generated using an inducible MEA-glucocorticoid receptor system) with pollen from 80 Cvi-0/Ler recombinant inbred lines (RILs) for which a detailed genetic map is available [38]. The percentage of plump seeds that originated from these crosses followed a continuous distribution (Fig 3A), indicating that the rescue of mea seeds is a polygenic trait. The broad-sense heritability H2 (the percentage of total phenotypic variance that can be explained by genetic factors) is 85%, indicating that mea seed rescue is under strong genetic control. We used maximum likelihood standard interval mapping to identify regions that are significantly associated with mea seed rescue. As expected from the continuous phenotype distribution, we identified multiple QTL peaks on several chromosomes (Fig 3B). Using a multiple-QTL approach [39], we narrowed down these regions to six QTLs, located on chromosome 1 (64.3cM and 101cM), chromosome 2 (65cM), chromosome 3 (77cM) and chromosome 5 (21 and 63.5cM) (Fig 3C). The six QTLs contribute independently to seed rescue (i.e. there was no evidence for epistatic interaction between QTLs) and together explain 73.1% of the phenotypic variance. Each QTL has a relatively small effect and explains a small proportion (5–11%) of the overall phenotypic variation (Fig 3D and Table 1). Nevertheless, the effect of multiple QTLs increases exponentially: every additional Cvi-0 QTL increases the rescued seed rate by roughly 50% (e.g. pollen donors with two, three or four Cvi-0 QTLs generate on average 18%, 28% or 42% plump seeds, respectively) (Fig 3E).
We then crossed mea/mea homozygous plants with pollen from a population of Cvi-0/Ler near-isogenic lines (NILs) [40]. Unlike RILs, which have mosaic genomes with a similar proportion of Ler and Cvi-0 genetic backgrounds, these NILs contain only one or a few small introgressions of Cvi-0 in an otherwise homogenous Ler genetic background. We used 33 NILs that together cover 93–98% of the genome with isogenic Cvi-0 fragments. While Cvi-0 pollen gave rise to 85% mea plump seeds, almost all the NILs showed no significant differences from Ler (3% viable seeds) (Fig 4A). The three NILs that clearly showed an effect (13–23% viable seeds) actually contain multiple Cvi-0 fragments that overlap two or three of the identified QTLs (Fig 4B). Together, the RIL and NIL analyses suggest the existence of at least six loci in Cvi-0 that contribute to the rescue of mea seeds.
We also scored seed abortion in the F2 progeny of a cross between mea-2 and C24 (genotyped at 14 markers throughout the genome). Despite the low statistical power caused by the segregation of MEA in this population, we found evidence for one QTL at the bottom of chromosome 1 that could explain 15% of the observed phenotypic variation (S4 Table). Thus, even this analysis with limited power identified one of the loci on chromosome 1 that was mapped using RILs and NILs.
To independently validate the results of the QTL analyses, we developed a novel method for mapping parent-of-origin effects using whole-genome sequencing. The strategy is to create an F2 population that contains one set of chromosomes from one parent but inherits two segregating sets from the other parent. These two sets should have opposing effects in pre- or post-fertilisation fitness or viability, so that they will not be equally transmitted. DNA is then extracted from pools of viable F2 seedlings, and whole-genome sequencing is used to identify genomic regions that exhibit biased transmission of the two segregating paternal (or maternal) genotypes.
In this case, we took advantage of the differential survival of mea seeds depending on the inheritance of Cvi-0 against Ler paternal alleles. First, we generated F1 hybrid plants by reciprocally crossing Ler and Cvi-0 plants (Fig 5A). The Ler/Cvi-0 hybrids were then used to pollinate (1) Ler plants and (2) homozygous mea/mea plants (Ler background). The resulting F2 progenies will therefore exclusively inherit Ler chromosomes from the mother (with and without mea) but different combinations of Ler and Cvi-0 alleles from the father (due to recombination and segregation of chromosomes during male gametogenesis in the F1 plant). Only mea F2 seeds that inherit Cvi-0 alleles that rescue mea are able to generate viable seedlings: therefore, genomic regions that are linked to these Cvi-0 alleles will be predominantly transmitted to viable F2 plants. We can identify these by sequencing pools of viable plants and quantifying the relative proportion of Cvi-0 and Ler SNPs throughout the genome. To account for biases in the paternal transmission of Cvi-0 and Ler SNPs that occur independently of mea, we determined the transmission of Cvi-0 SNPs in a control cross using wild-type Ler instead of mea plants. In this wild-type (WT) control ('WT pool'), we expect the percentage of Cvi-0 reads throughout the genome to be close to 25%; in the 'mea pool', regions that are associated with mea rescue will be enriched in Cvi-0 reads (up to 50%).
We pooled genomic DNA from a total of 2400 viable mea seedlings and 1400 WT seedlings in three biological replicates. We then used a dataset of known Ler and Cvi-0 polymorphisms [41,42] to estimate the proportion of Cvi-0 reads throughout the genome (Fig 5B and S4 Fig). In the WT pool, there is clear evidence for segregation distortion in several genomic regions, including a low proportion of Cvi-0 reads at the top of chromosome 3 and the middle of chromosome 1, and a high proportion at the bottom of chromosome 1 and the top of chromosomes 2 and 4. Most of these regions were previously shown to exhibit segregation distortion in crosses between Ler and Cvi-0 [38]. To identify the regions that are associated with mea seed rescue, we calculated the difference in the proportion of Cvi-0 reads between the mea and WT pools (Fig 5C). There was an overall increase in Cvi-0 reads throughout the genome in the mea pool, likely reflecting the highly polygenic nature of mea seed rescue; but the enrichment in Cvi-0 reads was particularly pronounced in the middle and bottom of chromosome 1, the bottom of chromosomes 2 and 3, and in the top and middle of chromosome 5: in these regions there was an increase of 15–30% in the proportion of Cvi-0 alleles relative to the WT pool (Fig 5C, Table 2). These peaks were reproducible between the three biological replicates (S4 Fig). Each of the peaks identified by Bulk-Seq is located in the vicinity of the QTLs identified by the RIL-QTL analysis (Table 1); some of the peaks (particularly b, d, and g) are also close to SNPs that were identified by the GWAS analysis as associated (although non-significantly) with mea rescue (Fig 2D). Taken together, the Bulk-Seq analysis provides strong support to the existence and predicted location of the multiple Cvi-0 alleles that underlie the rescue of mea seeds.
Our results demonstrate that there is a pool of hidden variation in the paternal regulation of seed development in Arabidopsis. This paternal variation is released upon maternal loss of mea, suggesting that the maternal genome actively buffers the manifestation of paternal effects during seed development. While in the past the effects of the paternal genotype on seed growth were found to be very small or non-existent [23–30], our results clearly indicate that paternal effects exist but are buffered by the maternal genome. This observation is consistent with predictions of the paternal conflict theory, which proposes that the maternal genome counteracts the effect of paternally inherited alleles that would otherwise place extra demands on seed growth.
We hypothesize that, in a maternal Ler background, the (potential) paternal growth demands of Cvi-0 and C24 are lower than the ones of most other accessions (including Ler itself). Upon maternal loss of the buffering mechanism mediated by MEA, the paternal growth demands of Ler (and most accessions) lead to excessive seed growth, resulting in mea seed collapse; however, the paternal growth demands of Cvi-0 and C24 are not as strong and allow mea seeds to complete development.
Interestingly, paternal effects on mea seed development are, in turn, dependent on the maternal genetic background: we showed that C24 paternal alleles can strongly rescue mea seeds in a maternal Ler or C24 background, but this rescue is much weaker in a Cvi-0 maternal background (Fig 1B). This indicates that there are multiple reciprocal interactions between maternal and paternal alleles in the regulation seed growth.
In many ways, this paternal variation is a classic example of cryptic genetic variation (CGV). Natural genotypes often harbour extensive CGV that is only released upon severe environmental or genetic perturbations [43–45]. Typical examples of CGV include variation in the number of Drosophila bristles in a scute mutant background [46], inflorescence architectures in maize crossed to its wild ancestor teosinte [47], body size in oceanic stickleback upon exposure to low salinity environments [48], or genetic background-dependent phenotypic variation upon disruption of the heat shock protein Hsp90 in Drosophila and Arabidopsis [49,50]. CGV usually has no or little effects on phenotypical variation, but it can modify phenotypes under atypical environmental conditions or following the introduction of novel alleles. By acting as a standing pool of genetic information, CGV has been hypothesized to play an important role in adaptation and the evolution of novel characters [51,52]. One explanation for the origin of CGV is that as new mutations appear, their potential phenotypic effect is suppressed by existing buffering mechanisms [44]. During Arabidopsis seed development, MEA could act as a buffering mechanism that prevents the expression of mutations that would otherwise disrupt the balance of paternal genomes.
Another possibility that could explain the hidden paternal variation is the predominantly self-fertilizing behaviour of Arabidopsis. Although imprinting can be maintained in species with a low outcrossing rate [53,54], high kin genetic relatedness is predicted to decrease the intensity of parental conflict [55]. The transition of Arabidopsis from an outcrossing to a self-fertilizing species around one million years ago [56], could have resulted in an erosion of the functional importance of imprinting mechanisms and the strength of the parental effects. This could make Arabidopsis seeds more resistant to unbalanced crosses and mask the manifestation of parental effects. Supporting this hypothesis, Arabidopsis seeds are unusually tolerant of unbalanced interploidy crosses [1,6].
Our genetic analyses demonstrate that multiple loci contribute independently to the paternal rescue of mea seeds, but the effect of each individual locus is small. We predict that the underlying genes encode factors that fine-tune embryo and endosperm growth during the early stages of embryogenesis, particularly endosperm cellularization and the transition from radial (globular stage) to bilateral (heart stage) embryo symmetry. We showed that the rescue is directional (e.g. Cvi-0mea x Ler seeds abort, but Lermea x Cvi-0 seeds are rescued). This suggests that the rescue is parent-of-origin-specific, and should therefore be meditated by imprinted genes. However, at this point we cannot distinguish whether the underlying alleles are paternally expressed or paternally repressed in Cvi-0. The MADS-box gene PHE1, a paternally expressed gene that is a direct target of MEA, is located close to the QTL peak at 101cM in chromosome 1 (peaks b and c in the Bulk-Seq analysis). phe1 mutants develop slightly lighter seeds and can partially rescue mea seeds [57–59], suggesting that variation in PHE1 may underlie this QTL.
mea seeds can be paternally rescued by the ddm1 mutant or an anti-sense met1 line [19,60]. Such lines have lower levels of CG methylation, which has been hypothesised to act antagonistically to H3K27 trimethylation to regulate the expression of paternally-derived alleles [61]. Interestingly, Cvi-0 has lower levels of CG methylation in embryos and endosperm than Ler or Col-0 [11]. However, we found no correlation between mea seed rescue and global levels of CG methylation in vegetative tissues of over 50 A. thaliana accessions [62].
The overgrowth phenotype of mea seeds strongly resembles the phenotype of interploidy crosses where the ploidy of the male is higher than that of the female [6,63], suggesting that MEA is an important contributor to maternal genome dosage. Accordingly, paternal excess crosses can be rescued by increasing the expression of MEA [63], while mea seeds are viable in maternal excess crosses [64]. Interestingly, hypomethylated pollen can also rescue seeds resulting from unbalanced crosses where the ploidy of the male is higher than the female [65], confirming the existence of overlaps between mechanisms that regulate parental dosage, Polycomb activity, and DNA methylation in developing Arabidopsis seeds.
Overall, we demonstrate here that there is a large pool of hidden intra-specific variation in the paternal control of seed development. Recent transcriptome studies have shown that 5–15% of A. thaliana and maize imprinted genes have allele-specific imprinting [11,66,67], while MEA has been found to be under positive selection in the genus Arabidopsis [68–70]. This suggests that the balancing of parental information during seed development is a very dynamic evolutionary process, and provides strong support to the parental conflict theory for the evolution of imprinting. Importantly, this standing pool of cryptic genetic variation in wild and domesticated species could have important uses in plant breeding programs [71] that aim to regulate seed size or overcome inter-specific hybridizations [72–75].
The mea-1 and mea-2 [21], fis2-1 [76] and fie (SALK_042962) [77,78] mutants, as well as the RIL and NIL Ler/Cvi populations [38,40] were previously described. The Landsberg erecta (Ler-1), Cape Verde Islands (Cvi-0), and C24 accessions used in this study are derived from lines N22618, N22614, and N22620, respectively, and were a gift of Ortrun Mittelsten Scheid (GMI Vienna). All plants were grown on standard soil (ED73, Einheitserde, Germany) in a greenhouse chamber with 16h light at 20°C and 8h dark at 18°C with an average of 60% humidity.
For seed viability assays, individually crossed siliques were harvested 1–2 days before dehiscence; seeds were then examined under a stereomicroscope and categorised as plump or aborted based on shape, size, and colour. For embryo quantification, seeds at different stages of development were fixed overnight at -20°C with 90% acetone, cleared with chloral hydrate/water/glycerol (8:2:1 w/v/v), and analysed under a Leica DMR microscope.
For the generation of Cvi-0mea/MEA and C24mea/MEA lines, mea-2/MEA plants (Ler background) were used to pollinate Cvi-0 and C24. The F1 progeny was selected in Murashige and Skoog (MS) medium supplemented with 50μg/ml kanamycin (mea-2 is marked by a kanamycin resistance gene) and backcrossed to Cvi-0 and C24, respectively. Individual backcrossed (BC1) kanamycin-resistant plants were used to pollinate Cvi-0 or C24 and generate independent BC2 lines. Backcrossing was performed for another four generations; BC6 plants were genotyped with a sequence length polymorphism marker linked to the MEA locus (3.286 Mbp distance) using primers 5'-AATTGAAGCTTTTCTGC-3' and 5'-AGAAAATGAAAAACTTATGG-3' to select plants with homozygous Cvi-0 introgressions close to mea. Seed abortion was scored in the progeny of a single BC6 individual from each of three independently generated Cvi-0mea/MEA and C24mea/MEA lines.
Cvi-0, Hs-0, C24, and Lom-1 were crossed with pollen from mea-2, the F1 populations were allowed to self-fertilize, and DNA was extracted from viable seedlings. mea-2 plants were genotyped using primers 5'-CCAATGCACAAATCGACAATG-3' and 5'-CACCAAGAGTGCCATCTCCA-3' (WT genomic DNA), and 5'-CGATTACCGTATTTATCCCGTTCG-3' (Ds insertion tightly linked to the mea-2 allele [21]).
To obtain pMEA::MEA-GR, an 8.6 kb genomic fragment (4.4kb upstream region + 4.2 kb MEA ORF) was amplified from the previously generated plasmid pCambia 1381Z [20] using primers proMEA-BPfor (AAAAAGCAGGCTCACTAAGATATGTTGGGTC) and MEA-BP-GRrev (AGAAAGCTGGGTCTGCTCGACCTGCCCGA), and recombined into pDONR207 using Gateway cloning (Invitrogen). The resulting entry vector was subsequently recombined into pDEST-GR (pARC146 without 35S promoter) [79], and the fusion between MEA and the GR domain sequence was confirmed by sequencing. The construct was introduced into the Agrobacterium strain GV3101::pMP90, and transformed into Ler plants using floral dip [80]. Lines expressing the construct were crossed with mea-1, and the offspring was continuously watered with 10 μM dexamethasone (DEX) (Sigma cat. D1756) to identify reduced seed abortion and thus rescue by the construct. Several lines were identified that complemented the mea-1 seed abortion phenotype to a large extent only in the presence of 10 μM DEX. The best complementing line (line 18) was used to raise the homozygous mea-1/mea-1 plants used in this study.
A total of 47,619 seeds derived from crosses between homozygous mea-1/mea-1 plants and 80 Ler/Cvi RILs were scored. The genotype map for these lines included 144 markers [38,81] and was kindly provided by Joost Keurentjes (Wageningen University and Research Centre). Broad-sense heritability was estimated with an analysis of variance of a linear mixed-effects model, using the lmer function of the 'lme4' R package [82]. Means and confidence intervals for each RIL were estimated using a binomial regression, and normalised using a cubic root transformation. QTL analyses were performed using the 'R/qtl' R package [39]. Genotype data across the genome was estimated using multiple imputation at a 1cM density, and interval mapping was calculated using standard maximum likelihood estimation; the LOD (logarithm of odds) threshold at 1% was calculated using a permutation test with 5000 replicates. Multiple-QTL models were selected using a combination of automated stepwise model selection and iterative individual QTL location refinement as implemented in 'R/qtl'; the penalized LOD scores used to guide model selection were derived using a permutation test with 1000 replicates of a two-dimensional, two-QTL genome scan. The best fit-model we identified had six loci and no epistatic interactions between the QTLs. Finer localizations of each of the QTLs of the best-fit model along the chromosomes were estimated using a Bayesian approach, as implemented in the bayesint function of 'R/qtl'. The effect of individual QTLs was estimated as the proportion of phenotypic variance they explain. The approximate physical location was estimated using the physical location of genetic markers [38,81].
For the mea-2 x C24 QTL analysis, an F2 population of 247 individuals was generated and genotyped using 14 sequence length polymorphism markers. An ANOVA regression was calculated for each marker: the values on S4 Table are the p-values for a regression made using a subset of 35 homozygous mea/mea plants.
Ler and Cvi-0 were reciprocally crossed to generate Ler/Cvi-0 hybrids. These were then used to pollinate homozygous mea/mea or Ler plants. Three independent replicates were generated, each using different parental individuals and at different days. In total, around 10,000 and 4,000 F2 seeds were generated from mea/mea and Ler mothers, respectively. Seeds were surface sterilised for 10 minutes using 1% sodium hypochlorite, washed extensively with water, and sown in MS medium. After 10–14 days of growth at 22°C, leaves from viable seedlings were collected (825, 800, and 775 seedlings for each of the WT pool replicates; 600, 400, and 400 individuals for each of the mea pool replicates). DNA was extracted in groups of 50 leaves using the DNeasy Plant Mini Kit (Qiagen cat. 69104), and quantified using the Qubit dsDNA HS Assay kit (Life Technologies cat. Q32854). DNA from the different extractions was pooled in equi-amounts, precipitated using sodium acetate and isopropanol, and resuspended in 25 μl TE buffer for each replicate at a final concentration of 100–150 ng/μl.
The TruSeq DNA Sample Prep Kit v2 (Illumina) was used in the succeeding steps. DNA samples (1 μg) were sonicated and the fragmented DNA samples end-repaired and polyadenylated. TruSeq adapters containing the index for multiplexing were ligated to the fragmented DNA samples. The ligated samples were run on a 2% agarose gel and the desired fragment length was excised (50bp +/- the target fragment length). DNA from the gel was purified with MinElute Gel Extraction Kit (Qiagen). Fragments containing TruSeq adapters on both ends were selectively enriched with PCR. The quality and quantity of the enriched libraries were validated using Qubit (1.0) Fluorometer and the Caliper GX LabChip GX (Caliper Life Sciences). The product is a smear with an average fragment size of approximately 260 bp. The libraries were normalized to 10nM in Tris-Cl 10 mM, pH8.5 with 0.1% Tween 20. The TruSeq SR Cluster Kit (Illumina) was used for cluster generation using 5 pM of pooled normalized libraries on the cBOT. Sequencing of single reads was performed on one lane of the Illumina HiSeq 2000 using the TruSeq SBS Kit v3-HS (Illumina Inc, USA).
Reads were quality-checked using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and trimmed with the FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/). Processed reads were aligned to the TAIR10 version of the Arabidopsis genome (Col-0 accession) using Bowtie2 [83] and the SAMtools package [84]. SNPs were called using the mpileup command from SAMtools. Known SNPs from Ler-1 and Cvi-0 [41,42] were retrieved from http://1001genomes.org/data/MPI/MPISchneeberger2011/releases/current//Ler-1/Marker/Ler-1.SNPs.TAIR9.txt (461,070 Ler-1/Col-0 SNPs) and from http://signal.salk.edu/atg1001/data/Salk/quality_variant_filtered_Cvi_0.txt (657027 Cvi-0/Col-0 SNPs). After removing common Ler/Cvi-0 SNPs, we obtained a list with 765,643 SNPs. This list was combined with the SNP calls from the bulk sequencing dataset; for each of the datasets (three replicates of each pool), we then removed positions that were (1) ambiguous in the reference sequence; (2) did not match between the Bulk-Seq samples and the published SNPs; and (3) had very high coverage (above the 99% percentile of a Poisson distribution with λ = median coverage of the sample). These quality-filtering steps resulted in an average of 390,000 SNPs for each of the six datasets. The three replicates were combined by summing up the number of reads at each SNP position. SNP positions that had no Ler or no Cvi-0 reads in the combined dataset were discarded, as were the top and low 1% quantiles of coverage. This resulted in a final combined dataset of 352,491 SNPs with an average read coverage of 16 Ler and 6 Cvi-0 reads, respectively (S3 Table). The same quality filtering steps were performed in each of the three replicates. To calculate the relative proportion of Cvi-0 and Ler reads across the genome, we first summed the read counts of groups of 50 neighbouring SNPs across the genome using a rolling window. We then calculated the proportion of Cvi-0 and Ler reads, and the relative enrichment of Cvi-0 reads in the mea pool relative to the WT pool:
%Cvi enrichment=%Cvimeapool−%CviWT pool%CviWT pool
Finally, we smoothed Cvi enrichment across neighbouring positions by computing the median of Cvi-0 enrichment with a rolling window of 100 SNPs.
We used mea-2 to pollinate 167 Arabidopsis accessions obtained from the Nottingham Arabidopsis Stock Centre (NASC) or as a kind gift from Ortrun Mittelsten Scheid (Gregor Mendel Institute) and Takashi Tsuchimatsu (University of Zurich). Accessions are detailed in S1 Table. The F1 progeny was selected in MS medium supplemented with 50μg/ml kanamycin and allowed to self-fertilize. We collected individual fruits 1–2 days before dehiscence and visually scored seed phenotypes; some fruits had a high proportion of autonomous seeds (mea autonomous endosperm development depends on the genetic background [85]); to avoid a bias in the aborted/plump seed ratio calculations, we did not use fruits that had more than 8% autonomous seeds (225 out of 2046 fruits; there was no correlation between mea rescue and autonomous seed development). We discarded three outliers (accessions with a low number of scored seeds or relatively high percentage of autonomous seed development) and obtained a final dataset consisting of 93,884 seeds from 164 accessions. Genotypic information (250k snp data v3.06) was downloaded from https://cynin.gmi.oeaw.ac.at/home/. For correlation analysis with the dataset of 107 phenotypes [35], we calculated Pearson correlations and corrected p-values for multiple testing using the Benjamini-Hochberg false discovery rate procedure.
Genome-wide association mapping was performed with compressed mixed linear models [86] implemented in the 'GAPIT' R package [87] and in the web-based GWAPP portal [88] using the proportion of plump seeds as a phenotype (cubic root normalised). Plots were generated using the 'ggplot2' R package [89].
|
10.1371/journal.pgen.1006717 | Larval crowding accelerates C. elegans development and reduces lifespan | Environmental conditions experienced during animal development are thought to have sustained impact on maturation and adult lifespan. Here we show that in the model organism C. elegans developmental rate and adult lifespan depend on larval population density, and that this effect is mediated by excreted small molecules. By using the time point of first egg laying as a marker for full maturity, we found that wildtype hermaphrodites raised under high density conditions developed significantly faster than animals raised in isolation. Population density-dependent acceleration of development (Pdda) was dramatically enhanced in fatty acid β-oxidation mutants that are defective in the biosynthesis of ascarosides, small-molecule signals that induce developmental diapause. In contrast, Pdda is abolished by synthetic ascarosides and steroidal ligands of the nuclear hormone receptor DAF-12. We show that neither ascarosides nor any known steroid hormones are required for Pdda and that another chemical signal mediates this phenotype, in part via the nuclear hormone receptor NHR-8. Our results demonstrate that C. elegans development is regulated by a push-pull mechanism, based on two antagonistic chemical signals: chemosensation of ascarosides slows down development, whereas population-density dependent accumulation of a different chemical signal accelerates development. We further show that the effects of high larval population density persist through adulthood, as C. elegans larvae raised at high densities exhibit significantly reduced adult lifespan and respond differently to exogenous chemical signals compared to larvae raised at low densities, independent of density during adulthood. Our results demonstrate how inter-organismal signaling during development regulates reproductive maturation and longevity.
| The nematode C. elegans is one of the most highly developed models for the elucidation of conserved mechanisms connecting environmental cues to the regulation of animal lifespan and development. Surprisingly, the effects of larval population density on developmental timing and adult lifespan have not been investigated, although population density is known to affect developmental dynamics and survival in many species. We here describe a novel phenotype in C. elegans: population density-dependent acceleration of development. That high population density would accelerate development is unexpected, since at high population density accumulation of dauer pheromone, a developmental arrest signal, would be expected to slow down development. However, we found that C. elegans development is regulated by a pull-push mechanism, based on at least two different types of pheromone-like signals: the developmental acceleration signal we first describe in this manuscript, and its antagonist, the dauer pheromone, whose chemical make-up has gradually emerged over the past 10 years. We further show that both developmental acceleration and deceleration are mediated by two nuclear hormone receptors that have close mammalian homologs. Finally we demonstrate that larval population density predetermines adult lifespan in C. elegans hermaphrodites, including responses to hormonal stimuli during adulthood.
| Sensing environmental conditions is of central importance for organismal development and survival and has been recognized as a major driver of adaptive evolution [1,2]. The nematode Caenorhabditis elegans is a particularly useful model for studying the effects of complex environmental inputs on development and lifespan [3,4], and C. elegans has become one of the most important models for studying conserved mechanisms of aging. Several lines of evidence indicate that development and lifespan in C. elegans are controlled by interorganismal signaling. High population density under conditions of dietary restriction triggers arrest of development as dauer larvae, a long-lived, highly stress resistant alternate larval stage that can persist under adverse environmental conditions for many months [4]. The dauer-inducing population density signal was shown to consist of a synergistic mixture of small molecules, the ascarosides [5–7]. Perception of dauer-inducing ascarosides downregulates insulin, cGMP and TGF-β signaling, which in turn downregulates expression of enzymes involved in the biosynthesis of steroid hormones called dafachronic acids (DAs), the endogenous ligands of the nuclear receptor DAF12, a homolog of vertebrate vitamin D and liver-X receptors [8,9].
Binding of DAs to DAF-12 is required for normal reproductive development, whereas abolishment of DA biosynthesis results in unliganded DAF-12 triggering larval arrest at the long-lived and non-feeding dauer stage [9,10]. Parallel work showed that C. elegans and other nematodes produce a large diversity of ascaroside-based small molecules, the nematode-derived modular metabolites (NDMMs), which are derived from combinatorial assembly of chemical building blocks from lipid, amino acid, carbohydrate, citrate, neurotransmitter, and nucleoside metabolism [7,11]. NDMMs control many aspects of nematode life history, including mating and aggregation behaviors as well as adult lifespan.
Whereas dauer formation and associated signaling pathways have been investigated in great detail, other effects of population density on developmental progression and adult lifespan remain poorly understood. For example, the lifespan of C. elegans males starkly decreases at high worm densities, and the presence of male worms has been shown to accelerate aging in hermaphrodite worms [12,13]. Furthermore, decreased sensory stimulation at low worm densities has been shown to slow down development and the begin of egg laying [14]. Based on serendipitous observations during worm synchronization for aging studies, we hypothesized that population density may regulate the rate of developmental maturation in C. elegans via an excreted chemical signal. Using the time point of first egg laying as a marker for full developmental maturity we show that developmental rate in C. elegans is controlled by a push-pull mechanism that relies on several different types of interorganismal small-molecule signals that are excreted by the worms. We show that high population conditions accelerate development by an as-yet unidentified chemical signal, and that this effect is counteracted by specific ascaroside pheromones and DAs. In addition, we found that population density experienced during larval development strongly affects adult lifespan.
To facilitate comparing developmental rates across different conditions and genotypes, we considered using the time point of first egg laying as an easily observable marker for reproductive maturation. Egg laying has been used previously as part of a suite of assays to characterize the effects of decreased sensory stimulation on behavior and neuronal development in C. elegans [14]. In an initial experiment, we measured the time point of first egg laying on plates with 1 to ~900 worms per plate (Fig 1A, S1 Table). Results from this assay suggested that the time point of first egg lay is strongly dependent on the number of worms per plate; however, the chance of observing an egg on a plate obviously increases with the number of worms per plate, which made it difficult to ascertain the significance of the effect.
Therefore, we established a more elaborate protocol (Fig 1B) in which we raised worms from eggs, either in isolation (one worm per plate, "ISO worms") or on high density plates ("HD worms", raised at ~100 worms/plate), and then transferred both ISO and HD worms at late L4 larval stage (after 59 h development on ISO or HD plates), a few hours before egg laying would start, onto fresh plates, at one worm per plate. Using this protocol we confirmed that HD worms started egg laying significantly earlier than ISO worms (Fig 1C, S1 Table), whereby the magnitude of the effect depends on the amount of time the worms were kept at different density before transfer (Fig 1C, S1 Fig, S1 Table, S2 Table). To assess whether population density-dependent egg laying is associated with overall faster development of HD worms, we compared vulva development and outgrowth of the gonad arms of ISO and HD worms 52 h after hatching. Whereas ISO worms were predominantly in the early larval stage 4 (L4), most HD worms had developed to mid L4 stage (Fig 1D, S1 Fig, S3 Table), corresponding to a 2 to 3 h developmental lead of HD animals. We also tested whether ISO and HD worm differed with regard to embryo volume and overall fecundity. We found that the total number of progeny on the first day of egg laying was increased in HD worms (S2 Fig); however, total numbers of eggs laid were similar for ISO and HD worms (S2 Fig). Taken together, these observations suggest that earlier egg laying of worms raised at high densities is indicative of overall faster development. We refer to this phenotype as population density-dependent acceleration of development (Pdda).
Next, we screened genes in several pathways related to the regulation of development and lifespan for their potential roles in Pdda. Because developmental times of different mutants vary considerably, we focussed on differences in time points of first egg laying between ISO and HD worms relative to the time difference observed in ISO and HD wildtype worms, which we report in percent of the effect observed for wildtype (Figs 1–3, also see Methods). First, we considered the possibility that differences in food intake may underlie Pdda. Although the bacterial lawn on HD plates did not appear to be significantly depleted by the time worms reached adulthood, it seemed possible that Pdda is caused by differences in feeding behavior or food availability, which have been shown to affect many aspects of development and germline proliferation [15]. Therefore we compared developmental rates of HD and ISO eat-2 mutant worms, which exhibit reduced pharyngeal pumping rates [16]. We found that, although eat-2 mutant worms start laying eggs later than wild type, the time difference between the start of egg laying of eat-2 ISO and HD worms, relative to total time to maturity, is similar to that of wildtype (Fig 1E, S4 Table). Pdda was also unchanged in egl-4 mutant worms, which are egg laying-defective due to egg retention [14] (Fig 1E, S4 Table), supporting that Pdda is unrelated to egg laying defects or inappropriate egg retention. In addition, Pdda was not affected by mutations of the NADH dependent histone deacetylase SIR-2.1, which may play a role for some aspects of dietary restriction-mediated lifespan changes [3] and is required for ascaroside-mediated increased lifespan and stress resistance [17], the nuclear receptor NHR-49, which is required for entering the starvation-induced adult reproductive diapause (ARD, [18]), and the transcription factor heat shock factor-1 (HSF-1), a general regulator of stress-induced gene expression [19] that is also required for aspects of dietary restriction-mediated lifespan extension [20] (S3 Fig, S4 Table). Taken together, these results suggest that differences in food intake are unlikely to be the primary cause of the observed population-density dependent effects on development.
We were initially surprised by the finding that C. elegans develop faster under HD conditions than in isolation, given that dauer-inducing ascarosides accumulate at much higher concentrations on HD plates [7], which we hypothesized would slow down, but not accelerate development. Alternatively, given the wide range of functions that chemically different ascarosides serve in C. elegan s[11], it also seemed possible that development on HD plates is faster because of accumulation of specific ascarosides. To clarify whether ascaroside production plays a role for Pdda, we investigated the effect of population density on development of daf-22 and dhs-28 mutant worms. daf-22 and dhs-28 encode two enzymes of the peroxisomal β-oxidation pathway that are required for the biosynthesis of all known ascaroside-based signaling molecules in C. elegans [21], and therefore daf-22 and dhs-28 mutant worms are devoid of all known ascaroside-based signalling molecules. We found that Pdda is starkly increased in daf-22 and dhs-28 mutants, with daf-22 and dhs-28 HD worms developing 8–17 h faster than ISO mutant worms, compared to a time difference of typically 2–5 h for WT (Fig 2A and 2B, S7 Fig, S5 Table). The strong enhancement of Pdda in ascaroside biosynthetic mutants distinguishes this phenotype from the effect of males on hermaphrodite lifespan (male-induced demise, MID) [13], which is partially abolished in daf-22 mutants.
Next we tested the effect of adding synthetic ascarosides to both HD and ISO plates. We found that Pdda in WT is partially suppressed by physiological concentrations of the strongly dauer inducing ascaroside ascr#2 [7] (Fig 2C and 2D, S7 Table). Supplementation of ascaroside-defective daf-22 worms with synthetic ascarosides also suppressed the starkly enhanced Pdda phenotype in this mutant (Fig 2E, S7 Table). These results suggested that accumulation of dauer-inducing ascarosides in the case of wildtype worms results in less Pdda compared to the ascaroside-free mutant daf-22.
Ascarosides are perceived via several ciliated sensory neurons [6,22,23], and therefore impairment of chemosensation should also remove the Pdda-suppressing effect of ascarosides. We tested mutants of osm-6, which is required for proper cilia formation in sensory neurons [24] and tax-4, which encodes a subunit of a cyclic nucleotide-gated channel expressed in ciliated sensory neurons where it localizes to the cilia [25]. Previous work has shown that both osm-6 and tax-4 mutants are defective in their responses to dauer-inducing ascarosides [25]. We found that Pdda is strongly increased in both osm-6 and tax-4 mutants, to a similar extent as in peroxisomal β-oxidation mutants (Fig 2F, S8 Table). We also showed that in tax-4 and osm-6 mutants ascr#2 does not affect Pdda, whereas ascr#2 abolishes Pdda in wildtype (S4 Fig, S9 Table). In contrast, in worms mutant for odr-3, which encodes a G-protein alpha subunit that is required for general chemotaxis [26], but not for dauer pheromone perception, Pdda was unchanged compared to wildtype (Fig 2F, S8 Table). Taken together, our results indicate that chemosensation of dauer-inducing ascarosides reduces the effect of population density on developmental rate, and that, in wildtype, dauer-inducing ascarosides partially mask the effects of another population-density dependent factor that accelerates development and represents the proximal cause of Pdda.
Accelerated development at high population density could be due to increased mechanosensory stimulation, which has been shown to affect egg laying [14,27], or result from accumulation of excreted or secreted small molecules other than ascarosides. Pdda was not significantly reduced in mechanosensory defective mec-4 mutant worms (Fig 2F, S8 Table). To test whether excreted small molecules are responsible for Pdda, we conditioned plates with HD daf-22 mutant worms (grown from larval stage L1 to L4), removed these worms, and then compared development of daf-22 ISO worms on HD daf-22-conditioned plates with daf-22 ISO and HD worms on regular plates (Fig 2G, S6 Table). We found that daf-22 ISO worms on HD-conditioned plates developed faster than daf-22 ISO worms on regular plates, though not as fast as daf-22 HD worms. Similarly, isolated wildtype worms grown on plates that have been pre- incubated with N2 larvae started egg laying significantly earlier than isolated N2 worms grown on non pre-incubated plates (S5 Fig, S6 Table). Thus, it appears that one or more compounds excreted by worms on HD plates accelerate development.
Downstream of ascaroside perception, the insulin/IGF signaling pathway plays a central role in regulating C. elegans development and lifespan, in part by regulating the biosynthesis of the dafachronic acids (DAs), the steroidal ligands of the nuclear receptor DAF-12 [8,9,28,29]. The DAs are produced from dietary cholesterol via a multi-step enzymatic pathway that is being studied extensively [30]. Three different endogenous DA's have been identified, named dafa#1-dafa#3 (Fig 3A), and binding of DAs to DAF-12 suppresses developmental arrest at the dauer stage and promotes reproductive development [31–33]. Notably, higher concentrations of DA have been shown to accelerate development [34]. One of the last steps of the biosynthesis of the DAs requires the cytochrome P450, DAF-9, and, correspondingly, developing daf-9 null mutant worms arrest as dauer-like larvae, unless supplied with exogenous DA [35]. DAF-9 expression, and thus DA biosynthesis, is downregulated in mutants of the insulin/insulin-like growth factor receptor DAF-2, whereas DAF-9 expression and DA biosynthesis are increased in knockout mutants of daf-16, the FOXO transcription factor acting downstream of daf-2 [8,34]. We found that Pdda is abolished in both daf-2 and daf-16 mutants, in contrast to many other aging- or developmental phenotypes [29] (Fig 3B, S10 Table, S11 Table). We also tested a null mutant of the insulin-like peptide ins-11, which is partially required for MID [13]. We found that Pdda is unchanged in ins-11 mutants, providing additional evidence that hermaphrodite population density affects development in a manner that is distinct from the effects of males on lifespan (Fig 3B, S10 Table). Abolishment of Pdda in daf-2 as well as daf-16 mutants argued that while insulin/IGF signaling plays a role, it may not act exclusively via regulation of DA production [8,9,36], which is oppositely regulated in daf-2 and daf-16 mutants.
To determine whether DAF-12 or DAs are required for Pdda, we tested the null-allele daf-12(rh61; rh411) and the double mutant daf-9(dh6); daf-12(rh61; rh411), which does not produce any DAs. We found that Pdda is similar to wildtype in daf-9(dh6); daf-12(rh61 rh411) double mutant worms and slightly reduced in daf-12(rh61 rh411) worms (Fig 3C, S12 Table). These results indicated that DA signaling via DAF-12 is not required for Pdda. However, we found that exogenous addition of DAs largely abolished Pdda in wild type as well as the enhanced Pdda in daf-22 mutant worms (Fig 3D–3F, S7 Fig, S14 Table). Exogenous addition of DAs had no significant effect on Pdda in daf-9(dh6);daf-12(rh61 rh411) double mutant worms, indicating that abolishment of Pdda in wildtype is in fact dependent on DAF-12 (Fig 3D, S15 Table).
Recent work has shown that steroid hormone signaling in C. elegans also involves NHR-8, a nuclear hormone receptor that is closely related to DAF-12. Although DAs are not known to act as ligands of NHR-8, DA signaling through NHR-8 is required for aspects of dietary-restriction-mediated longevity [37] and regulates cholesterol and DA homeostasis [38]. We asked whether NHR-8 plays a role for Pdda, or for abolishment of Pdda by DAs. We found that Pdda is largely abolished in the loss-of-function mutants nhr-8(ok186) and nhr-8(tm1800) (Fig 3C). Next we tested the effect of one specific DA, dafa#3, on nhr-8 mutants. Surprisingly, whereas dafa#3 abolished Pdda in wildtype, we found that in nhr-8 null mutants dafa#3 largely restored Pdda (Fig 3E, S15 Table). Because NHR-8 loss changed the response to dafa#3, a ligand of DAF-12, we then tested the double mutant, nhr-8(ok186); daf-12(rh61;rh411). We found that Pdda is strongly increased in nhr-8; daf-12 double mutant worms, to a similar extent as in ascaroside-deficient daf-22 worms (Fig 3F and 3G, S7 Fig, S15 Table). Addition of dafa#3 did not change Pdda in nhr-8;daf-12 worms, consistent with the notion that DAs act via binding to DAF-12 (Fig 3F, S8 Fig, S15 Table).
The strong Pdda observed for nhr-8;daf-12 double mutant worms further provided the opportunity to corroborate that accumulation of worm-deposited chemical signals underlie this phenotype. For this purpose we compared time points of first egg laying of HD and ISO nhr-8;daf-12 double mutant worms with the time point of first egg laying of ISO nhr-8;daf-12 double mutant worms raised on plates pre-incubated with daf-22 worms. We found that ISO nhr-8;daf-12 double mutant worms raised on daf-22-pre-incubated plates developed as fast as HD nhr-8 daf-12 double mutant worms (Fig 3H, S6 Table). This result confirms that Pdda is due to accumulation of a worm-deposited chemical signal and that this signal is responsible for the enhanced Pdda phenotype of both peroxisomal β-oxidation mutant and nhr-8;daf-12 double mutant worms. Taken together, our results show that NHR-8 plays a central role for Pdda, and that although DAF-12 is only partially required for Pdda, developmental timing is affected by DAs in an NHR-8- and DAF-12-dependent manner.
Previous studies demonstrated that population density affects C. elegans lifespan in a sex-specific manner. Lifespan of C. elegans males starkly decreases at high worm densities, whereas population density did not have a significant effect on the lifespan of hermaphrodites [12]. In addition, the presence of male worms has been shown to starkly accelerate aging in hermaphrodite worms (male-induced demise, MID [13]). Based on our finding that population density affects the rate of larval development, we decided to re-investigate the effect of population density on hermaphrodites. For this purpose, we used two different experimental setups. In protocol I, we transferred hermaphrodite worms from high-density plates at the young adult stage onto the experimental plates, at densities of 1, 10, 20, and 50 worms per plate. This setup closely mimics the experimental design of the earlier study of Gems and Riddle [12] and most published aging studies. In protocol II, we set up plates with 1, 10, 20, and 50 eggs, and worms were kept at these population densities throughout development and adulthood. Under protocol I, population density did not significantly affect adult lifespan (Fig 4A, S16, S17, S18, S19 and S20 Tables), in agreement with earlier results [12]. In contrast, for worms set up under protocol II, mean adult lifespan was negatively correlated with population density. For example, the mean lifespan of isolated worms was about 5 days longer than that of worms that were raised in groups of 50 worms (Fig 4A, S8 Fig, S16, S17, S18, S19 and S20 Tables). Comparing the lifespan of isolated worms (1 worm/plate), worms raised in isolation from hatching lived almost 30% longer than worms raised on high density plates before isolation at the young adult stage. These results indicate that population density experienced during larval development contributes significantly to determining hermaphrodite lifespan.
Next we tested the effect of added dafa#3 on the adult lifespan of hermaphrodites at different population densities. Previous studies indicate that added DAs do not directly affect the adult lifespan of hermaphrodites [31]. In contrast, we found that dafa#3 significantly decreased the median lifespan of hermaphrodites at low population densities, when set up as eggs. Whereas the lifespan of hermaphrodites raised at a density of 20 or more worms per plate was similar to that of mock treated controls, lifespan of isolated worms on dafa#3-treated plates was about 20% shorter than for mock-treated worms (Fig 4A and 4B, S17 Table). This indicates that larval exposure to dafa#3 reduces adult lifespan of isolated worms. These results do not directly contradict earlier studies of the effects of DAs on adult lifespan [31], since these studies were performed with 20–40 worms per plate. At these worm densities, accumulation of other signals, for example ascarosides or the chemical signals that underlie Pdda, may mask the lifespan-shortening effect of DAs. Larval exposure to dafa#3 thus counteracts the effects of population density on both larval development and adult lifespan.
The nematode C. elegans is an important model for the study of metazoan development and aging. However, the effects of population density on developmental progression and adult lifespan have not been studied extensively, except for conditions that induce developmental arrest at the L1 or dauer stage. Our results demonstrate that C. elegans hermaphrodite development is regulated by larval population density. The finding that developmental acceleration at high population densities is starkly enhanced in ascaroside-deficient peroxisomal β-oxidation and chemosensory mutants indicates that in wildtype worms, Pdda is partially suppressed or antagonized by population density-dependent accumulation of dauer-inducing ascarosides (for a simplified model, see Fig 4C). We further show that Pdda is due to population density-dependent accumulation of a worm-deposited chemical signal. Therefore, it appears that C. elegans development under non-dauer inducing conditions is regulated by a push-pull mechanism based on at least two different types of pheromone-like signals: (i) chemosensation of ascarosides, which slows down development, possibly via the same pathways known to induce dauer entry by downregulating steroid hormone (DA) biosynthesis, including the daf-2 insulin/insulin-like growth factor and the daf-7 TGF-β signaling pathways [9], and (ii) a second pheromone-like chemical signal, named compound X in Fig 4C, that accelerates development and whose population density-dependent accumulation is the proximal cause of Pdda.
In contrast to ascarosides, compound X does not appear to be perceived via the cilia of sensory neurons, given that cilia-defective osm-6 mutants, which do not respond to ascarosides, show increased Pdda, similar to the enhanced Pdda phenotype in ascaroside-defective daf-22 or dhs-28 mutants (Fig 2). Abolishment of Pdda in both daf-2/insulin receptor and daf-16/FOXO mutants argues that insulin/IGF signaling plays a complex role for Pdda, as could be expected given that both dauer-inducing ascarosides and dafachronic acids reduce Pdda (Figs 2D and 3D–3F, S6 Fig) and mutations of daf-2 and daf-16 affect both DA production [34,36] and sensitivity to ascarosides [8,9]. Furthermore, developmental acceleration by compound X may interact with the insulin/IGF signaling pathway via DA- and ascaroside-independent mechanisms.
Significantly, our results indicate that the effects of the Pdda signal on development involve two nuclear hormone receptors, DAF-12, which has been studied in great detail as the master switch between dauer arrest and reproductive development[9,39], and the much less well characterized NHR-8. Like most nuclear receptors in C. elegans and many NHRs in higher animals, NHR-8 remains to be de-orphaned, that is, no endogenous ligands have been identified. However, NHR-8 has been shown to be involved in cholesterol and DA homeostasis [38] and, more recently, this nuclear receptor been shown to play a role for dietary restriction-mediated longevity [37]. Notably, NHR-8 was shown to act downstream of DA biosynthesis in this pathway, even though there is no evidence that DAs activate or directly bind to this nuclear receptor. Our observation that Pdda is starkly enhanced in the nhr-8; daf-12 double mutant indicates that the pheromone-like signal compound X does not represent a ligand of either NHR-8 or DAF-12. However, given that Pdda is fully abolished in nhr-8 null mutants, enhanced in the nhr-8; daf-12 double mutant, but similar to wild type in daf-12 and daf-9; daf-12 mutants, NHR-8 appears to play a central role for the regulation of developmental timing. Particularly curious is the finding that even though Pdda is abolished in nhr-8 null mutants and abolished in wildtype by addition of exogenous DA, the addition of exogenous DA in nhr-8 null mutants largely restores Pdda, that is, the effect of added DA is opposite in nhr-8 mutants compared to wildtype. It is not known whether NHR-8 and DAF-12, like their closest mammalian homologs, e.g. the vitamin D receptor (VDR), act as heterodimers. In mammals, VDR forms a heterodimer with the retinoid X receptor (RXR), and binding of their respective endogenous ligands to either VDR or RXR directly affects recruitment of transcriptional coregulators by the unliganded partner receptor [40]. Correspondingly, it seems possible that Pdda is mediated in part via direct interactions of NHR-8 and DAF-12 within a heterodimer. Clarification of the mechanisms by which NHR-8 regulates developmental timing will benefit from identification of endogenous ligands of this NHR, in addition to identification of compound X as the primary cause of Pdda. Taken together, our results show that the role of nuclear hormone receptors for regulating developmental timing extends beyond DAF-12 and the dafachronic acid signalling cascade.
Inspired by earlier work on the effects of the population density of C. elegans males on male [12] and hermaphrodite lifespan ("MID") [13], we also investigated whether population density affects adult lifespan in C. elegans hermaphrodites. We found that high larval population density, but not high population density experienced during adulthood, significantly reduces hermaphrodite lifespan. In this regard, population density-dependent shortening of hermaphrodite lifespan differs from MID, since male-derived small molecules shorten hermaphrodite lifespan through exposure during adulthood [13]. It is unclear whether Pdda and the effects of larval population density on adult lifespan derive from a common molecular mechanism, e.g. whether compound X is responsible for the reduced lifespan of hermaphrodites raised under crowded conditions, and to what extent accumulation of lifespan-extending ascarosides, e.g. ascr#2 and ascr#3 [17], antagonize hermaphrodite progeria on high density plates. However, our lifespan assays with added DA demonstrate that larval population density not only regulates hermaphrodite adult lifespan, but also affects responses to added small molecules, suggesting that responses to other environmental stimuli or genetic interventions may be affected as well. Therefore, differences in larval population densities prior to set-up of the actual aging assays, generally at the young adult stage, provides a potential explanation for conflicting findings about the roles of specific environmental and genetic factors on lifespan in C. elegans.
Our work further shows that measuring onset of egg laying as a marker for full maturity represents an effective means to investigate the effect of environmental or genetic changes on development. Developmental maturation in animals is highly plastic, and a great variety of environmental factors have been shown to affect rates of development [9], although the underlying mechanisms are unclear. In humans, the onset of sexual maturation has continued to trend toward younger ages, likely as the result of yet unidentified environmental stimuli [41,42]. Elucidation of the signaling pathways mediating Pdda in C. elegans may provide insights in the control of developmental timing in higher animals, including humans.
Unless indicated otherwise, worms were maintained on Nematode Growth Medium (NGM) 6 cm diameter Petri dish plates with E. coli OP50 (http://www.wormbook.org/methods, Brenner 1974). The following C. elegans strains were used: wild type Bristol N2, daf-16(mu86), daf-16(m26), daf-2(e1368), daf-2(e1370), daf-12(rh61 rh411), ins-11(tm1053), daf-22(ok693), daf-22(m130), dhs-28(hj8), dhs-28(2581), daf-9(rh50), daf-9(dh6); daf-12(rh61 rh411), sir-2.1(ok434), odr-3(n2150), tax-4(ks28), osm-6 (p811), egl-4(n477), nhr-49(nr2041), hsf-1(sy441), nhr-8(ok186), nhr-8(tm1800), nhr-8(ok186);daf-12 (rh61 rh411), mec-4(e1399).
Wildtype (N2) or daf-22(ok693) mutants worms were synchronized by letting 50–100 young adults lay eggs for 1 h on each of 20–30 6 cm NGM plates with E. coli OP50 bacteria. This resulted in plates with 100–150 wildtype or daf-22 eggs per plate. Worms were allowed to grow until reaching larval stage 4, before egg laying started. At this time all worms were carefully picked off the plates. These wildtype- or daf-22-preincubated plates were then used in subsequent Pdda assays with wildtype (using wildtype-preincubated plates), daf-22 mutant (using daf-22-preincubated plates) or nhr-8;daf-12 double mutant (using daf-22-preincubated plates) worms. For these assays, wildtype, daf-22(ok693) mutant, or nhr-8(ok186); daf-12(rh61 rh411) double mutant worms were synchronized by egg laying of 50–100 young adults for 1 hour on three 6 cm-NGM plates. Freshly laid single eggs were either transferred to pre-incubated 6 cm NGM plates ("HD-preincubated ISO", 20–30 plates per assay) or to untreated OP50 coated NGM plates ("ISO", 20–30 plates per assay). Simultaneously, cohorts of 50–60 freshly laid eggs were transferred to OP50 coated 6 cm NGM plates ("HD", 3 plates per assay). Single worms from the ISO, HD-preincubated ISO, and HD plates were isolated to fresh NGM plates (20–30 plates per condition) at 59 h. Starting at 60 h, worms were scored every hour for laying of the 1st egg.
50–100 hermaphrodite N2 worms from synchronized plates were placed onto NGM plates with fresh E. coli OP50 bacteria and allowed to lay eggs for 1 h. Protocol I. From plates with freshly synchronized eggs, 100 eggs were transferred to fresh plates seeded with E. coli OP50 bacteria. At young adult stage, 62 h after birth, worms were transferred to fresh plates in groups of 1, 10, 20, and 50 worms per plate. Protocol II. From plates with freshly synchronized eggs, 1, 10, 20, and 50 eggs were transferred to fresh plates seeded with E. coli OP50 bacteria. For each assay, 20–30 plates with 1 worm, 5–10 plates with 10 worms, and 3–5 plates with 20 or 50 worms were set up. For ageing studies with dafachronic acids, worms were set up on dafa#3 containing plates prepared as described above according to protocol II. Worms were transferred daily onto fresh plates until cessation of egg laying (~ day 8) and every 3 days after that until the experiment was completed. Animals were scored as dead if they failed to respond to a tip on the head and tail with a platinum wire. Worms with internal hatching, exploders or animals that crawled off the plate were excluded. Aging experiments were conducted in two different laboratories, in Kiel, Germany, and Ithaca, NY, as indicated in the Supporting Tables.
The Pdda metric was calculated from time differences of mean times to first egg laying ("t"), as follows: Pdda [%] = [t(ISOmutant)-t(HDmutant)]/[t(ISOwildtype)-t(HDwildtype)]*t(ISOwildtype)/t(ISOmutant)*100. Whereas absolute developmental times exhibit a high degree of variability, acceleration of development under HD conditions as measured by this Pdda metric is much less variable (see Supporting Tables and S10 Fig). The standard deviation of the difference in times between onset of egg laying under isolated versus high density conditions was calculated: STDtime difference = sqrt(STDiso^2/Niso+STDHD^2/NHD); P-values were calculated using t-tests, and the Bonferroni correction was used to assess significance for multiple comparisons. Box plots were generated with the “R-Studio” software. Lifespan data were analysed using the SPSS software for determination of mean and median lifespan, generation of life span curves and log-rank tests.
Supplementary Information is linked to the online version of the paper at http://journals.plos.org/plosgenetics
|
10.1371/journal.pmed.1002700 | Reducing chronic disease through changes in food aid: A microsimulation of nutrition and cardiometabolic disease among Palestinian refugees in the Middle East | Type 2 diabetes mellitus and cardiovascular disease and have become leading causes of morbidity and mortality among Palestinian refugees in the Middle East, many of whom live in long-term settlements and receive grain-based food aid. The objective of this study was to estimate changes in type 2 diabetes and cardiovascular disease morbidity and mortality attributable to a transition from traditional food aid to either (i) a debit card restricted to food purchases, (ii) cash, or (iii) an alternative food parcel with less grain and more fruits and vegetables, each valued at $30/person/month.
An individual-level microsimulation was created to estimate relationships between food aid delivery method, food consumption, type 2 diabetes, and cardiovascular disease morbidity and mortality using demographic data from the United Nations (UN; 2017) on 5,340,443 registered Palestinian refugees in Syria, Jordan, Lebanon, Gaza, and the West Bank, food consumption data (2011–2017) from households receiving traditional food parcel delivery of food aid (n = 1,507 households) and electronic debit card delivery of food aid (n = 1,047 households), and health data from a random 10% sample of refugees receiving medical care through the UN (2012–2015; n = 516,386). Outcome metrics included incidence per 1,000 person-years of hypertension, type 2 diabetes, atherosclerotic cardiovascular disease events, microvascular events (end-stage renal disease, diabetic neuropathy, and proliferative diabetic retinopathy), and all-cause mortality. The model estimated changes in total calories, sodium and potassium intake, fatty acid intake, and overall dietary quality (Mediterranean Dietary Score [MDS]) as mediators to each outcome metric. We did not observe that a change from food parcel to electronic debit card delivery of food aid or to cash aid led to a meaningful change in consumption, biomarkers, or disease outcomes. By contrast, a shift to an alternative food parcel with less grain and more fruits and vegetables was estimated to produce a 0.08 per 1,000 person-years decrease in the incidence of hypertension (95% confidence interval [CI] 0.05–0.11), 0.18 per 1,000 person-years decrease in the incidence of type 2 diabetes (95% CI 0.14–0.22), 0.18 per 1,000 person-years decrease in the incidence of atherosclerotic cardiovascular disease events (95% CI 0.17–0.19), and 0.02 decrease per 1,000 person-years all-cause mortality (95% CI 0.01 decrease to 0.04 increase) among those receiving aid. The benefits of this shift, however, could be neutralized by a small (2%) increase in compensatory (out-of-pocket) increases in consumption of refined grains, fats and oils, or confectionaries. A larger alternative parcel requiring an increase in total food aid expenditure by 27% would be more likely to have a clinically meaningful improvement on type 2 diabetes and cardiovascular disease incidence.
Contrary to the supposition in the literature, our findings do not robustly support the theory that transitioning from traditional food aid to either debit card or cash delivery alone would necessarily reduce chronic disease outcomes. Rather, an alternative food parcel would be more effective, even after matching current budget ceilings. But compensatory increases in consumption of less healthy foods may neutralize the improvements from an alternative food parcel unless total aid funding were increased substantially. Our analysis is limited by uncertainty in estimates of modeling long-term outcomes from shorter-term trials, focusing on diabetes and cardiovascular outcomes for which validated equations are available instead of all nutrition-associated health outcomes, and using data from food frequency questionnaires in the absence of 24-hour dietary recall data.
| Type 2 diabetes mellitus and cardiovascular disease have become leading causes of morbidity and mortality among Palestinian refugees.
Traditional food aid is often primarily composed of grain, flour, and rice, which may increase the risk of such chronic diseases.
It remains unclear how much changes in food aid in delivery method and composition may alter chronic disease risk.
We constructed a microsimulation model to estimate the impact of food aid delivery methods on food consumption, type 2 diabetes, and cardiovascular disease outcomes among the 20- to 79-year old subpopulation receiving food aid (43%) among 5.3 million registered Palestinian refugees in Syria, Jordan, Lebanon, Gaza, and the West Bank.
The simulation took into account the proportion who received food aid and the dietary patterns among them, including the portion of food acquired and consumed from other sources besides food aid.
Using the simulation, we did not observe that a change from food parcel to electronic debit card delivery of food aid or to cash aid led to a meaningful change in consumption, biomarkers, or disease outcomes. We did observe that a shift an alternative food parcel with less grain and more fruits and vegetables may avert hypertension and type 2 diabetes but be neutralized through minimal compensatory (out-of-pocket) increases in consumption of refined grains, fats and oils, or confectionaries.
A larger alternative parcel requiring an increase in total food aid expenditure by 27% would be more likely to have a clinically meaningful improvement on type 2 diabetes and cardiovascular disease.
An alternative food parcel would be more effective than transitioning from traditional food aid to either debit card or cash delivery from the perspective of reducing chronic disease.
But compensatory increases in consumption of less healthy foods may neutralize the improvements from an alternative food parcel unless total aid funding were increased substantially.
| Increasingly, refugee camps worldwide have become semipermanent or permanent, accompanied by an epidemiologic transition with fewer traumatic injuries, infectious diseases, or malnutrition as well as a much higher rate of chronic disease [1,2]. Type 2 diabetes mellitus and cardiovascular disease have become leading causes of morbidity and mortality among refugees in the Middle East [3–9], and are particularly prevalent among the >5 million registered Palestine refugees who live in Syria, Jordan, Lebanon, Gaza, and the West Bank, with a prevalence of 12.1% for type 2 diabetes and 18.6% for hypertension among adults over 40 years old [10]. Treatment for these chronic diseases now consumes a substantial part of the healthcare system budgets of the United Nations (UN) agency responsible for support of this population [1].
Chronic diseases affecting long-term refugee settlements are commonly nutrition-related conditions such as diabetes and cardiovascular disease; in addition, >40% of refugees (and, in some “fields” [refugee settlement areas designated by the UN], >80%) receive food aid. This aid is primarily composed of grain, flour, and rice, under a traditional model that emphasizes maintenance of calorie intake rather than dietary diversity [11]. The reason for this composition of aid is 2-fold: first, because aid was traditionally designed for emergency situations to address acute under-nutrition needs and second, because of supply-side pressures to dump excess United States agricultural production on international markets through food aid [12]. Reducing chronic disease among refugees through modification of food aid packages for primary prevention has become a subject of intensive discussion and preliminary policy changes, particularly because healthcare systems are so under-resourced in the region that funding for secondary prevention is severely constrained [13,14]. At least three reforms have been proposed and piloted to different degrees in field-based trials: (i) partial replacement of traditional food parcel delivery (“in-kind” food aid) with electronic debit cards (“e-vouchers”) restricted to purchasing non-tobacco/alcohol foods [14–16], (ii) partial replacement of traditional food parcel delivery with unrestricted cash [14,17], and/or (iii) an alternative food parcel that replaces a portion of grain with fruits and vegetables (canned, dried, or—if locally available—fresh) [18–20].
How much reduction in chronic disease, if any, would be expected from the proposed food aid policy changes remains unclear because field-based trials or pilot demonstration projects cannot be sustained for sufficient periods to track long-term outcomes. Increased dietary diversity, food security, sense of agency, and local economic growth have been observed from field-based randomized trials and pilot demonstration projects of alternative food aid. Diversion towards non-food products, theft, corruption, or food price inflation have been minimal or undetected in these studies, even under the cash delivery policy [14–16,21]. Hence, agencies such as the World Food Program have begun to expand debit card or cash programs, despite opposition from the US government and associated private contractors invested in traditional food parcel delivery [22–24]. The World Food Program, World Health Organization, and others working with large refugee populations in the Middle East have called for simulation models to help estimate the longer-term chronic disease consequences of alternative food aid [14,20].
The principal aim of this study was to identify the potential changes in type 2 diabetes and cardiovascular disease among Palestinian refugees transitioning from traditional food aid to an alternative aid (debit card, cash, or alternative food parcel). Our primary a priori hypothesis was that transitioning from traditional food parcel delivery to a debit card for food would reduce the overall incidence and complications from type 2 diabetes and cardiovascular disease, given prior evidence that the debit card intervention was particularly effective at increasing measures of dietary diversity [15,16]. The secondary hypotheses were that the transition to debt card would be more effective than substitution to cash, but less effective than transition to an alternative food parcel.
Approval for this study was obtained from the Institutional Review Board of Stanford University (eProtocol number IRB-39274).
Study design and reporting was based on the Modeling Good Research Practice Guidelines by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) [25]. S1 Text details the data underlying the results and provides the prospective analysis plan.
An individual-level microsimulation was created to estimate the impact of food aid delivery methods on food consumption, type 2 diabetes, and cardiovascular disease outcomes among the 20- to 79-year-old subpopulation receiving food aid (43%) among 5.3 million registered Palestinian refugees in Syria, Jordan, Lebanon, Gaza, and the West Bank. The simulation took into account the proportion who received food aid and the dietary patterns among them, including the portion of food acquired and consumed from other sources besides food aid. Individual people were simulated by sampling from the large datasets specified below to capture the correlations among demographic, food consumption, and cardiometabolic variables, including age, sex, nutrition profile, diabetes and cardiovascular disease biomarkers, chronic disease history, and medication use.
The microsimulation (Fig 1) had the following 3 key components: (i) a demographic component specifying features of the refugee population as of 2017, including age, sex, and location; (ii) a food consumption component reflecting individual food consumption in major categories based on the person’s demographics, incorporating both aid-based and non–aid-based food, and conditional on whether the person received food parcel delivery or electronic debit card delivery of food aid; and (iii) a health component, reflecting both demographics and food consumption as well as detailing diabetes and cardiovascular disease history, biomarkers, medications, and the estimated probability of subsequent outcome events, listed below.
Demographic data were obtained from the UN (2017) on 5,340,443 registered refugees in Syria, Jordan, Lebanon, Gaza, and the West Bank, including age, sex, and location (the 5 locations above; Table 1) [26].
Food consumption data were obtained from the World Food Program (2011–2017) on registered refugees receiving food aid by the demographic variables listed above, categorized by quantity and caloric value of major food group procured and consumed per person per day, including aid and nonaid food (Table 1) [15,16]. Food procurement and consumption were separately tabulated for persons receiving traditional food parcel delivery of food aid (n = 1,507 households) and for persons obtaining electronic debit card delivery of food aid (n = 1,047 households), as detailed further below.
Health data were obtained from a random 10% sample of people in the electronic health record system of the UN Relief and Works Agency (UNRWA, the UN body responsible for care of registered refugees in Syria, Jordan, Lebanon, Gaza, and the West Bank, who are treated through 143 primary health facilities), stratified by age, sex, principal diagnosis, and location (2012–2015; n = 516,386; Table 1). The health data included the most recent value of age, sex, height, weight, systolic and diastolic blood pressure, hemoglobin A1c (percent), total and high-density lipoprotein (HDL) cholesterol (mmol/L), tobacco smoking status, serum creatinine, urine microalbumin-to-creatinine ratio, cardiovascular and metabolic medications (anticoagulants, antihypertensives, lipid therapies, and glycemic treatment), and diabetes and cardiovascular history (history of myocardial infarction or stroke).
Outcome metrics included key outcomes considered strongly attributable to changes in nutrition per a recent systematic review and meta-analysis [27]: incidence per 1,000 person-years of hypertension (systolic blood pressure ≥130 mmHg or diastolic ≥80 mmHg or prescription of an antihypertensive medication), type 2 diabetes (hemoglobin A1c ≥6.5%, fasting plasma glucose ≥126 mg/dL, random plasma glucose ≥200 mg/dL, or prescription of a diabetes medication), atherosclerotic cardiovascular disease events (first nonfatal or fatal myocardial infarction or stroke), each of three microvascular complications (end-stage renal disease, diabetic neuropathy manifest as pressure sensation loss by 10 g monofilament test, and proliferative diabetic retinopathy leading to severe vision loss with Snellen test acuity <20/200), and all-cause mortality.
Two exposures were compared in the base case analysis: food parcel delivery of food aid and electronic debit card delivery of food aid. Food parcel delivery involved standardized monthly food parcels for each household that could be obtained by the designated recipient for each household at central locations in the community by participants. Each parcel contained wheat flour, white rice, sugar, vegetable oil, chickpeas, lentils, dried whole milk, and canned sardines. Electronic delivery involved $30 per household member per month in value (in 2017 US dollars; roughly the same value as the food parcel) added via automated bank transfer to an electronic debit card, which could be used with an individual PIN code at participating store locations in the community. Delivery was accompanied by a text message reminder to the assigned card bearer’s phone for each household. Debit cards were electronically restricted to permit food purchases, excluding chocolates, alcohol, cigarettes, soda, or non-food items [28]. Both exposures were directed to refugees with household income below the location-specific definitions of absolute (approximately $3.7/person/day) or abject poverty (approximately $1.5/person/day) by a means test, along with those enrolled in parallel aid programs or meeting other definitions of socioeconomic marginalization (e.g., elderly, widowed, and living alone) [29,30].
The change in food consumption attributable to a change from food parcel to electronic debit card delivery was estimated from a series of field-based studies by the World Food Program, comparing households after piloted roll-out of electronic debit card food aid to similar households remaining on food parcel delivery within each location (n = 2,554 households) [15,16]. Consumption in kilocalories (kcal) per person per day was obtained from quantitative household food frequency questionnaires among registered refugees, specifying daily per person consumption in each of 250 food categories (S1 Table). Major changes in consumption in each of the major Food and Agriculture Organization food groups (cereals; tubers, pulses, legumes, and nuts; fruits and vegetables; animal products; additional oils and fats (apart from animal products); and sugars [31]) are summarized in Table 1.
The estimated change in food consumption attributable to a change from food parcel to electronic debit card delivery was used to estimate changes in each outcome metric. Specifically, 4 intermediate variables were calculated to translate change in food consumption to change in each outcome metric: change in total calories, change in sodium and potassium intake, change in fatty acids (polyunsaturated, monounsaturated, and saturated fats), and change in overall dietary quality (Mediterranean Dietary Score [MDS] [32]). The MDS was chosen because it is predictive of incident type 2 diabetes independently from total calorie intake [33], and dietary interventions corresponding to MDS improvements have resulted in reduced type 2 diabetes incidence in prior trials among population in the Middle East [34,35]. Change in total calories was calculated by summing change in each food category from the field-based studies cited above. Change in sodium, potassium, and fatty acids were estimated based on the per-unit content estimates of each food from standardized international databases [36,37], with bootstrapping across the distribution of content variations within each food category to construct 95% confidence intervals (CIs) (S2 Table). Change in MDS was estimated by calculating the score from the food patterns reported in quantitative food frequency questionnaires (S1 Text).
Change in total calories was used to estimate changes in body mass index (BMI), using the National Institutes of Health/Hall model of body weight change, which conditions body weight change estimates on age, sex, height, and starting weight, based on results of randomized controlled trials (S1 Text) [38–40]. Changes in sodium and potassium were used to estimate changes in systolic blood pressure using regression models from a large, international prospective epidemiological study (S1 Text) [41]. Changes in fatty acids were used to estimate changes in total and HDL cholesterol using regression models from a meta-analysis of randomized controlled trials (S1 Text) [42]. Changes in BMI and MDS were used to estimate change in hemoglobin A1c based on regression models from meta-analyses of prospective studies and risk projections (S1 Text) [43,44].
The health implications of the change from food parcel to electronic debit card delivery of food aid were estimated among a simulated refugee population of 5.3 million people. To produce the simulated population, Monte Carlo sampling with copulas was used [45], which is a strategy to repeatedly sample with replacement from the distribution of each variable in the demographic and health datasets: baseline age, sex, height, weight, systolic and diastolic blood pressure, hemoglobin A1c (percent), total and HDL cholesterol (mmol/L), tobacco smoking status, serum creatinine, urine microalbumin-to-creatinine ratio, cardiovascular and metabolic medications, and diabetes and cardiovascular history as of 2015. When sampling, the approach of Monte Carlo sampling with copulas captures both the marginal distribution of each factor as well as the correlations among the factors (S1 Fig).
Following current modeling guidelines [46], the full life course of the population was simulated for any persons alive during a 10-year policy planning horizon (2016–2025). The outcome metrics among the population were simulated with traditional food aid delivery (baseline simulation), and then after transition to electronic debit card delivery of food aid (intervention simulation). In each simulation, UN birth rate estimates were used to simulate new births, the age- and sex-specific linear secular trend in each biomarker was used to simulated biomarker changes over time in each population, and both cardiovascular and all-cause mortality equations calibrated to the death statistics at each location were used to estimate deaths conditional on demographics and biomarkers [47]. Incidence of type 2 diabetes and hypertension were defined each year of the simulation based on blood pressure and hemoglobin A1c, respectively [48,49]. The incidence of cardiovascular disease events—first nonfatal or fatal myocardial infarction, first nonfatal or fatal stroke—was estimated from the Globorisk equations for each location, which used age, sex, smoking status, diabetes status, systolic blood pressure, and total cholesterol to estimate risk [50,51]. The incidence of each of three microvascular complications—end-stage renal disease, diabetic neuropathy, and diabetic retinopathy—and all-cause mortality were estimated by recalibrating the baseline hazard rate in the Risk Equations for Complications of Type 2 Diabetes (RECODe) to the observed rate of each complication from the most recent UNRWA health audits (Table 1) [52,53]. To account for uncertainty in the effect sizes of diet on health outcomes, we bootstrapped across the distribution of 95% CIs around each coefficient in the input equations to construct 95% CIs around each outcome metric. We note these estimation approaches attempt to focus on previously validated physiological models of causality from specific nutritional components to the outcomes of interest, at the risk of underestimating indirect correlational effects for which causal pathways are not yet clearly established.
As an alternative to the debit card program, we simulated 2 other potential interventions. First, we simulated transitioning from debit card delivery of food aid to transitioning to cash aid, which has been introduced at several sites. The change in food consumption attributable to a change from food parcel to cash aid was estimated by food frequency questionnaires in a field-based study (n = 1,041 households receiving a debit card compared to 1,507 receiving cash aid) and included analysis of diversion of cash for non-food uses [14].
Second, we simulated a proposed alternative food parcel that would reduce the amount of grain within the food parcel but increase the amount of fruits and vegetables (canned, dried, and—where possible—fresh) included by way of both regional and local sourcing [20,24]. We specifically performed a threshold analysis to estimate how much decrease in grain and increase in fruits and vegetables would be needed to reduce the rate of the chronic disease outcomes in our model without increasing the total aid budget, given the exchange rate in budgetary dollars of bundled grain costs versus bundled fruit and vegetable costs. The latter costs incorporated the additional procurement, labor, transport, and storage costs of increased fruit and vegetable inclusion; correction for seasonal price variations; and (for noncanned/nondried fresh fruits and vegetables) additional refrigerated transport and storage and increased delivery frequency to avoid spoilage (Table 1) [54,55]. We restricted the threshold analysis to ensure that the lower 95% CI bounds of total available food per person per day from the alternative food parcel would always remain above the World Health Organization/Food and Agricultural Organization goals to avoid food insecurity (>2,100 kcal/person/day, >10% from protein, >17% from fat [11]).
Following recently updated cost-effectiveness guidelines [46], the incremental cost-effectiveness of transitioning away from traditional food parcel delivery of food aid to alternative strategies (debit card, cash, or alternative food parcel) was calculated from both a healthcare system and a societal perspective (S4 Table). Costs from each of the healthcare and societal perspectives are itemized in Table 2 and were computed from the same life-course perspective over a 10-year policy planning horizon, as with the impact simulation above. Costs for both delivery mechanisms included formal and informal healthcare sector expenditures (reflecting observed trends in service provision, pharmaceutical dispensation, and utilization at UNRWA facilities), and nonhealthcare expenditures (including overhead expenditure and costs of in-kind food or debit card expenditures). Disability-adjusted life-years (DALYs; the sum of years of life lost and utility-weighted years of life lived with disability) were calculated based on utility weights for each outcome metric from a prior international survey (Table 2) [58]. Both costs and DALYs were discounted at a 3% annual rate, and 2 incremental cost-effectiveness ratios (one for the healthcare sector perspective, the other for the societal perspective) were expressed in 2017 US dollars per DALY.
Uncertainty analysis was performed by Monte Carlo sampling with replacement 10,000 times from the probability distribution of all input parameters listed in the Tables, from which 95% CIs were calculated [59]. Analyses were weighted by frequency weights estimated to convert the healthcare input data to a representative Palestinian population sample (S5 Table). Simulations were performed in R (version 3.3.3; The R Foundation for Statistical Computing, Vienna, Austria) with the MASS [60] and tidyverse [61] packages, using the code uploaded at https://sdr.stanford.edu concurrent with publication.
Demographic data from the UN indicated that, among the 5,340,443 registered refugees in Gaza, West Bank, Syria, Lebanon, and Jordan, the mean age was 32.5 years old (interquartile range [IQR] 15.6–43.2 years), with 60.1% female and 25.3% located in Gaza, 15.1% in the West Bank, 10.2% in Syria, 8.7% in Lebanon, and 40.7% in Jordan; 43.0% were enrolled in food aid programs (Table 1).
Food consumption data from the World Food Program among the population of refugees receiving traditional food parcel delivery of food aid averaged 2,296 kcal/person/day (IQR 845–2,362), including 1,239 kcal/person/day from cereals; 137 kcal/person/day from tubers, pulses, legumes, and nuts; 230 kcal/person/day from fruits and vegetables; 343 kcal/person/day from animal products; 163 kcal/person/day from additional/added oils and fats; and 184 kcal/person/day from sugars. Food consumption among the population of refugees receiving electronic debit card delivery of food aid averaged 2,441 kcal/person/day (IQR 1,258–3,623), including 1,209 kcal/person/day from cereals; 100 kcal/person/day from tubers, pulses, legumes, and nuts; 271 kcal/person/day from fruits and vegetables; 526 kcal/person/day from animal products; 157 kcal/person/day from additional oils and fats; and 178 kcal/person/day from sugars (Table 1).
Health data from UNRWA indicated a mean BMI of 29.2 kg/m2 (IQR 23.5–34.1), a mean systolic blood pressure of 124.6 mmHg (IQR 111.7–136.5), a mean diastolic blood pressure of 76.5 mmHg (IQR 66.7–85.3), a mean total cholesterol of 4.9 mmol/L (IQR 4.6–5.2), a mean HDL cholesterol of 1.1 mg/dL (IQR 0.8–1.4), and a current prevalence of 12.2% for type 2 diabetes mellitus, with people with type 2 diabetes mellitus averaging a hemoglobin A1c of 8.2% (IQR 6.7–9.6) (Table 1). We note that the UNRWA data cover all locations with Palestinian refugees (not only the West Bank and Gaza but also the fields in Syria, Jordan, and Lebanon).
We did not find that the change from food parcel to electronic debit card delivery of food aid led to a meaningful change in consumption, biomarkers, or disease outcomes in our simulation model (Table 3). Among those receiving food aid in our model, the change from food parcel to electronic debit card delivery of food aid was estimated to produce an overall 145 kcal/person/day increase in calorie consumption (95% CI 647 decrease to 929 increase, from the baseline of 2,296), 641 mg/person/day increase in sodium consumption (95% CI 2,321 decrease to 2,793 increase, from a baseline of 4,288), 258 mg/person/day increase in potassium consumption (95% CI 2,277 decrease to 3,102 increase, from a baseline of 3,834), 2,014 mg/person/day increase in saturated fatty acid consumption (95% CI 7,507 decrease to 11,429 increase, from a baseline of 26,124), 2,689 mg/person/day increase in monounsaturated fatty acid consumption (95% CI 10,039 decrease to 15,293 increase, from a baseline of 28,567), 938 mg/person/day increase in polyunsaturated fatty acid consumption (95% CI 5,860 decrease to 7,657 increase, from a baseline of 26,871), and a 0.1 point increase in the MDS on a scale from 0 to 14, such that higher values indicate better adherence to the Mediterranean diet (95% CI −1.0 to 1.0, from a baseline of 7.9).
The changes in food consumption among those receiving food aid, attributable to the change from food parcel to electronic debit card delivery of food aid, were estimated by our model to produce a 2.5 kg/m2 increase in BMI (95% CI 11.4 decrease to 17.0 increase, from a baseline of 29.2), 0.58 mmHg increase in systolic blood pressure (95% CI 5.61 decrease to 6.81 increase, from a baseline of 124.6), 0.29 mmHg increase in diastolic blood pressure (95% CI 2.14 decrease to 2.71 increase, from a baseline of 76.5), 0.06 mmol/L increase in total cholesterol (95% CI 0.17 decrease to 0.29 increase, from a baseline of 4.9), and a 0.11 mmol/L increase in HDL cholesterol (95% CI 0.55 decrease to 0.75 increase, from a baseline of 1.1; Table 4).
The changes in biomarkers, attributable to the change from food parcel to electronic debit card delivery of food aid, were estimated by our model to produce a 0.14 per 1,000 person-years increase in the incidence of hypertension (95% CI 3.03 decrease to 2.61 increase, from a baseline of 35.06), 0.77 per 1,000 person-years increase in the incidence of type 2 diabetes (95% CI 3.14 decrease to 6.98 increase, from a baseline of 12.24), 0.40 per 1,000 person-years increase in the incidence of atherosclerotic cardiovascular disease events (95% CI 1.03 decrease to 1.14 increase, from a baseline of 6.58), 0.11 per 1,000 person-years incidence in the incidence of end-stage renal disease (95% CI 0.34 decrease to 0.72 increase, from a baseline of 1.56), 0.24 per 1,000 person-years incidence in the incidence of diabetic neuropathy (95% CI 1.17 decrease to 2.10 increase, from a baseline of 4.41), 0.36 per 1,000 person-years incidence in the incidence of proliferative diabetic retinopathy (95% CI 1.52 decrease to 2.66 increase, from a baseline of 4.53), and 0.16 increase per 1,000 person-years all-cause mortality (95% CI 0.51 decrease to 1.11 increase, from a baseline of 3.50) among those receiving food aid (Table 5).
Among those receiving food aid in our model, a change from food parcel to cash aid was estimated to produce an overall 238 kcal/person/day increase in calorie consumption (95% CI 603 decrease to 1,077 increase, from the baseline of 2,296), 865 mg/person/day increase in sodium consumption (95% CI 2,236 decrease to 3,861 increase, from a baseline of 4,288), 517 mg/person/day increase in potassium consumption (95% CI 2,194 decrease to 3,179 increase, from a baseline of 3,834), 2,773 mg/person/day increase in saturated fatty acid consumption (95% CI 7,407 decrease to 12,840 increase, from a baseline of 26,124), 3,779 mg/person/day increase in monounsaturated fatty acid consumption (95% CI 9,837 decrease to 17,268 increase, from a baseline of 28,567), 1,588 mg/person/day increase in polyunsaturated fatty acid consumption (95% CI 5,623 decrease to 8,728 increase, from a baseline of 26,871), and a 0.1 point increase in the MDS on a scale from 0 to 14 such that higher values indicate better adherence to the Mediterranean diet (95% CI −1.0 to 1.0, from a baseline of 7.9).
The changes in food consumption among those receiving food aid, attributable to the change from food parcel to cash aid, were estimated by our model to produce a 4.2 kg/m2 increase in BMI (95% CI 10.6 decrease to 19.8 increase, from a baseline of 29.2), 0.35 mmHg increase in systolic blood pressure (95% CI 6.15 decrease to 6.89 increase, from a baseline of 124.6), 0.23 mmHg increase in diastolic blood pressure (95% CI 2.33 decrease to 2.77 increase, from a baseline of 76.5), 0.07 mmol/L increase in total cholesterol (95% CI 0.17 decrease to 0.32 increase, from a baseline of 4.9), and a 0.17 mmol/L increase in HDL protein cholesterol (95% CI 0.53 decrease to 0.85 increase, from a baseline of 1.1).
The changes in biomarkers, attributable to the change from food parcel to cash aid, were estimated by our model to produce a 0.04 per 1,000 person-years increase in the incidence of hypertension (95% CI 3.33 decrease to 2.60 increase, from a baseline of 35.06), 1.28 per 1,000 person-years increase in the incidence of type 2 diabetes (95% CI 2.95 decrease to 8.05 increase, from a baseline of 12.24), 0.52 per 1,000 person-years increase in the incidence of atherosclerotic cardiovascular disease events (95% CI 0.94 decrease to 1.31 increase, from a baseline of 6.58), 0.22 per 1,000 person-years increase in the incidence of end-stage renal disease (95% CI 0.33 decrease to 0.84 increase, from a baseline of 1.56), 0.43 per 1,000 person-years increase in the incidence of diabetic neuropathy (95% CI 1.09 decrease to 2.42 increase, from a baseline of 4.41), 0.60 per 1,000 person-years increase in the incidence of proliferative diabetic retinopathy (95% CI 1.45 decrease to 3.05 increase, from a baseline of 4.53), and 0.26 increase per 1,000 person-years all-cause mortality (95% CI 0.45 decrease to 1.31 increase, from a baseline of 3.50) among those receiving aid (Table 5).
We observed that a food aid parcel would need to divert 8.5% of the current grain parcel to fruits and vegetables to reduce all chronic disease outcome measures while maintaining the same total aid budget, after accounting for uncertainty in fruit and vegetable procurement and delivery costs. In particular, such a shift would be expected to reduce supply of grain in food aid parcels by 40 kcal/person/day and increase fruit and vegetable supply by 7 kcal/person/day in food aid parcels. If the supply were fully consumed and did not result in any shift in non–aid-based food consumption, the change from the traditional to alternative food parcel would be expected to produce an overall 33 kcal/person/day decrease in calorie consumption (95% CI 27–38, from the baseline of 2,296), 64 mg/person/day decrease in sodium consumption (95% CI 60–68, from a baseline of 4,288), 1.3 mg/person/day decrease in potassium consumption (95% CI 18 decrease to 15 increase, from a baseline of 3,834), 345 mg/person/day increase in saturated fatty acid consumption (95% CI 340–350, from a baseline of 26,124), 358 mg/person/day increase in monounsaturated fatty acid consumption (95% CI 349–367, from a baseline of 28,567), 342 mg/person/day increase in polyunsaturated fatty acid consumption (95% CI 336–347, from a baseline of 26,871), and a 0.2 point increase in the MDS on a scale from 0 to 14 such that higher values indicate better adherence to the Mediterranean diet (95% CI 0.0–2.0, from a baseline of 7.9).
The changes in food consumption among those receiving food aid, attributable to the change from traditional food parcel to alternative parcel, were estimated by our model to produce a 0.57 kg/m2 decrease in BMI (95% CI 0.37–0.84, from a baseline of 29.2), 0.12 mmHg decrease in systolic blood pressure (95% CI 0.08–0.16, from a baseline of 124.6), 0.05 mmHg decrease in diastolic blood pressure (95% CI 0.03–0.07 from a baseline of 76.5), 0.004 mmol/L decrease in total cholesterol (95% CI 0.014 decrease to 0.006 increase, from a baseline of 4.9), and a 0.03 mmol/L decrease in HDL cholesterol (95% CI 0.02–0.04, from a baseline of 1.1).
The changes in biomarkers, attributable to the change from traditional food parcel to alternative parcel, were estimated by our model to produce a 0.08 per 1,000 person-years decrease in the incidence of hypertension (95% CI 0.05–0.11, from a baseline of 35.06), 0.18 per 1,000 person-years decrease in the incidence of type 2 diabetes (95% CI 0.14–0.22 from a baseline of 12.24), 0.18 per 1,000 person-years decrease in the incidence of atherosclerotic cardiovascular disease events (95% CI 0.17–0.19, from a baseline of 6.58), 0.14 per 1,000 person-years decrease in the incidence of end-stage renal disease (95% CI 0.04–0.24, from a baseline of 1.56), 0.05 per 1,000 person-years decrease in the incidence of diabetic neuropathy (95% CI 0.03–0.08, from a baseline of 4.41), 0.08 per 1,000 person-years decrease in the incidence of proliferative diabetic retinopathy (95% CI 0.05–0.11, from a baseline of 4.53), and 0.02 decrease per 1,000 person-years all-cause mortality (95% CI 0.01 decrease to 0.04 increase, from a baseline of 3.50) among those receiving aid (Table 5).
Because transitioning from food parcel to either electronic debit card delivery or cash aid delivery of food aid was not effective at improving the chronic disease outcomes, we restricted the cost-effectiveness analysis to examining the transition from food parcels to alternative food parcels. The changes in health outcomes—attributable to a 10-year policy change from food parcels to alternative food parcels—were estimated by our model to avert 3,034 discounted DALYs per 100,000 population, over the life courses of the simulated individuals both receiving and not receiving food aid (95% CI 2,532–3,617, from a baseline of 1,194,868 DALYs accumulated in the overall population; Table 6). From a healthcare sector perspective that included the 10-year costs of disease management but not the costs of the parcel food aid or debit card expenditures and infrastructure, the change from traditional to alternative food parcels was estimated to save $1,255,370 in discounted healthcare expenditures per 100,000 population (95% CI $1,047,389–$1,496,447, from a baseline of $150,631,285 in expenditures for the overall population; Table 6), including both out-of-pocket and healthcare delivery system costs. From a societal perspective that included the 10-year cost savings of traditional to alternative food parcel procurement and delivery, the cost-effectiveness of alternative food parcel was estimated to save the same amount, $1,255,370, because (by design) the alternative food parcel was designed to have the same overall food aid budget in discounted expenditures per 100,000 population (including overhead), from a baseline of $336,848,228 when including food aid and associated overhead costs (Table 6). The change from traditional to alternative food parcels was estimated to have an incremental cost-effectiveness ratio of $414 saved per DALY averted from either a healthcare sector or a societal perspective (95% CI $290–$591). If the alternative food parcel was increased in funding level by 27% (as in the “enhanced aid budget” scenario described above), the incremental cost-effectiveness ratio increased to $70,223 spent per DALY averted (95% CI $58,223–$121,746).
Contrary to the supposition in the literature, our findings do not robustly support the theory that transitioning from traditional food aid to either debit card or cash delivery alone would necessarily reduce chronic disease outcomes. Rather, an alternative food parcel would be more effective, even after matching current budget ceilings. But compensatory increases in consumption of less healthy foods may neutralize the improvements from an alternative food parcel unless total aid funding were increased substantially.
In this study, we created a simulation model for forecasting changes in type 2 diabetes and cardiovascular disease incidence and complications from a transition from traditional food aid to an alternative aid format (debit card, cash, or alternative food parcel) among Palestinian refugees. One contribution of simulation models is to quantify the degree of certainty or uncertainty in making a policy decision based on available evidence. We incorporated the most comprehensive data available from the comprehensive healthcare provider for Palestinian refugees and from the main food aid providers among Palestine refugees in the Middle East; by repeatedly sampling from these data sources, we rejected our primary a priori hypothesis that transitioning from traditional food parcel delivery to a debit card for food would reduce the overall incidence and complications from type 2 diabetes and cardiovascular disease. Contrary to our hypothesis and to the supposition in the literature [22–24], we observed large variation in the possible trajectories of chronic disease from the changes in diet observed among refugees receiving alternative food aid delivery strategies, spanning the spectrum from both increased to decreased chronic disease morbidity and mortality. This addresses the request from policymakers to understand the degree to which proxy outcomes observed from short-term field-based trials may translate into long-term chronic disease outcomes and costs, particularly accounting for uncertainties around the estimated results [14,20]. Our findings suggest that the dietary measures and observations in extant studies do not robustly support the idea that chronic disease incidence would necessarily be reduced by a transition from traditional food aid to either debit card or cash delivery alone.
We did find, however, that transitioning to an alternative food parcel could reduce chronic disease morbidity and mortality with less uncertainty in the projected outcomes if the alternative food parcel transitioned to slightly less grain and more fruits and vegetables. We found, in particular, that small changes at the individual level in cardiometabolic risk factors would be expected to translate into larger population-level health gains and cost savings given the high prevalence of these diseases. The magnitude of the gains are somewhat higher than that anticipated, for example, from general sodium reduction in the food supply [62]. The alternative food parcel would be expected to be both effective and cost-effective even if the total aid budget—including additional infrastructure costs for fruit and vegetable procurement and delivery—were unchanged. Uncertainty was lower in this simulation due in part to the narrowed set of food content available for the food aid parcel, as opposed to the large range of content available in local stores. Our analyses importantly accounted for the proportion of the overall refugee population who receive food aid and the dietary patterns among them, including the portion of food acquired and consumed from other sources besides food aid. The microsimulation was also strengthened by using individual participant data from large datasets capturing the correlations among demographic, food consumption, and cardiometabolic variables, including age, sex, nutrition profile, diabetes and cardiovascular disease biomarkers, chronic disease history, and medication use.
Our analysis has important caveats and limitations. There is much uncertainty around our estimates due to the task of modeling long-term outcomes from shorter-term trials while accounting for the wide variation in nutrient profiles of available foods and their cardiometabolic consequences. While we focused on the chronic disease implications of transitions in food aid, there are other important reasons why food aid transitions might be important—including improved agency and choice among refugees, stimulus to local economies in semipermanent or permanent settlements, and avoidance of international dumping of undesired or extra crops under the guise of food aid [12]. Importantly, the primary analysis assumed that the nonaid portion of food consumption would not change with the change in aid, such that people would not compensate for the change by buying more grain and less fruits or vegetables with their own limited income; this assumption must be tested in field studies. An additional limitation is that we focused on a narrow subset of nutritional outcomes—type 2 diabetes and cardiovascular disease risk factors and outcomes—for which validated equations are available to translate food intake estimates into outcomes [50, 52–53]. Nutrition plays an important role in additionally reducing cancers and other diseases and improving infant health, for which equations are not yet well established [62,63]. Furthermore, our results are based on food frequency questionnaires, which are known to be less reliable and subject to underreporting compared to 24-hour dietary recalls. Finally, it is important to note that interventions that are effective in providing nutritional support in many locales may not be effective in the Gaza Strip, where blockades, conflict, and funding crises can limit quantities of goods distributed through the UN and humanitarian organizations. Additionally, prior detailed qualitative studies suggest that the supply chains for locally produced food versus internationally supplied food must undergo substantial review and modification to achieve the aim of an alternative food aid intervention [64–67].
The next logical step following this analysis is to perform field-based randomized studies of the alternative food parcel to identify unanticipated barriers to achievement, the role of seasonality and spoilage, and the degree to which refugees do or do not alter their diets in the context of an alternative food parcel. Prior to such randomized studies, our findings do not robustly support the theory that transitioning from traditional food aid to either debit card or cash delivery alone would necessarily reduce chronic disease outcomes. We did find, however, that transitioning to an alternative food parcel could be effective and should be explored further.
|
10.1371/journal.pgen.1003430 | Altered Splicing of the BIN1 Muscle-Specific Exon in Humans and Dogs with Highly Progressive Centronuclear Myopathy | Amphiphysin 2, encoded by BIN1, is a key factor for membrane sensing and remodelling in different cell types. Homozygous BIN1 mutations in ubiquitously expressed exons are associated with autosomal recessive centronuclear myopathy (CNM), a mildly progressive muscle disorder typically showing abnormal nuclear centralization on biopsies. In addition, misregulation of BIN1 splicing partially accounts for the muscle defects in myotonic dystrophy (DM). However, the muscle-specific function of amphiphysin 2 and its pathogenicity in both muscle disorders are not well understood. In this study we identified and characterized the first mutation affecting the splicing of the muscle-specific BIN1 exon 11 in a consanguineous family with rapidly progressive and ultimately fatal centronuclear myopathy. In parallel, we discovered a mutation in the same BIN1 exon 11 acceptor splice site as the genetic cause of the canine Inherited Myopathy of Great Danes (IMGD). Analysis of RNA from patient muscle demonstrated complete skipping of exon 11 and BIN1 constructs without exon 11 were unable to promote membrane tubulation in differentiated myotubes. Comparative immunofluorescence and ultrastructural analyses of patient and canine biopsies revealed common structural defects, emphasizing the importance of amphiphysin 2 in membrane remodelling and maintenance of the skeletal muscle triad. Our data demonstrate that the alteration of the muscle-specific function of amphiphysin 2 is a common pathomechanism for centronuclear myopathy, myotonic dystrophy, and IMGD. The IMGD dog is the first faithful model for human BIN1-related CNM and represents a mammalian model available for preclinical trials of potential therapies.
| The intracellular organization of muscle fibers relies on a complex membrane system important for muscle structural organization, maintenance, contraction, and resistance to stress. Amphiphysin 2, encoded by BIN1, plays a central role in membrane sensing and remodelling and is involved in intracellular membrane trafficking in different cell types. The ubiquitously expressed BIN1, altered in centronuclear myopathy (CNM) and myotonic dystrophy (DM), possesses a muscle-specific exon coding for a phosphoinositide binding domain. We identified splice mutations affecting the muscle-specific BIN1 isoform in humans and dogs presenting a clinically and histopathologically comparable highly progressive centronuclear myopathy. Our functional and ultrastructural data emphasize the importance of amphiphysin 2 in membrane remodeling and suggest that the defective maintenance of the triad structure is a primary cause for the muscle weakness. The canine Inherited Myopathy of Great Danes is the first faithful mammalian model for investigating other potential pathological mechanisms underlying centronuclear myopathy and for testing therapeutic approaches.
| Amphiphysin 2 is one of the key factors in muscular membrane remodeling. The gene, BIN1, has recently been associated with two different muscle disorders: centronuclear myopathy (CNM, MIM #255200) [1] and myotonic dystrophy (DM, MIM #160900 and #602668) [2]. However, the muscle-specific role of the ubiquitous protein amphiphysin 2 and the pathological mechanisms underlying the muscle disorders are not well understood. This is mainly due to the lack of faithful animal models.
Centronuclear myopathies are characterized by a generalized muscle weakness, atrophy, predominance of type I fibers, and aberrant positioning of nuclei and mitochondria [3]. The different genetic forms are not or are only moderately progressive. The X-linked and dominant CNM forms result from mutations in the phosphoinositide phosphatase myotubularin (MTM1) and the large GTPase dynamin 2 (DNM2), respectively [4], [5]. The autosomal recessive form (ARCNM) is caused by mutations in BIN1, probably involving a partial loss-of-function as the protein level was found to be normal in previously described patients [1]. Amphiphysin 2, encoded by BIN1, contains a N-terminal amphipathic helix, a BAR (Bin/Amphiphysin/Rvs) domain, able to sense and maintain membrane curvature, a Myc-binding domain and a SH3 domain, both implicated in protein-protein interactions [6], [7], [8]. There are at least 12 different isoforms, mainly differing by the presence or absence of a phosphoinositide-binding domain and a clathrin-binding domain encoded by exon 11 and exons 13–16, respectively [9], [10]. The clathrin-binding domain is present in the brain isoforms, while the phosphoinositide-binding (PI) domain is found almost exclusively in skeletal muscle isoforms [10], [11], [12]. Sequencing of cDNA demonstrated that all BIN1 skeletal muscle isoforms contain exon 11 [12]. All ARCNM mutations described to date are in ubiquitously expressed exons [1], [13], [14], [15], raising the question about the molecular basis of the muscle-specificity of the disease. The BAR domain mutations strongly decreased the amphiphysin 2 membrane tubulating properties when expressed in cultured cells, while SH3 truncating mutations were shown to impair the binding and recruitment of dynamin 2 [1].
Mis-splicing of the BIN1 muscle-specific exon 11 was reported in different forms of myotonic dystrophy (DM) [2]. DM is one of the most common muscular dystrophies in neonates and adults, and results from the expression of mutant RNAs with expanded CUG or CCUG repeats leading to the sequestration of splicing factors and subsequent defects in RNA splicing. Splicing alterations of the muscle chloride channel CLCN1 are suggested to be responsible for the myotonia, whereas aberrant splicing of the insulin receptor INSR gene is thought to cause insulin resistance in DM patients. Complete or partial skipping of BIN1 exon 11 in congenital and adult DM was shown to involve structural T-tubule abnormalities and subsequently muscle weakness [2]. However, there are numerous splicing defects in DM. It is therefore challenging to assess the exact contribution of BIN1 exon 11 skipping to the DM phenotype, even though severe hypotonia, respiratory failure and histopathological features such as fiber hypotrophy and centrally located nuclei in the congenital forms of DM show intriguing similarities to CNM.
Amphiphysins are key regulators of membrane curvature and trafficking [16]. They can sense membrane curvature and presumably promote the curvature and fission of membranes [17]. Membrane binding occurs via BAR domain dimers, presenting a positively charged concave site that interacts with the negative membrane charges [17]. Amphiphysins also bind and recruit other regulators of endocytosis to sites of plasma membrane inward budding [18]. Amphiphysin 1 expression is restricted to neuronal tissues and the protein regulates synaptic vesicle recycling in the brain [19]. Amphiphysin 2 is highly expressed in adult striated muscle and its expression increases during muscle cell maturation [10], [11], [20], [21]. The polybasic residues encoded by BIN1 exon 11 are required for amphiphysin 2-induced membrane tubulation when exogenously expressed in cultured cells [1], [22]. In skeletal muscle, amphiphysin 2 is localized at the T-tubules, which are deep sarcolemmal invaginations enabling excitation-contraction coupling [11], i.e. the process converting an electrical stimulus into mechanical muscle work. This specific localization, together with the membrane tubulation properties of the muscle-specific isoform containing the PI domain, called iso8 or M-amphiphysin, has led to the suggestion that amphiphysin 2 is implicated in T-tubule biogenesis [22]. This is sustained by defects in the localization of nascent T-tubule markers such as caveolin 3 following BIN1 downregulation in cultured cells [23], and by the abnormal T-tubule structure seen in drosophila with null mutations in amph, the unique ortholog of mammalian amphiphysins 1 and 2 [24]. While faithful animal models were previously characterized for the MTM1 and DNM2 related CNM forms [25], the perinatal lethality of Bin1-null mice precludes the analysis of the role of amphiphysin 2 in skeletal muscle [26]. Therefore, critical questions concerning the muscle-specific function of amphiphysin 2 in mammals and the pathological mechanism of BIN1-related CNM remain unanswered. The lack of a faithful animal model for autosomal recessive centronuclear myopathy is a hurdle for a better comprehension of the pathological mechanisms and for the development of therapeutic approaches.
In this study, we identified and characterized the first human BIN1 mutation affecting the muscle-specific PI domain. We also identified a novel spontaneous canine model reproducing the human pathology and allowing investigations on the physiological role of amphiphysin 2 in skeletal muscle after birth. Characterization of the dog model revealed an important role for amphiphysin 2 in triad structure, and we provide the evidence for a physiological function of the membrane-deforming properties of amphiphysin 2 and its alternative splicing-dependent activity. Our data support the hypothesis that the alteration of the muscle-specific function of amphiphysin 2 on membrane remodeling is a common pathomechanism underlying several canine and human myopathies.
To identify BIN1 mutations affecting its function in skeletal muscle, we sequenced the muscle-specific exon 11 and the adjacent splice-relevant intronic regions in a cohort of 84 patients with various forms of centronuclear myopathy and without mutations in MTM1, DNM2, or in the other BIN1 exons. We identified a homozygous BIN1 exon 11 splice acceptor mutation (IVS10-1G>A) in two affected members from a consanguineous family from Turkey (Figure 1A and 1B). DNA was not available for the third affected member, who is expected to carry the same homozygous BIN1 mutation as her monozygotic twin sister. The parents are healthy and do not present clinical signs of a muscle disorder. They are first-degree cousins and were found to be heterozygous for the BIN1 exon 11 splice acceptor mutation, confirming autosomal recessive inheritance of the disease. The mutation was not found in unaffected individuals from different origins, including 37 DNAs from an ethnically matched control population, and is not listed in the SNP databases as dbSNP, 1000 genomes, or the NHLBI exome variant server.
Patients 1 and 2 are dizygotic twins. Pregnancy and birth, as well as motor and speech development were normal. General muscle weakness was diagnosed at 3.5 years. Hypotonia, muscle weakness (predominantly of the lower limbs), respiratory distress (VC 50%) and loss of motor skills were rapidly progressive and the twins died from acute pneumonia involving cardiac failure at age 5 and 7, respectively. Patient 3 is the younger brother, and as for his sisters, pregnancy, birth, motor and speech development were normal. Age of onset was 3.5 years and the myopathy was highly progressive, contrasting the slow progression of muscle weakness in the reported CNM cases with BIN1 mutations in ubiquitous exons [1], [13], [14], [15]. Patient 3 presented with predominant proximal muscle weakness of the lower limbs requiring a wheelchair since the age of 5 years, facial weakness, but no respiratory distress. Eye movement defects, as seen in the majority of the MTM1, DNM2 and BIN1 patients, were not noted. Deep tendon reflexes were absent and the patient had progressive contractures in knees and ankles. Electrophysiological evaluation was normal or showed only unspecific myopathic changes, with normal nerve conduction velocity. Cardiac defects were not noted and CK levels were normal. Patient 3 is now 9 years old and presented at his last medical exam in April 2012 with low MRC grades for both upper and lower limbs.
The BIN1 IVS10-1G>A variation changes the AG acceptor splice site into AA, and is predicted to impair exon 11 splicing by various algorithms. The wild-type acceptor site was recognized by NNSPLICE (score 0.84) and Human Splice Finder (88.5), while no acceptor splice site was predicted in the mutated sequence. To confirm an impact on exon 11 splicing, we performed RT-PCR after RNA isolation from a muscle biopsy of patient 1, amplified a fragment encompassing exons 10 to 12, and obtained a shorter product compared to the control (Figure 1C). To analyze the transcript(s), we cloned the PCR-products and sequenced the resulting clones. As we and others previously reported, the skeletal muscle BIN1 isoforms contain exon 11, but lack exons 7 and 13 to 16. Exon 17 can be either present or absent, corresponding to isoform 8 or M-amphiphysin [10], [11], [12]. Among the 30 analyzed clones, only a single clone contained exon 11. Twenty-nine clones did not contain exon 11 and directly combined exon 10 with exon 12, demonstrating a major skipping of the in-frame exon 11 in the patient muscle (Figure 1D). The impact of the mutation on the amphiphysin 2 protein level in skeletal muscle was investigated by Western blot (Figure 1E). Using an anti-PI domain antibody, we detected several bands in the control as previously reported [1], most probably reflecting post-translational modifications of the different isoforms containing exon 11. In the patient muscle, we found a significant decrease of the level of the amphiphysin 2 isoform containing the PI domain, confirming exon 11 skipping in most BIN1 muscle transcripts. The amphiphysin levels detected with the anti-SH3 domain antibody were similar in patient 1 and control. Together with the genetic data, we conclude that the rapidly progressive CNM form results from a splice mutation involving the skipping of the muscle-specific exon 11.
Previous publications demonstrated the importance of the amphiphysin 2 PI domain in PtdIns(4,5)P2 binding and membrane tubulation [1], . We transfected C2C12 cells with BIN1 constructs including or excluding exon 11, and we induced the differentiation of the murine myoblasts into myotubes. Myotubes expressing the exon 11 containing isoform showed tubulation [22], [27], whereas the isoform without exon 11 did not induce this effect (Figure 2). Quantification revealed statistical significance. Immunolabelling of actin, caveolin-3 and RYR1 did not reveal obvious differences between the differentially transfected myotubes (data not shown), suggesting that the amphiphysin 2 PI domain is important for late muscle development or maintenance, rather than for early muscle development. This hypothesis is supported by the fact that the patients were unaffected at birth and during early childhood.
The perinatal lethality of Bin1-null mice precludes investigations on the role of amphiphysin 2 in skeletal muscle maintenance [26]. To identify and characterize an animal model for BIN1-related CNM, we analyzed canine pedigrees with molecularly unsolved myopathies. The canine Inherited Myopathy of Great Danes (IMGD) is characterized by rapidly progressive muscle atrophy and exercise intolerance with an age of onset of about 6 months. Histological examinations of muscle biopsies from autosomal recessive cases from the UK, Canada and Australia revealed increased nuclear internalization and centralization [28], [29], [30], consistent with centronuclear myopathy. We excluded mutations in MTM1 [31] and PTPLA [32] before sequencing the coding regions and intron/exon boundaries of the canine BIN1 gene (XM_540990.3). We identified a homozygous AG to GG substitution of the BIN1 exon 11 acceptor splice site in five dogs from Canada, US and UK (IVS10-2A>G; Figures 3A and 3B). CK values for the dogs were normal or slightly elevated. Pedigree reconstruction revealed a distant relationship between the US and one UK dog (Figure 3C) and a previous publication reported a common ancestor for all IMGD dogs in the UK [29]. The BIN1 IVS10-2A>G mutation was not found in 112 healthy Great Danes and in 35 dogs from 12 other breeds, strongly suggesting its pathogenicity.
Like the human BIN1 IVS10-1G>A mutation, the canine BIN1 IVS10-2A>G variation affects the exon 11 acceptor splice site. To assess its impact on splicing, we performed RT-PCR on RNA isolated from skeletal muscle biopsies and found a strong reduction of the BIN1 RNA level compared to healthy controls and compared to a control gene (MTM1, Figure 3D). We however detected a faint signal of expected size and cloned the amplicon. All three clones contained exon 11 with 27 additional upstream nucleotides, encoding the amino acid sequence ASASRPFPQ (Figure 3E). This in-frame extension results from the disposition of a weak cryptic 5′ acceptor site. The intronic sequence upstream of exon 11 slightly differs between human and dog, possibly explaining the cryptic splicing in dogs versus exon skipping in human patients (Figure 3F). To confirm the impact of the splice mutation on the amphiphysin 2 protein level, canine muscle extracts were probed with an anti-PI domain antibody on Western blot. Compared to the healthy control, amphiphysin 2 was significantly reduced in the affected dog (Figure 3G). Using an anti-SH3 antibody we detected a strong reduction of all skeletal muscle amphiphysin isoforms (Figure S1) in accordance with the RT-PCR data. We conclude that the canine Inherited Myopathy of Great Danes results from a BIN1 exon 11 splice mutation, provoking a strong reduction of the exon 11/PI domain-containing RNA and protein.
Vastus lateralis muscle biopsies were performed for patient 1 as well as for patient 3 at the age of 3.5 years. H&E staining revealed prominent nuclear centralization (>60%, arrow), fiber atrophy and endomysial fibrosis (Figure 4), consistent with centronuclear myopathy. Similarly, H&E staining of biceps femoris muscle biopsies from affected dogs revealed nuclear internalization (>40%) and fiber atrophy. The central areas devoid of staining reflect perinuclear regions lacking myofibrils. Of note, the transverse muscle sections of patients and affected dogs showed an unusual lobulated appearance with indentations of the sarcolemma (arrowheads). NADH staining of human and canine sections revealed dense central areas in most fibers and “spoke of wheel” appearance in 5% of the fibers. ATPase staining showed no or only a slight predominance of type I muscle fibers as compared to the age matched controls. Gomori trichrome staining did not reveal any further abnormalities (data not shown). Taken together, human and canine histopathologies were comparable.
To uncover the pathological defects underlying this highly progressive form of centronuclear myopathy and to validate the canine model, we analyzed human and dog muscle biopsies by electron microscopy. Ultrastructural analysis of the human muscle biopsy revealed centralized nuclei surrounded by an area devoid of myofibrils and containing glycogen granules and other organelles (Figure 5A, Figure S2), as commonly seen in MTM1, DNM2 and BIN1-related CNM. Myofibrillar disintegration with occasional Z-band streaming (arrow, Figure 5A) was seen in the adjacent sarcomeres. Triad structures were found to be aberrant and we observed frequent enlarged structures, most probably originating from the sarcoplasmic reticulum (arrow, Figure 5D). We also noted other membrane alterations, including accumulations of membranes and tubules, vacuoles containing whorled membranes (arrow, Figure 5B), as well as a high number of myelin-like membranous structures suggestive of autophagosomes (arrow, Figure 5C). Likewise, ultrastructural analysis of muscle biopsies from an affected Great Dane dog showed nuclear internalization, mitochondrial accumulations around the internalized nuclei and myofibrillar disarray (Figure 5E, Figure S3). We furthermore found membranous whorls (arrow, Figure 5F) as reported for the X-linked CNM Labrador retriever model with MTM1 mutation [31], deep membrane invaginations (arrowhead, Figure 5F), lipofuscin granules (arrow, Figure 5G), and abnormal triads in almost all fibers (arrow, Figure 5H). Sarcolemmal invaginations contained basement membranes and often pointed towards centralized nuclei. Taken together and considering the histological analysis described above, histopathology of IMGD dogs and human patients appear strikingly similar, emphasizing common alterations of membrane structures.
To further characterize the pathophysiology of the rapidly progressive human CNM and canine IMGD, we performed immunolocalization experiments on muscle biopsies. Using the R3062 antibody recognizing most amphiphysin isoforms or the PI-domain specific R2405 antibody, signals were detected as an intracellular network in transverse sections of human and canine controls (Figure 6). Signals were also detected in sections of muscles from patient and affected dog, reflecting the presence of different amphiphysin 2 isoforms as shown by Western blot. Despite the decrease of BIN1 RNA in affected dogs, the remaining mis-spliced in-frame transcripts can explain the detection of amphiphysin 2 on muscle sections, especially because immunohistochemistry is not quantitative. The amphiphysin 2 network appeared however abnormal in patient and IMGD sections. In some fibers we noted central areas without any signal, while in other fibers accumulations around centralized nuclei were observed (arrows). To determine whether these anomalies were specific for the BIN1 exon 11 splice mutation or rather a general BIN1-related CNM feature, we analyzed a muscle biopsy from a patient with the previously reported BIN1 p.Asp151Asn mutation and a classical ARCNM phenotype [1]. We observed similar accumulations of amphiphysin 2 (Figure 6A), suggesting that different BIN1 mutations in humans and dogs lead to similar amphiphysin 2 mis-localization in muscle.
Amphiphysin 2 has been proposed to be implicated in T-tubule biogenesis, but the exact link has barely been documented in mammalian skeletal muscle [22]. We therefore examined the skeletal muscle triad using antibodies against the junctional sarcoplasmic calcium channel RYR1 and the T-tubule marker DHPR in human and dog (Figure 7). Both proteins were profoundly altered, showing focal accumulations or central areas without signal in the fibers. Compared to the control longitudinal sections, the transversal orientation of RYR1-labeled triads was lost in patient and canine muscle. Similarly, the longitudinal sarcoplasmic calcium pump SERCA1 was mislocalized in sections from affected dogs.
We next wanted to know whether the aberrant triad structure was concurrent with more generalized membrane defects. Dysferlin and caveolin 3, key regulators of membrane repair and trafficking [33], [34], were found to be mainly located at the sarcolemma in control muscle sections. In contrast, transverse sections of patient 1 and of an affected Great Dane dog revealed striking intracellular accumulations of dysferlin, mainly around central nuclei (Figure 7). Labeling of the sarcolemmal markers dysferlin, caveolin 3 and dystrophin confirmed the presence of numerous fibers with unusual lobulated and indented sarcolemma, representing deep sarcolemmal invaginations pointing towards the center of the fibers (arrows, Figure 7). Taken together, our data correlate the highly progressive human CNM and canine IMGD with general membrane alterations at the triad, the sarcolemma and within the fibers. However, these defects did not reflect a general disorganization of the sarcomere, as alpha-actinin labeling appeared largely normal (not shown). Staining of developmental myosin revealed no difference between affected and control dogs, indicating that there is no excessive fiber regeneration in IMGD dogs (Figure S3).
As MTM1 is mutated in X-linked human and canine CNM, we investigated the localization of myotubularin in muscle sections of IMGD dogs. Myotubularin formed an intracellular network in control sections and the signal was stronger in type II fibers labeled with the SERCA1 antibody (Figure 8). In both analyzed IMGD muscles, myotubularin was mainly located as concentric strands pointing to the center in both type I and type II fibers. We conclude that altered splicing of BIN1 has a strong impact on myotubularin localization in muscle, revealing a potential link between IMGD and X-linked CNM.
In this study we identified and characterized BIN1 mutations affecting the splicing of the muscle-specific exon 11, resulting in a rapidly progressing myopathy in humans and dogs. The IMGD dog is the first faithful mammalian model for BIN1-related centronuclear myopathy and particularly for the highly progressive form, and is the only characterized mammalian model available for preclinical trials of potential therapies for this severe congenital myopathy. Our data provide strong evidence for muscle-specific functions of amphiphysin 2 in membrane structural organization and remodelling and allow novel insights into the overlapping pathogenesis of centronuclear myopathy and myotonic dystrophy. A schematic representation of the amphiphysin 2 protein domains and of the position of the mutations and splicing alterations causing classical autosomal recessive centronuclear myopathy, rapidly progressive human CNM and canine IMGD as well as myotonic dystrophy is shown in Figure 9.
Classical BIN1-related ARCNM has been described with neonatal or childhood onset, hypotonia and ptosis and all mutations were found in ubiquitously expressed exons [1], [13], [14], [15]. The muscle weakness was mildly to moderately progressive, and some patients could still walk at older age. In contrast, our patients with a splice mutation affecting the muscle-specific exon 11 showed a rapid disease progression involving strong care-dependence and leading to death within a few years, despite normal motor development and disease-onset not before 3.5 years. The histopathological findings of our patients and of the previously reported ARCNM cases partially overlap, including atrophy, prominent nuclear internalization and central dense areas upon NADH-TR staining of muscle sections. However, there is no evidence for type I fiber predominance in the muscle biopsies of our patients. Previous RT-PCR experiments demonstrated a progressive integration of exon 11 in BIN1 mRNA during human skeletal muscle development [2]. We therefore hypothesize that the muscle-specific exon 11 plays a major role in muscle maintenance, rather than in early muscle development. This is in accordance with the highly progressive phenotype of humans and dogs with a disease onset several months or years after birth. Consistently, we detected amphiphysin 2 in muscle tissue, but RNA analysis revealed major skipping of BIN1 exon 11. This suggests that the patients mainly express an embryonic BIN1 isoform, which might not be able to assume the function of the adult BIN1 isoform, possibly explaining the more progressive phenotype compared to patients with BIN1 mutations in the ubiquitously expressed exons.
The characterization of the pathological mechanisms leading to BIN1-related CNM and the development of potential therapeutic approaches is obviated by the lack of a faithful animal model. Bin1-null mice are perinatally lethal [26], so that a comprehensive analysis of skeletal muscle alterations during disease development is not possible. We sought for dog breeds with molecularly unsolved congenital myopathies and we identified the canine Inherited Myopathy of Great Danes as a disease model reproducing the histological and physiological defects observed in BIN1-related CNM patients. IMGD has been reported for cases in Canada, Australia and UK and is characterized by generalized muscle atrophy, exercise intolerance, exercise-induced tremor and muscle wasting [29]. The disease typically starts before 10 months of age, is highly progressive, and most of the affected dogs are euthanized before 18 months of age due to severe debilitating muscle weakness. Histological examinations revealed internalized or central nuclei without evidence of inflammation, disruption of the sarcomeric architecture with central fiber areas devoid of myofibrils, and central accumulations of mitochondria and glycogen granules ([28], [29], [30] and our data). In addition, type I fiber predominance in combination with an increased expression of genes implicated in the slow-oxidative metabolism was described [35]. In this study we demonstrate that IMGD and progressive CNM have a comparable etiopathology and both conditions result from mutations of the AG acceptor splice site of the BIN1 muscle-specific exon 11. The histopathology and the cellular organization defects of the human and canine muscle disorders are almost identical, we therefore consider IMGD as a faithful mammalian model for BIN1-related centronuclear myopathy.
Some dogs of our IMGD cohort were found to be negative for BIN1 mutations, suggesting that IMGD encompasses several disorders with similar clinical and overlapping histopathological features. The proven relationship of two affected Great Dane dogs demonstrates a common origin of the BIN1 exon 11 splice mutation, and it is likely that all five affected dogs described here can be traced back to a common ancestor. As the muscle disorder is inherited as a recessive trait, and as canine pedigrees are generally highly inbred, it is likely that the mutation can be found in Great Dane dog populations from all over the world, as recently demonstrated for another autosomal recessive CNM form in Labrador retrievers [36]. It is therefore of veterinary medical interest to sequence BIN1 exon 11 in Great Dane dogs. Also, veterinarians and veterinary pathologists should consider BIN1 mutations as a possible cause of any unexplained progressive myopathy in dogs, especially when the biopsy displays internal nuclei and lobulated or indented sarcolemma.
Detailed immunohistochemical and ultrastructural analyses of muscles from patients and affected Great Dane dogs revealed common membrane alterations and abnormal accumulations of proteins regulating membrane trafficking. Similar findings were observed on biopsies from patients with DNM2 or MTM1 mutations [12], suggesting that mislocalization of triad proteins reflects common aberrations in CNM and that the amphiphysin 2 muscle-specific isoform plays an important role in triad formation and/or maintenance. This is in accordance with the known biochemical function of amphiphysin 2 and other N-BAR domain proteins to sense membrane curvature and to potentially induce curvature through the insertion of an amphipathic helix into the membrane bilayer. In vitro and cell culture experiments have led to the suggestion that the exon 11 encoded PI-binding motif is essential for membrane tubulation in cultured muscle cells [22]. Indeed, Drosophila mutated for amphiphysin, the ortholog of both amphiphysin 1 and amphiphysin 2, display an abnormal T-tubule system [24]. T-tubule alterations and muscle weakness were reproduced in murine Tibialis anterior injected with a U7 small nuclear RNA construct harboring an antisense sequence promoting BIN1 exon 11 skipping [2]. However, nuclear centralization and atrophy were not observed, contrasting with the IMGD model. This difference might be species-related, is possibly due to a low efficacy of the AAV-U7 method or alternatively to the examination time point 4 months post injection. As the triad is the membrane structure controlling excitation-contraction coupling, this suggests that impaired excitation-contraction coupling and subsequent calcium homeostasis defects are a primary cause of the myopathy. Of note, abnormal intracellular calcium release was observed in isolated murine muscle fibers after BIN1 shRNA-mediated knock-down [37]. Together with the present characterization of the IMGD model, these data indicate that amphiphysin 2 has an important muscle-specific role in triad structural maintenance, and provide additional evidence that triad modifications are a common defect in centronuclear myopathies, IMGD and myotonic dystrophies [2], [12].
Triads are not the only membrane compartment affected in patients and dogs harboring BIN1 exon 11 mutations. We also noted central accumulations of caveolin 3 and dysferlin, two key regulators of membrane trafficking in skeletal muscle, numerous membranous whorls, and a peculiar remodeling of the sarcolemma, manifesting an indented fiber perimeter and invaginations towards the center of the fibers. Caveolin 3 regulates membrane tension at the sarcolemma and dysferlin controls membrane exocytosis in sarcolemmal membrane repair [33], [34]. As both proteins are also present on regenerating T-tubules [38], their mislocalization resulting from a BIN1 mutation would be in accordance with defective T-tubule regeneration. Moreover, data mainly obtained in cultured cells support a key role of amphiphysins in the formation of endocytic vesicles [16], and a study in Caenorhabditis elegans suggested a role of amphiphysin in vesicle recycling [39]. Defects in intracellular signaling resulting from calcium defects and impaired transport of ion channels and growth factor might explain the muscle weakness and atrophy in BIN1-related CNM.
Our findings on the IMGD model uncovered possible links between BIN1-related and other forms of CNM. Altered triads and the presence of membranous whorls were reported for MTM1 dog, mouse and zebrafish models as well as for patients with MTM1 mutations involving protein loss [12], [31], [40], [41], [42]. Abnormal triad markers were also reported for MTM1-related and DNM2-related CNM [12], [43]. Dysferlin localization was not extensively studied in MTM1-CNM but was increased in the cytoplasm of a mouse model and in patients with DNM2-CNM [44]. Moreover, we found myotubularin localization was strongly impaired in IMGD muscles. These findings suggest that myotubularin and amphiphysin 2 are in the same pathway regulating membrane remodeling in skeletal muscle and strengthen the hypothesis of a common pathological mechanism of the X-linked and the autosomal recessive CNM forms.
Alternative splicing of BIN1 exon 11 is mis-regulated in patients with myotonic dystrophy [2]. The parallel inclusion of exon 7 was noted, but its impact has not been assessed yet. Here we report the first mutation affecting the muscle-specific exon 11 of BIN1 and having an impact on splicing. The major clinical and histological aspects of the patients and IMGD dogs include general muscle weakness, atrophy and nuclear centralization, consistent with the muscle phenotype in DM patients. Our data therefore support the hypothesis that mis-splicing of BIN1 exon 11 partially accounts for the muscle-specific signs in myotonic dystrophy.
Human sample collection was performed with informed consent from the patients according to the declaration of Helsinki and experimentation was performed as part of routine diagnosis. All dogs were examined with the consent of their owners. Blood and biopsies were obtained as part of routine clinical procedures for diagnostic purposes. Cheek cells were collected by owners or veterinarians using non-invasive swabs. As the data were from client-owned dogs undergoing normal veterinary exams, there was no “animal experiment” according to the legal definitions in Europe and the US. All local regulations related to clinical procedures were observed. Cryopreserved muscle specimens were processed and stored at the University of California, San Diego, under a tissue transfer approval from the institutional Animal Care and Use Committee.
Human Genomic DNA was prepared from peripheral blood by routine procedures and sequenced for all coding exons and intron/exon boundaries of MTM1, DNM2, and BIN1 as described elsewhere [1], [4], [5]. Patient 1 had a normal CTG repeat length at the DMPK locus (7 and 13 repeats) and was therefore excluded for myotonic dystrophy. Control DNAs were from healthy individuals of Turkish origin.
Dog DNA samples were extracted from cheek cells, venous blood or muscle biopsy specimens (cryosections or paraffin embedded tissue) by routine procedures and sequenced for all coding exons and intron/exon boundaries of canine MTM1 [31], PTPLA [32] and BIN1 (primer sequences in Table S1). Control samples were from a world-wide collection of healthy Great Danes as well as from healthy individuals of 13 other breeds.
RNA was extracted from muscle biopsies by routine procedures and reverse transcribed using the SuperScript III kit (Invitrogen, Carlsbad, USA). Human and dog amplicons were cloned into the pGEM-T Easy vector (Promega, Madison, USA) and transfected into E.coli DH5α cells. Blue/white selection, repeated twice, resulted in 30 clones for the human cDNA and 3 clones for the canine cDNA. Control dog was an unaffected Drahthaar (German Wirehaired Pointer). Primer sequences are listed in Table S1.
Western blot and immunofluorescence were performed using routine protocols. Biceps femoris and tibialis anterior biopsies from two affected dogs (14 months and 22 months, respectively) and from healthy age-matched Golden Retrievers or Belgian Shepherds as controls have been used for the analysis. Following antibodies were used for the study: R2406 (home-made rabbit anti-BIN1 PI binding domain), R2444 (home-made rabbit anti-BIN1 SH3 domain), R3062 (home-made rabbit anti-BIN1 exon 12 epitope), R2867 and R2868 (home-made rabbit anti-MTM1), mouse anti-GAPDH (Merck Millipore, Darmstadt, Germany), mouse anti-ryanodine receptor 1 (Affinity BioReagents, Golden, USA), mouse anti-SERCA 1 (Affinity BioReagents, Golden, USA), rabbit anti-dysferlin (Euromedex, Souffelweyersheim, France), goat anti-caveolin-3 (Tebu-BIO, Le-Perray-en-Yvelines, France), rabbit anti-caveolin-3 (Affinity BioReagents, Golden, USA), mouse anti-DHPR (Affinity BioReagents, Golden, USA), and mouse anti-dystrophin (Leica Microsystems, Germany). For immunohistofluorescence, transverse cryosections were prepared, fixed and stained by routine methods. Nuclei were stained with Hoechst or DAPI (Sigma-Aldrich, St. Louis, USA). Sections were mounted with slowfade antifade reagent (Invitrogen, Carlsbad, USA) and viewed using a laser scanning confocal microscope (TCS SP2; Leica Microsystems, Wetzlar, Germany) or a a Zeiss Axio Observer Z.1 microscope equipped with a 20×, 40× or 63× lens and Axioplan imaging with structured illumination (Carl Zeiss, Jena, Germany).
For histochemical analyses, transverse sections of muscle cryosections (8 µm) of vastus lateralis and biceps femoris muscle biopsies were stained with hematoxylin-eosin, modified Gomori trichrome, NADH-TR and myofibrillar ATPase and then assessed for centralized nuclei, fiber morphology, fiber type distribution, cores, protein accumulation and cellular infiltrations.
Muscle biopsies were processed for electron microscopy as described previously [45]. Briefly, the tissue was fixed either in 6% phosphate-buffered glutaraldehyde (human patient) or in 2.5% paraformaldehyde, 2.5% glutaraldehyde, and 50 mM CaCl2 in 0.1 M cacodylate buffer at pH 7.4 (dog), and post-fixed with 2% OsO4, 0.8% K3Fe(CN)6 in 0.1 M cacodylate buffer (pH 7.4) for 2 h at 4°C and incubated with 5% uranyl acetate for 2 h at 4°C. Samples were dehydrated in graded series of ethanol and embedded in epoxy resin 812. Ultrathin sections (70 nm) were contrasted with uranyl acetate and lead citrate.
Murine C2C12 myoblasts were seeded on coverslips and transfected at 50–60% confluency using Lipofectamine 2000 (Invitrogen, Carlsbad, USA) either with GFP-BIN1 isoform 8 (including exon 11) or isoform 9 (excluding exon 11, both were a kind gift from Pietro de Camilli, Howard Hughes Medical Institute, USA). Cells were differentiated after 24 h by changing to medium containing 2% horse serum instead of FCS and fixed and stained after 5 days of differentiation by routine methods. Nuclei were stained with Hoechst/DAPI (Sigma-Aldrich, St. Louis, USA) and sections were mounted with slowfade antifade reagent and viewed using a laser scanning confocal microscope (TCS SP2; Leica Microsystems, Wetzlar, Germany).
1000 genomes - A Deep Catalog of Human Genetic Variation (URL: http://www.1000genomes.org/)
Database of Single Nucleotide Polymorphisms (dbSNP). Bethesda (MD): National Center for Biotechnology Information, National Library of Medicine. (dbSNP Build ID: 134).
(URL: http://www.ncbi.nlm.nih.gov/SNP/)
Exome Variant Server, NHLBI Exome Sequencing Project (ESP), Seattle, WA (URL: http://evs.gs.washington.edu/EVS/)
Online Mendelian Inheritance in Man (OMIM) (URL: http://www.omim.org/)
NNsplice: prediction of splice mutations (URL: http://www.fruitfly.org/seq_tools/splice.html)
Human Splicing finder (URL: http://www.umd.be/HSF/)
|
10.1371/journal.pcbi.1006967 | Recentrifuge: Robust comparative analysis and contamination removal for metagenomics | Metagenomic sequencing is becoming widespread in biomedical and environmental research, and the pace is increasing even more thanks to nanopore sequencing. With a rising number of samples and data per sample, the challenge of efficiently comparing results within a specimen and between specimens arises. Reagents, laboratory, and host related contaminants complicate such analysis. Contamination is particularly critical in low microbial biomass body sites and environments, where it can comprise most of a sample if not all. Recentrifuge implements a robust method for the removal of negative-control and crossover taxa from the rest of samples. With Recentrifuge, researchers can analyze results from taxonomic classifiers using interactive charts with emphasis on the confidence level of the classifications. In addition to contamination-subtracted samples, Recentrifuge provides shared and exclusive taxa per sample, thus enabling robust contamination removal and comparative analysis in environmental and clinical metagenomics. Regarding the first area, Recentrifuge’s novel approach has already demonstrated its benefits showing that microbiomes of Arctic and Antarctic solar panels display similar taxonomic profiles. In the clinical field, to confirm Recentrifuge’s ability to analyze complex metagenomes, we challenged it with data coming from a metagenomic investigation of RNA in plasma that suffered from critical contamination to the point of preventing any positive conclusion. Recentrifuge provided results that yielded new biological insight into the problem, supporting the growing evidence of a blood microbiota even in healthy individuals, mostly translocated from the gut, the oral cavity, and the genitourinary tract. We also developed a synthetic dataset carefully designed to rate the robust contamination removal algorithm, which demonstrated a significant improvement in specificity while retaining a high sensitivity even in the presence of cross-contaminants. Recentrifuge’s official website is www.recentrifuge.org. The data and source code are anonymously and freely available on GitHub and PyPI. The computing code is licensed under the AGPLv3. The Recentrifuge Wiki is the most extensive and continually-updated source of documentation for Recentrifuge, covering installation, use cases, testing, and other useful topics.
| Whether in a clinical or environmental sample, metagenomics can reveal what microorganisms exist and what they do. It is indeed a powerful tool for the study of microbial communities which requires equally powerful methods of analysis. Current challenges in the analysis of metagenomic data include the comparative study of samples, the degree of uncertainty in the results, and the removal of contamination. The scarcer the microbes are in an environment, the more essential it is to have solutions to these issues. Examples of sites with few microbes are not only habitats with low levels of nutrients, but also many body tissues and fluids. Recentrifuge’s novel approach combines statistical, mathematical and computational methods to tackle those challenges with efficiency and robustness: it seamlessly removes diverse contamination, provides a confidence level for every result, and unveils the generalities and specificities in the metagenomic samples.
| Studies of microbial communities by metagenomics are becoming more popular in different biological arenas, like environmental, clinical, food and forensic studies [1–3]. New DNA and RNA sequencing technologies are boosting these works by dramatically decreasing the cost per sequenced base. Scientists can now analyze sets of sequences belonging to microbial communities from different sources and times to unravel longitudinal (spatial or temporal) patterns in the microbiota (see S1 Fig for an example model). In shotgun metagenomic sequencing (SMS) studies, researchers extract and purify nucleic acids from each sample, sequence them, and analyze the sequences through a bioinformatics pipeline (see S2 and S3 Figs for detailed examples). With the development of nanopore sequencing, portable and affordable real-time SMS is a reality [4].
In the case of low microbial biomass samples, there is very little native DNA from microbes; the library preparation and sequencing methods will return sequences whose principal source is contamination [5, 6]. Sequencing of RNA requiring additional steps introduces still further biases and artifacts [7], which in case of low microbial biomass studies translates into a severe problem of contamination and spurious taxa detection [8]. The clinical metagenomics community is stressing the importance of negative controls in metagenomics workflows and, recently, raised a fundamental concern about how to subtract the contaminants from the results [9].
From the data science perspective, this is just another instance of the importance of keeping a good signal-to-noise ratio [10]. When the signal (inherent DNA/RNA, target of the sampling) approaches the order of magnitude of the noise (acquired DNA/RNA from contamination and artifacts), particular methods are required to tell them apart.
The roots of contaminating sequences are diverse, as they can be traced back to nucleic acid extraction kits (the kitome) [11, 12], reagents and diluents [13, 14], the host [15], and the post-sampling environment [16], where contamination arises from different origins such as airborne particles, crossovers between current samples or DNA remains from past sequencing runs [17]. Variable amounts of DNA from these sources are sequenced simultaneously with native microbial DNA, which could lead to severe bias in magnitudes like abundance and coverage, particularly in low microbial biomass situations [18]. If multiplex sequencing uses simple-indexing, false assignments could be easily beyond acceptable rates [19]. Even the metagenomic reference databases have a non-negligible amount of cross-contamination [15, 17, 20].
Regarding the kitome, it varies even within different lots of the same products. For example, the DNeasy PowerSoil Kit (formerly known as PowerSoil DNA Isolation Kit), a product that usually provides significant amounts of DNA and has been widely used, including Earth Microbiome Project and Human Microbiome Project, often yields a background contamination by no means negligible [6]. The lower the biomass in the samples, the more essential it is to collect negative control samples to help in the contamination background assessment because, without them, it would be almost impossible to distinguish inherent microbiota in a specimen —signal— from contamination —noise—.
Assuming that the native and contaminating DNA are accurately separated, the problem of performing a reliable comparison between samples remains. In general, the taxonomic classification engine assigns the reads from a sequencing run to different taxonomic ranks, especially if the method uses a more conservative approach like the lowest common ancestor (LCA) [21]. While LCA drastically reduces the risk of false positives, it usually spreads the taxonomic level of the classifications from the more specific to the more general. Even if the taxonomic classifier does not use the LCA strategy, each read is usually assigned a particular score or confidence level, which should be taken into account by any downstream application as a reliability estimator of the classification.
On top of these difficulties, it is still more challenging to compare samples with very different DNA yields, for instance, low microbial biomass samples versus high biomass ones, because of the different resolution in the taxonomic levels. This sort of problem also arises when the samples, even with DNA yields in the same order of magnitude, have an entirely different microbial structure so that the minority and majority microbes are fundamentally different between them [18]. Finally, a closely related problem emerges in metagenomic bioforensic studies and environmental surveillance, where it is essential to have a method prepared to detect the slightest presence of a particular taxon [3, 22, 23] and provide quantitative results with both precision and accuracy.
From the beginning, the application of SMS to environmental samples supplied biologists with an insight of microbial communities not obtainable from the sequencing of Bacterial Artificial Chromosome (BAC) clones or 16S rRNA [24, 25]. The scientific community soon underlined the need and challenges of comparative metagenomics [26, 27]. MEGAN [28], one of the first metagenomic data analysis tools, provided in its initial release a very basic comparison of samples, which has improved with an interactive approach in more recent versions [29]. In general, metagenomic classification and assembly software is more intra- than inter-sample oriented [30]. Several tools have tried to fill this gap, starting with CoMet [31], a web-based tool for comparative functional profiling that combines different methods such as multi-dimensional scaling and hierarchical clustering analysis to predict functional differences in a collection of metagenomic samples. Soon after, a different approach appeared with the discovery of the crAssphage thanks to the crAss software [32], which provides reference-independent comparative metagenomics using cross-assembly. The following year, Community-analyzer was released, a tool for visually comparing microbial community structure across microbiomes using correlation-based graphs to infer differences in the samples and predict microbial interactions [33]. In 2014, yet another alternative came, COMMET [34], a piece of software that goes a step further by enabling the combination of numerous metagenomic datasets through a scalable method based on efficient indexing. Two years later, a parallel computation method called Simka was published [35], which performs de novo comparative metagenomics by counting k-mers concurrently in multiple datasets.
In 2015, a highly publicized report on the metagenomics of the New York subway suggested that the plague and anthrax pathogens were part of the normal subway microbiome. Soon afterward, several critics arose [36] and, later, reanalysis of the New York subway data with appropriate methods did not detect the pathogens [37]. As a consequence of this and other similar problems involving metagenomic studies, a work directed by Rob Knight [38] emphasized the importance of validation in metagenomic results and issued a tool based on BLAST (Platypus Conquistador). This software confirms the presence or absence of a taxon of interest within SMS datasets by relying on two reference sequence databases: one for inclusions, with the sequences of interest, and the other for exclusions, with any known sequence background. Another BLAST-based method for validating the assignments made by less precise sequence classification programs has been recently announced [22].
The approach of Recentrifuge to increased confidence in the results of taxonomic classification engines follows a dual strategy. Firstly, it accounts for the score level of the classifications in every single step. Secondly, it uses a robust contamination removal algorithm that detects and selectively eliminates various types of contaminants, including crossovers. Recentrifuge directly supports the following high-performance taxonomic classifiers: Centrifuge [7], LMAT [21], CLARK [39], CLARK-S [40], and Kraken [41]. Other classification software is supported through a generic parser. The interactive interface of Recentrifuge enables researchers to analyze the results of those taxonomic classifiers using scored Krona-like charts [42]. In addition to the plots for the raw samples, Recentrifuge generates four different sets of scored charts for each taxonomic level of interest: control-subtracted samples, shared taxa (with and without subtracting the controls), and exclusive taxa per sample. This battery of analysis and plots permits robust comparative analysis of multiple samples in metagenomic studies, especially useful in case of low microbial biomass environments or body sites.
Recentrifuge enables robust contamination removal and score-oriented comparative analysis of multiple samples, especially in low microbial biomass metagenomic studies, where contamination removal is a must.
Just as it is essential to accompany any physical measurement by a statement of the associated uncertainty, it is desirable to attend any read classification with a confidence estimation of the assigned taxon. Recentrifuge reads the score given by a taxonomic classification software to the reads and uses this valuable information to calculate an average confidence level for each taxon in the taxonomic tree associated with the sample analyzed. This value may also be a function of further parameters, such as read quality or length, which is especially useful in case of significant variations in the length of the reads, like in the datasets generated by nanopore sequencers.
Only a few codes, such as Krona [42] and MetaTreeMap [43], are hitherto able to handle a score assigned to the classification nodes. In Recentrifuge, the calculated score propagates to all the downstream analysis and comparisons, including the interface, an interactive framework for a straightforward assessment of the validity of the taxonomic assignments. That is an essential advantage of Recentrifuge over other metagenomic dataset analysis tools.
For each sample, according to the NCBI Taxonomy [44], Recentrifuge populates a logical taxonomic tree, with the leaves usually belonging to the lower taxonomic levels like species, variety or form. The methods involving trees were implemented as recursive functions for compactness and robustness, making the code less error-prone. One of such methods is essential for understanding the way Recentrifuge prepares samples before any comparison or operation such as control subtraction. It recursively ‘folds the tree’ for any sample if the number of assigned reads to a taxon is under the mintaxa setting (minimum reads assigned to a taxon to exist in its own), or because the taxonomic level of interest is over the assigned taxid (taxonomic identifier). See Fig 1A for a working example of the method in action for two samples. The same procedure applies to the trees of every sample in the dataset. This method does not just ‘prune the tree’, on the contrary, it accumulates the counts ni of a taxon in the parent ones np and recalculates the parent score σp as a weighted average taking into account the counts and score of both. In general, the new score of parent taxa, σ p ′ is calculated as follows:
σ p ′ = 1 n p + ∑ i D n i ( σ p n p + ∑ i D σ i n i ) ∀ ( σ i , n i )
where 0 < ni < mintaxa and D is the number of descendant taxa that are to be accumulated in the parent one and σi their respective scores. This is done recursively until the desired conditions are met. This method is applied, at a given taxonomic level, to the trees of every sample before being compared in search for the shared and exclusive taxa. For a sample, the mintaxa parameter defaults to the nearest integer of the decimal logarithm of the number of reads passing the minimum score threshold (minscore) filter, thus growing with the order of magnitude of the effective size of the sample. However, the user can modify such automatic value for mintaxa and set it independently for control and real samples.
In addition to the input samples, Recentrifuge includes some sets of derived samples in its output. After parallel calculations for each taxonomic level of interest, it adds hierarchical pie plots for CTRL (control subtracted), but also for EXCLUSIVE, SHARED and SHARED_CONTROL samples, defined below.
Let T mean the set of taxids in the NCBI Taxonomy and Ts the collection of taxids present in a sample s. If Rs stands for the set of reads of a sample s and Cs for the group of them classifiable, then the taxonomic classification c is a function from Cs to T, i.e., C s → c T, where Cs ⊆ Rs and c [ C s ] = T s ⊆ T. The set L of the 32 − 1 different taxonomic levels used in the NCBI Taxonomy (see S5 Fig) [44] is ordered in accordance with the taxonomy, so (L, <) is a strictly ordered set, since form < variety < subspecies < ⋯ < domain. Then, T s = T s form ∪ ⋯ ∪ T s domain = ∪ L T s l, where T s l represents the collection of taxa belonging to a sample s for a particular taxonomic rank or level l. Related with this, we can write as T s → l the taxa of the sample s for a taxonomic level l once we have applied the ‘tree folding’ to such level l detailed in the previous subsection (and in Fig 1A).
For a taxonomic rank k of interest, in a series of S samples where there are N < S negative controls, Recentrifuge computes the sets of taxa in the derived samples CTRL ( CTRL T s k ), EXCLUSIVE ( EXCL T s k ), SHARED (SHARED Tk) and SHARED_CONTROL (SHARED_CTRL Tk) as:
CTRL T s k = T s → k \ ∪ n N T n → k EXCL T s k = T s → k \ ∪ m ≠ s S T m → k SHARED T k = ∩ m S T m → k SHARED _ CTRL T k = ∩ m > N S T m → k \ ∪ n N T n → k
Please see Fig 1B for examples. Finally, Recentrifuge generates in parallel a set of SUMMARY samples condensing the results for all the taxonomic levels of interest.
For a taxonomic rank k, after the ‘tree folding’ procedure detailed above, the contamination removal algorithm retrieves the set of candidates T ¯ s → k to contaminant taxa from the N < S control samples. Depending on the relative frequency (fi = ni/∑i ni) of these taxa in the control samples and if they are also present in other specimens, the algorithm classifies them in contamination level groups: critical, severe, mild, and other. Except for the latter group, the contaminants are removed from non-control samples. Then, Recentrifuge checks any taxon in the ‘other contaminants’ group for crossover contamination so that it eliminates any taxon marked as a crossover from every sample except the one or ones selected as the source of the pollution. In detail, the algorithm removes any taxon t s k ∈ T ¯ s → k from a non-control sample unless it passes the robust crossover check: a statistical test screening for overall outliers and an order of magnitude test against the control samples. See Fig 2 for an example of this procedure.
The robust crossover tests are defined as follows:
Outliers statistic test ( t s k ) : f t s k > median { f t 1 k , … , f t S k } + δ Q n Order of magnitude test ( t s k ) : f t s k > 10 ξ max { f t 1 k , f t 2 k , … , f t N k }
where Qn [45] is a scale estimator to be discussed below, and δ and ξ are constant parameters of the robust contamination removal algorithm. The parameter δ is an outliers cutoff factor, while ξ is setting the difference in order of magnitude between the relative frequency of the candidate to crossover contaminator in the sample s and the greatest of such values among the control samples. In Recentrifuge, δ typically ranges from 3 to 5, and ξ from 2 to 3.
Qn is the chosen scale estimator for screening the data for outliers because of its remarkably general robustness and other advantages compared to other estimators [45, 46], like the MAD (median absolute deviation) or the k-step M-estimators. It has a 50% breakpoint point, a smooth influence function, very high asymptotic efficiency at Gaussian distributions and is suitable for asymmetric distributions, which is our case, all at a reasonable computational complexity, as low as O(n) for space and O(n log n) for time. So, here:
Q n = d { | f t i k - f t j k | i < j ≤ S } ( m ) : m = ( S 2 + 1 2 ) = Γ ( S 2 + 2 ) 2 Γ ( S 2 ) = S 4 ( S 2 + 1 )
where d = 3.4760 is a constant selected for asymmetric non-gaussian models similar to the negative exponential distribution, m refers to the mth order statistics of the pairwise distances and Γ is the Gamma function.
Recentrifuge is a metagenomics analysis software with two different main parts: the computing kernel, implemented and parallelized from scratch using Python, and the interactive interface, based on interactive hierarchical pie charts by extending the Krona [42] 2.0 JavaScript library developed at the Battelle National Biodefense Institute. Recentrifuge’s novel approach combines robust statistics, arithmetic of scored taxonomic trees, and concurrent computational algorithms to achieve its goals. Fig 3 is a flow diagram of Recentrifuge that clearly shows three parallel regions in the code. In each of them, the work divides into concurrent processes attending to different variables: control and regular samples in the first region, the taxonomic ranks in the second, and the specimen along with the type of analysis in the last parallel region, which summarizes the results.
In any SMS study with related samples, including negative controls, Recentrifuge generates four additional sets of scored charts: the samples with the contamination subtracted, the exclusive taxa per sample, and the shared taxa with and without control taxa subtracted (see S4 Fig). Fig 4 summarizes the package context and data flows. Recentrifuge straightforwardly accepts output files from various taxonomic classifiers, thus enabling a scored-oriented taxonomic visualization for metagenomics. Recentrifuge directly supports output from Centrifuge [7], LMAT [21], CLARK [39], CLARK-S [40], and Kraken [41]. Alternative taxonomic classifiers are supported through a generic interface developed to handle different file formats with comma-separated values (CSV), tab-separated values (TSV), or space-separated values (SSV). The software also includes support for LMAT plasmid assignment system [15]. For implementation details of the Recentrifuge computing kernel please see S1 Appendix, S5 and S6 Figs.
To ensure the broadest portability for the interactive visualization of the results, the central outcome of Recentrifuge is a stand-alone HTML file which can be loaded by any JavaScript-enabled browser. Fig 5 shows a labeled screenshot of the corresponding Recentrifuge web interface for an example of SMS study (see S1 Fig). A vectorial screenshot in SVG format with the original font scheme is available for any sample using the “Screenshot” button of the user interface. The package also provides comprehensive statistics about the reads and their classification performance. Another Recentrifuge output is a spreadsheet collection detailing all the classification results in a compact way. This format is adequate for further data mining on the data, for example, as input for applications such as longitudinal (time or space) series analyzers like Dynomics (in development). Besides, the user can choose between different score visualization algorithms, some of which are more interesting for datasets containing variable length reads, for example, the ones generated by Oxford Nanopore sequencers.
Finally, some filters are available, like the minimum score threshold (minscore), which can be set independently for the control and real samples. The minscore filter can be used to generate different output sets from a single run of the classifier with a low minimum hit length (MHL) setting, saving computational resources. Other filters are mintaxa, described in the scored taxonomic trees subsection, and the lists of identifiers to exclude or include a taxon and all its children in the taxonomic tree.
The additional tools in the Recentrifuge package (see Fig 4) can generate further products and results. Rextract is a script which helps in extracting a subset of classified reads of interest from the single or paired-ends FASTQ input files. This set of reads can be used in any downstream application, such as genome visualization and assembling. Remock is a script for easily creating mock Centrifuge samples, which is useful not only for testing and validation purposes but also for introducing a list of previously known contaminants to be taken into account by the robust contamination removal algorithm. Retest is the code used for continuous integration (CI) testing and algorithm verification procedures (see Section 2 of S4 Appendix for further details and S10 Fig for its flowchart).
We developed a synthetic dataset carefully designed to challenge the Recentrifuge algorithms (see S13 Fig and Section 2.3 of S4 Appendix for details), thus enabling a quantitative assessment of the capability of the method to cope with different kinds of contaminants. We also devised this mock dataset in order to evaluate the ability of the method to deal with cross-contamination between samples. This feature of Recentrifuge is one of the advantages of this novel approach. In addition, this synthetic dataset serves the purpose of the continuous integration framework of the software, as the results of processing these data are compared with a standard to check the reliability of the method after any change in the source code.
Fig 6 shows a comparison of abundances of taxa included in the synthetic dataset before and after the Recentrifuge robust contamination removal algorithm. The taxa belong to species or below in the NCBI taxonomy. The left column of the figure shows the abundance histogram for seven raw samples: four real samples (smpl1 to smpl4) plus three negative control samples (ctrl1 to ctrl3). Similarly, the right column shows the results after the algorithm intervention for the species taxonomic level, i.e., the corresponding CTRL_species samples (see ‘Derived samples’ subsection in Design and implementation). Native taxa are green-colored, crossover contaminants are colored in purple, and other colors indicate different classes of contaminants. The legend of S13 Fig details the complete color code.
We see in Fig 6 that Recentrifuge cleared the CTRL_species samples of the different contaminants (species and below) found in the negative control samples while retaining the particular native taxa, which accumulated up to the species level (see ‘Scored taxonomic trees’ subsection in Design and implementation for details). Examples of important contaminants removed were human reads and those belonging to Cutibacterium acnes. The algorithm also deleted more subtle contamination, such as the reads assigned to Malassezia globosa. Crossover contamination requires special mention. On the one hand, Methanosarcina mazei was ubiquitous among the samples, but it was only native to smpl1 and a contaminant in the rest. On the other hand, M. barkeri was present in the four real samples despite being only native to smpl3, but it was scarce in the control samples, even missing from ctrl2. Recentrifuge accurately detected which were the source sample of both Methanosarcina species, thus keeping the native reads there and clearing the cross-contamination from the rest of the samples.
Furthermore, we included an additional sample (smplH) in the synthetic dataset containing the 241 species of a high-complexity dataset used as a gold standard for benchmarking metagenomic software [47]. As with the other samples, this specimen combined contaminants as additional taxa. In addition, we spiked the controls with low abundances of native taxa from this and the other real samples in order to simulate statistical noise in negative control samples such as low-frequency misclassifications and sequencing errors. We used the complete synthetic dataset to obtain different ROC (receiver operating characteristic) plots. S11 Fig shows the evolution of the sensitivity and specificity from the raw specimens to the CTRL_species samples. Basically, this ROC presented a transition from a scenario of very low specificity, on account of the contamination misidentified as native taxa, to a situation characterized by very high specificity, thanks to the correct detection of contaminants, including crossovers. For some samples, this came at the expense of a slight loss in the sensitivity. The reason for that small decline in the recall rate was the intentional introduction in the synthetic dataset of the archaea Methanobacterium formicicum with two different strains, one native to the samples (M. formicicum DSM 3637) and another a contaminant (M. formicicum JCM 10132). At the species level, once the cross-contamination situation was ruled out, Recentrifuge followed a conservative strategy and deemed the archaeal species as a contaminant and, therefore, the native strain of M. formicicum became a false negative thus decreasing the sensitivity. For samples smpl1 to smpl4 and smplH, S12 Fig shows the ROC as a function of the mintaxa parameter. Results of Fig 6, S11 and S12 Figs can be easily replicated using retest (see Section 2 of S4 Appendix).
To confirm Recentrifuge’s ability to analyze complex metagenomes and provide new biological insight, we considered an ambitious but severely contaminated SMS study of RNA in plasma from individuals with Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS), alternatively diagnosed chronic Lyme syndrome (ADCLS), and systemic Lupus erythematosus (SLE) [48]. This research suffered from large batch and contamination effects and was unable to find a positive association between the plasma microbial content of sick individuals, thus highlighting the relevance of technical controls in metagenomics. More than 240 giga-base-pairs of raw genomic data distributed in 67 samples with paired-ends sequences were downloaded and analyzed using Bowtie2 [49], Centrifuge [7], and SAMtools [50] (see S2 Appendix for the procedure details). Recentrifuge analyzed the different datasets in this study using stricter parameters than the default ones: sequencing of RNA required extra steps than sequencing of DNA, including the reverse transcription of RNA and further purifications [48], which were additional sources of artifacts and contamination. In this case, an increase in the matching length to 60 was advisable [7], so Recentrifuge filtered the Centrifuge output with minscore raised to an even stricter value of 75 (unless otherwise indicated).
To further illustrate the difficulty of the dataset of the SMS study of plasma in ME/CFS patients regarding the contamination, just a couple of results. First, affecting the sequencing batch one, Recentrifuge detected crossover contamination in the negative control samples with the source in the positive control, consisting of human metapneumovirus (hMPV). Second, Recentrifuge reported quite more different taxa in the negative controls than in the normal samples: 65% and 22% more on average, respectively, for the batch two and three. The presence of generalized crossover contamination complicates the removal of the contaminants in the samples by merely excluding the taxa present in the controls. Here it is when the robust contamination removal algorithm of Recentrifuge is of great help: it detects the crossover contaminants (hMPV and other taxa) and removes them from all the samples except for the inferred source. Therefore, the positive control is still positive for hMPV after the contamination removal, as expected (see S7 Fig).
The Recentrifuge analysis of the entire collection of 67 samples revealed the presence of ubiquitous contaminants able to spread over different sequencing batches and type of samples (see S8 Fig). Most of the contaminating bacteria are known contaminants belonging to the kitome [11]. Other pervasive pollutants belong to the fungi orders Eurotiales, Helotiales, Hypocreales, Pleosporales, and Saccharomycetales. The contamination by Apicomplexa, in general, and Plasmodium vivax and Besnoitia, in particular, can be linked to database contamination [15, 20] and seems a negative hallmark of SMS RNA studies related to body fluids [8]. An interesting complementary analysis consisted of retrieving those taxa that are contaminating the negative control samples exclusively. S9 Fig shows the genera contaminating all the control samples but no other specimen along the second batch, representing contaminants which entered the workflow in some procedure or material exclusive to the control samples.
In concordance with the main conclusion of the study of plasma in individuals with ME/CFS [48], Recentrifuge did not find shared taxa after control removal (CONTROL_SHARED empty) when analyzing the samples rearranged in different batches and pathology/healthy groups. Nevertheless, the individual analysis of the samples after contamination removal presents interesting features in a case-per-case review. That is the case of sample 56 in Fig 7, which belongs to an ADCLS patient. It shows a collection of taxa with a high average score (114) in the classification, implying that a majority of sequences mapped in both reads from the pair, except for the contaminant genus Besnoitia, the lowest-scored one. This set of microbes seems compatible with bacteria translocated from the buccopharyngeal cavity into blood, apparently because of an oral chronic inflammatory polymicrobial disease. However, the clinically relevant taxa in this study go far beyond those of sample 56 shown in Fig 7. S3 Appendix portrays other representative bacteria, viruses, and fungi, present in the samples.
Research in recent years is overturning the commonly accepted paradigm which stated that, in healthy individuals, the tissues and body fluids not in contact with the environment are sterile. Healthy organs once thought to be free of microbes are crawling with bacteria, archaea, viruses, and eukaryotes. The shift of paradigm has spread to more and more tissues and fluids, like the deepest layers of the skin [51], the placenta [52], the urine [53], the blood [54, 55], the breast milk, or others [54, 56, 57].
The plasma is the part of the blood with the lower proportion of bacterial DNA, only 0.03% [55]. In the reanalyzed study of plasma in individuals with ME/CFS [48], the intrinsic difficulties of ultra-low microbial biomass joined the handicap of an RNA sequencing technique prone to further artifacts and biases, which resulted in severe widespread contamination. With the results of the research, the classical paradigm might seem supported, that is, the idea of the absence of a plasma microbiota in healthy individuals. However, the authors of the study believed that the limitation of the current techniques prevented them from revealing the microbial component in human plasma. Indeed, with the noise in the same order of magnitude of the signal, a robust method for contamination removal was required to tackle this complex dataset. Despite all the difficulties, the analysis with Recentrifuge has unveiled a meaningful plasma microbiota in the samples (Fig 7 and S3 Appendix). The results are in line with the recent research in the field, which points out the gut, the oral cavity, and the genitourinary tract as the primary sources of the blood microbiome [55, 58, 59].
In conclusion, thanks to the robust contamination removal and the score-oriented comparative analysis of multiple samples in metagenomics, Recentrifuge can play a key role, firstly, in the study of oligotrophic microbes in environmental samples, as it did by showing that microbiomes of Arctic and Antartic solar panels display similar taxonomic profiles [60]; secondly, in the more reliable detection of minority organisms in clinical or forensic samples. The relevant organisms found with a high score in the SMS study of plasma in ME/CFS patients [48] after the robust contamination removal are good examples. Finally, the mock dataset confirmed the worthiness of the developed methods, which demonstrated a radical improvement in specificity while retaining high sensitivity rates even in the presence of cross-contaminants.
Recentrifuge’s main website is www.recentrifuge.org. The data and source code are anonymously and freely available on GitHub at https://github.com/khyox/recentrifuge and PyPI at https://pypi.org/project/recentrifuge. The Recentrifuge computing code is licensed under the GNU Affero General Public License Version 3 (www.gnu.org/licenses/agpl.html). Recentrifuge’s continuous integration (CI) information is public on Travis CI at https://travis-ci.org/khyox/recentrifuge.
The wiki (https://github.com/khyox/recentrifuge/wiki) is the most extensive and updated source of documentation for Recentrifuge, including installation, testing, quick-start, and comprehensive use cases for the different taxonomic classification engines supported. In addition, Recentrifuge’s installation is explained in Section 1 of S4 Appendix, testing is detailed in Section 2 of S4 Appendix, and running Recentrifuge for Centrifuge, LMAT, CLARK flavors, Kraken, and other taxonomic classifiers are subsections of Section 3 of S4 Appendix. Similarly, Sections 4 and 5 of S4 Appendix describe running Rextract and the Recentrifuge command line, respectively. Finally, Section 6 of S4 Appendix includes troubleshooting subsections.
The full Centrifuge output and the detailed Recentrifuge results for the SMS study of plasma in individuals with ME/CFS are publicly available at som1.uv.es/plasmaCFS.
Just as the biochemical profile and cell count are currently usual blood tests, metagenomic analysis of the blood will probably become a standard in a few years. The methods that will pave the way for a well-established clinical practice of metagenomics are still to come. As an open-source project, the participation of the computational biology and clinical metagenomics community will determine the future of Recentrifuge considerably.
An important extension to Recentrifuge is under active development and will be released soon. It is “Regentrifuge”, the counterpart of Recentrifuge in the area of metagenomic functional analysis.
|
10.1371/journal.pgen.1006879 | Open chromatin profiling identifies AP1 as a transcriptional regulator in oesophageal adenocarcinoma | Oesophageal adenocarcinoma (OAC) is one of the ten most prevalent forms of cancer and is showing a rapid increase in incidence and yet exhibits poor survival rates. Compared to many other common cancers, the molecular changes that occur in this disease are relatively poorly understood. However, genes encoding chromatin remodeling enzymes are frequently mutated in OAC. This is consistent with the emerging concept that cancer cells exhibit reprogramming of their chromatin environment which leads to subsequent changes in their transcriptional profile. Here, we have used ATAC-seq to interrogate the chromatin changes that occur in OAC using both cell lines and patient-derived material. We demonstrate that there are substantial changes in the regulatory chromatin environment in the cancer cells and using this data we have uncovered an important role for ETS and AP1 transcription factors in driving the changes in gene expression found in OAC cells.
| Oesophageal adenocarcinoma is one of the ten most prevalent forms of cancer and is showing a rapid increase in incidence and yet exhibits poor survival rates. Understanding the molecular causes of this type of cancer will enable us develop more effective treatment strategies which will improve survival rates. Here we have investigated how the genes in cancer cells are packaged into chromatin. We then compare this packaging to normal cells and use this information to identify the molecular causes leading to changes in chromatin packaging in cancer cells. We have identified a regulatory factor called AP1 that acts as a molecular switch to alter gene expression and hence cause cells to adopt a cancer fate. Importantly either this regulatory factor or a coregulatory protein from the ETS family is upregulated in the majority of cancer cells. Our study has therefore uncovered an important regulatory pathway that is commonly activated in oesophageal adenocarcinoma cells.
| The incidence of oesophageal adenocarcinoma (OAC) in the Western world is increasing and five and ten year survival rates remain low [1,2]. In comparison to many other cancers, there is a general lack of knowledge about biomarkers and potential therapeutic targets, contributing to the poor prognosis for patients with this disease. More recently, this situation has improved with several studies employing genome-wide approaches to further our understanding of the molecular defects in OAC. Several microarray studies have identified gene signatures that are of prognostic value [3,4] and recent genome-sequencing studies have uncovered new mutations and genomic rearrangements commonly found in OAC samples and associated these with disease progression [5–9]. Importantly many of the defects detected are in genes encoding proteins that affect chromatin structure such as mutations found in genes encoding the chromatin remodeling complex components ARID1A and SMARCA4 [5,6]. However, these mutations are also detected in the pre-cancerous Barrett’s oesophagous stage, whereas transcription factor mutations or amplifications occur more often after the transition to adenocarcinomas [9]. These findings suggest a model for adenocarcinoma development that involves alterations to the chromatin structure accompanied with reprogramming of the gene expression profiles driven by changes in transcription factor activity.
Several studies have implicated different transcription factors as important drivers of oesophageal cancer, chiefly due to their overexpression in OAC cell lines and/or patient derived OAC samples. Well studied examples include GATA6 [10,11], and FOXM1 [12,13]. Members of the ETS transcription factor family are often implicated as oncogenic cancer drivers, and this is best exemplified by the role of ERG, and PEA3 subfamily members in prostate cancers [reviewed in 14]. Indeed, members of the PEA3 subfamily, ETV1, ETV4 and ETV5 have been implicated in a wide range of cancers [reviewed in 15] and ETV4 has been implicated in oesophageal adenocarcinomas [16]. However, the target genes and mechanisms used by PEA3 transcription factors to control their expression in OAC are not known. Indeed, despite these studies, our understanding of the transcriptional control networks that are deregulated in oesophageal cancer is not well developed.
In this study, we investigated the changes that occur in the regulatory chromatin landscape in oesophageal adenocarcinoma by an unbiased approach using ATAC-seq. We identified AP1 and ETS transcription factors as important regulators in OAC cells and targeted ChIP-seq analysis combined with knockdown experiments reinforced the role of the ETS protein ETV1 in driving OAC-specific gene expression programmes. Similarly, loss of function approaches validated a regulatory role for AP1. Our results therefore demonstrate an important role for AP1 in OAC and part of its action is through a regulatory module containing AP1 and PEA3 subfamily ETS transcription factors. Importantly one or both of these factors are commonly upregulated in patient-derived OAC samples, and both factors are implicated in regulating the active open chromatin environment in these cells.
Our previous studies focussed on ETV4 and its role in OAC but also demonstrated that the closely related transcription factor ETV1 is upregulated in OAC [16]. We were unable to identify any antibodies that were suitable for ChIP-seq analysis of ETV4, therefore we instead focussed on ETV1. We verified that ETV1 is important for OAC cell growth as depletion of ETV1 reduced the growth rate of both OE33 and OE19 cell lines (S1A and S1B Fig). To determine how this transcription factor might impact on OAC and identify the regulatory networks it participates in we used ChIP-seq analysis to identify the direct targets of ETV1. We focussed on OE33 cells because ETV1 is expressed to the highest level in OE33 cells among the OAC cell lines we tested (S1C Fig).
The DUSP6 promoter was demonstrated to be target for PEA3 family proteins in two previous ChIP-seq studies in ECC-1 endometrial carcinoma cells [17] and PC3 prostate cancer cells [18](S2B and S2C Fig). We therefore used the DUSP6 promoter as a positive control to optimise the ETV1 ChIP protocol (S2A and S2D Fig). Two ChIP-seq replicate experiments were performed in OE33 cells and the binding peaks showed a high degree of concordance (86% overlap at q value <0.01). We therefore merged the two datasets and recalled the peaks giving a total of 498 ETV1 binding regions (q-value <0.01)(S1 Table). Very few of these binding sites are located in proximal promoters (ie +/- 1kb from the TSS) and instead are largely located in intronic and intergenic regions, suggestive of enhancer binding (Fig 1A). ETV1 binding was identified in the DUSP6 promoter region and other newly identified ETV1 targets such as an intergenic region associated with the EGFR locus (Fig 1B). One of the two highest scoring motifs found in the ETV1 binding regions was closely related to the CCGGAA core motif recognised by ETS transcription factors [19], and was found in 76% of targets, thereby validating the quality of the ChIP-seq data (Fig 1C). However, unexpectedly, the highest scoring over-represented motif contained the core sequence TGAG/CTCA which is the recognition site for AP1 transcription factors and this was found at a very high frequency in 65% of ETV1 binding regions (Fig 1C). Several other motifs were also enriched but at much lower levels (S3 Fig).
Next we associated the ETV1 peaks with the closest gene (398 genes in total; S1 Table) and looked for enriched gene ontology terms. All of the top enriched biological process terms are related to controlling apoptosis and the response to growth factor signalling (Fig 1D), consistent with the cellular growth defects we observed upon ETV1 depletion. We compared the expression of these genes between HET1A (non-cancerous, SV40 large T antigen transformed oesophageal cells) and OE33 (OAC-derived cells) by microarray analysis (S2 Table). The ETV1 target genes are generally expressed to higher levels in OE33 cells (Fig 1E), which is suggestive of a role for ETV1 in promoting cancer cell-specific gene expression. Consistent with this, when we depleted ETV1 in OE33 cells, microarray analysis identified 121/398 ETV1-associated target genes (30%) which showed significant differences in expression (>1.5 fold difference; p<0.05). Of these differentially regulated genes, the largest proportion 75/121 (62%) showed decreased expression following ETV1 knockdown (Fig 1F). We validated these results by RT-qPCR analysis of a panel of targets using a different ETV1 siRNA construct which was not present in the original siRNA pool. Importantly, the majority of the targets (7/9) showed similar downregulation in the presence of either the new siRNA or the original pool of siRNAs against ETV1 (S4 Fig). Many of the ETV1 regulated genes also showed consistent deregulation upon depletion of the related transcription factor ETV4, suggesting some level of functional overlap (Fig 1F).
Collectively, these data are therefore consistent with ETV1 having an important role as a direct transcriptional activator of dozens of genes in OAC-derived cells largely through distal regulatory elements. DNA sequence motif analysis is strongly suggestive that AP1 transcription factors likely function alongside ETV1 in this role (Fig 1G).
Many of the ETV1 binding regions are located in distal intergenic regions. To begin to establish whether these are associated with open and hence potentially “active” chromatin regions, we mapped the open chromatin regions in OE33 cells using ATAC-seq. We also mapped the open chromatin regions in two other OAC-derived cell lines, OE19 and FLO1 [20] and compared these to two non-cancer-derived “normal” oesophageal cell lines, HET1A and HEEPIC. Replicate experiments demonstrated that the experiments were highly reproducible (S5A–S5C Fig) and comparisons with histone marks indicated that the distal intergenic regions identified by ATAC-seq are associated with active enhancer regions (ie marked with H3K27ac and H3K4me1; S5D–S5F Fig). To identify regions which are more accessible in cancer cells and hence potentially more active, we first combined all of the reads from the three OAC-derived cell lines and the two normal cell lines and recalled the peaks of open chromatin which are characteristic of the cancer or normal cell lines. This gave us a comprehensive set of regions that are accessible in one or more cell lines. We then took the top 50,000 regions and used a 500 bp region around the summit for these peaks as an accessibility window across which we determined the variations in accessibility between cell types. We identified 1580 regions which are differentially accessible between the normal and the cancer phenotype (5 fold difference; p<0.05); 987 regions were more accessible in the cancer condition and 593 were more accessible in the normal condition (S3 Table). Clustering analysis using these differentially accessible peaks demonstrated that the replicate experiments cluster tightly and the normal and cancer-derived cell lines are also clustered together (Fig 2A). There are however differences apparent among the cancer samples with OE19 and OE33 cells showing more similarity to each other than to FLO1 cells which are clustered independently (Fig 2A, left).
Next we identified the locations of the differentially accessible peaks and found that the majority are located in intergenic and intronic regions (88%) with only 8–9% found in promoter regions (Fig 2B). Examples of the promoter-proximal and intergenic peaks clearly demonstrate the cell type specificity in either subsets of the OAC-derived cell types (Fig 2C) or normal oesophageal cells (S6A Fig). We associated the differentially accessible regions to the nearest gene and performed gene ontology analysis. The results are consistent with the expected effects of deregulating the putative target genes ie regions which became open and active in cancer cells were associated with genes belonging to GO terms encompassing various adenocarcinomas (Fig 2D) and expression profiles associated with various intestinal organs (S6B Fig). Conversely, the regions which were closed in cancer cells are associated with different GO terms, chiefly associated with various signaling processes (S6C Fig). Finally we asked whether the regions which change accessibility are indicative of changes in gene expression between normal and cancer cells. Importantly we demonstrate that regions which become more open in cancer cells are associated with genes which are more highly expressed in OAC-derived OE33 cells (Fig 2E). On the contrary, regions which become closed in cancer are associated with genes whose expression is lower in the OE33 cells (Fig 2E). The analysis of differentially accessible peaks identified by ATAC-seq therefore reports on regulatory events that likely contribute to the changes in gene expression observed in cancer cells.
To identify potential regulatory factors responsible for generating the open chromatin regions in OAC-derived cells, we searched for sequence motifs that were enriched in the peaks that were either more open or more closed in cancer cells. The top ranked motif in the regions becoming more accessible in cancer cells was the AP1 transcription factor recognition site TGAG/CTCA (Fig 3A). Several other over-represented motifs recognised by the transcription factors, FOXA1, KLF5 and GATA4 were also identified. Importantly, we could detect a binding footprint across the AP1 binding motifs specifically in the cancer-derived ATAC-seq data from OE33 cells, indicating that the motifs are likely occupied in a large number of regions (Fig 3B; S7B Fig). Somewhat counter-intuitively, AP1 motifs were also the highest ranked over-represented motifs in the regions showing lower accessibility in OE33 cancer cells (S7A Fig). However a much less pronounced footprint at these AP1 motifs was obvious in the non-cancerous HET1A cells (S6B Fig) suggesting lower occupancy levels. To validate AP1 occupancy we tested the binding of the AP1 subunit JUN to six regions which show enhanced accessibility in OAC cells and contain an AP1 motif. All of these regions exhibited JUN binding (Fig 3C). ETS motifs were not among the top ranked over-represented motifs in the chromatin regions that are more open in cancer cells. However, we searched for the degenerate versions of the ETS motif CC/AGGAA/T and found this motif to be highly enriched around the centre of these open chromatin regions, although the local background levels were higher than observed for AP1 (Fig 3D).
These data are indicative of a role for AP1 in acting through the open chromatin regions to drive increases in gene expression in cancer cells. To determine whether this was the case, we disrupted AP1 activity by transducing OE33 cells with a lentivirus expressing a dominant-negative version of FOS (DN-FOS) which forms non-DNA binding heterodimers with endogenous JUN family members [21]. We observed 1527 upregulated and 2087 genes downregulated following DN-FOS expression (linear 1.5 fold change; q<0.01; FPKM >10)(Fig 3E; S4 Table). Importantly, of the 626 genes associated with the regions showing increased accessibility in OE33 cancer cells compared to HET1A cells and containing an AP1 motif, 27% show reduced expression in OE33 cells in the presence of DN-FOS. Next we partitioned the genes associated with regions that are more accessible in OE33 cancer cells into two groups according to the presence or absence of an AP1 motif. We found that there is a clear reduction in gene expression after expression of DN-FOS only in the group containing an AP1 motif (Fig 3F). These data are fully consistent with a model whereby AP1 is occupying and activating a significant proportion of the genes with regulatory regions which are becoming more accessible in cancer cells. Importantly inhibiting AP1 function affects cohorts of genes involved in cell growth, survival and movement which are all functions previously attributed to AP1 in other cellular environments ([22]; Fig 3G).
Given our identification of a regulatory module containing AP1 and ETS motifs, we asked whether ETV1 regulated genes were also regulated by AP1. Inhibition of AP1 function by DN-FOS inhibited a large proportion of the ETV1 activated genes (ie genes showing downregulation upon siETV1 treatment), with 32% (233/738) also showing downregulation after AP1 inhibition (Fig 3H; hypergeometric test, P = 3.8x10-50). The active chromatin state is usually coincident with the appearance of histone H3K27 acetylation. We therefore tested whether inhibition of AP1 activity by DN-FOS also affected the levels of this chromatin mark. At all four loci tested, DN-FOS reduced the levels of gene expression but with the exception of MMP1, failed to change the levels of H3K27ac (S8 Fig). Thus, the ETV1-AP1 binding module appears functionally relevant in OAC cells but is not generally required for maintaining histone acetylation levels.
Finally we investigated whether the genes controlled by AP1 in OE33 cells are relevant in the context of human OAC samples. To do this we defined a high confidence set of putative AP1 target genes, by first associating the differentially accessible peaks that are found to be more accessible when comparing all tumour and normal cell lines genes with the nearest gene and then selecting those peaks containing an AP1 motif within a 500 bp window of the peak summit. We then filtered for a reasonable level of expression (>5 FPKM), and >1.3 fold reduction in expression following DN-FOS expression which gave us a total of 58 genes which are high confidence direct AP1 targets. We then examined the expression of these genes in a panel of OAC samples [4] and plotted their expression according to AP1 status in each sample. The resulting heatmap demonstrated a clear correlation with AP1 expression levels across many of the target genes, with higher levels of AP1 expression generally being associated with high level target gene expression (both generally red on the left side of the heatmap) and reciprocally, low AP1 levels are associated with low target gene expression (both generally blue on the right side of the heatmap) (Fig 3I; S9 Fig).
Collectively, this data identifies the AP1 transcription factor as an important player in driving the gene expression profiles found in OAC cells.
Having established the open chromatin architecture of OAC-derived cells, we next asked whether the ETV1 binding regions we identified through ChIP-seq analysis are associated with open and potentially active chromatin. There is a strong overlap between the level of ETV1 binding and the presence of open chromatin in OE33 cells (Fig 4A). Furthermore, the ETV1-bound regions show increased accessibility compared to the non-tumourigenic HET1A cells (Fig 4B), suggesting an association with the acquisition of open chromatin. Indeed, further comparisons between the accessibility of the ETV1 binding regions in HET1A cells versus three different OAC-derived cell lines identified three broad clusters of sites (Fig 4C; S5 Table). Almost all the ETV1 binding regions show greater accessibility in OE33 cells than in the HET1A cell line. Indeed, one cluster shows increased accessibility which was largely limited to the OE33 cell line. However, the second cluster shows generally increased accessibility in all of the OAC cell models and this pattern is exemplified by binding at the DUSP6 promoter (Fig 4D, top) whereas the largest third cluster is defined by increased accessibility in both the OE19 and OE33 cell lines as observed at the ADAP1 locus (Fig 4D, bottom). Several of the genes showing increased chromatin accessibility across all samples have previously been shown to be associated with cancer. These include CTSB in OAC [23], BAIAP2L1 in ovarian cancer [24] and BCAR1 in a range of cancers [25]. We also examined whether we could see any transcription factor footprints in the regions occupied by ETV1 and as expected we could see a clear footprint at the ETS binding motif that is recognised by ETV1 (Fig 4E). We also asked whether we could observe a footprint at the AP1 motif which was identified as the most significantly over-represented motif in the ETV1 binding regions. Again we were able to identify a clear footprint in OE33 cells and at both the AP1 and ETS motifs, the footprints were clearly more enhanced in OE33 cells compared to the HET1A cell line (Fig 4E). Thus in addition to motif presence, this data indicates that these motifs are occupied in cancer cells, and further supports the existence of combinatorial actions of ETV1 and AP1 in controlling gene expression in OAC-derived cells.
We next examined whether the presence of associated ETV1 binding and potential AP1 co-occupancy had any relevance to gene expression in the context of human tumour biopsy samples. A high confidence set of ETV1 target genes were selected based on being associated with an ETV1 binding region which is open in OE33 cancer cells and also contains either an AP1 or ETS motif (or both). Target gene expression was then examined and samples clustered according to similarity in expression patterns (Fig 4F). This identified three major clusters, one small one with generally low level gene expression and two with generally elevated expression levels. Interestingly the latter two clusters, SC2 and SC3, are characterised by high level expression of PEA3 or AP1 family transcription factors respectively, suggesting that overexpression of either one of these classes of transcription factors is sufficient for enhanced target gene expression.
Together these results demonstrate that ETV1 binding occurs concomitantly with AP1 and is generally associated with chromatin which becomes more open and active in cancer cells. However, while some regions are specific to OE33 cells, others are commonly activated in several OAC-derived cell lines.
Our data from ChIP-seq and ATAC-seq studies in OAC-derived cell lines implicate ETV1 and AP1 transcription factors as important players in driving OAC-specific gene expression programmes. One prediction of these findings is that we should see greater expression of these transcription factors in patient-derived OAC samples compared to normal oesophageal tissue. AP1 is a heterodimeric transcription factor consisting of either homodimers of JUN family members (JUN, JUNB and JUND) or heterodimers of JUN family members with a FOS family member (FOS, FOSB, FOSL1 and FOSL2)[reviewed in 22]. ETV1 is structurally highly related to two other ETS proteins ETV4 and ETV5 which collectively form the PEA3 family and likely perform overlapping functions [15]. We therefore examined the expression of the individual AP1 family subunits and PEA3 family members in biopsies from 73 patients (S6 Table). In general all of the PEA3 and AP1 family members are expressed at low levels in normal samples with the exception of FOSL2 (Fig 5A). The expression across OAC samples is varied and three broad clusters could be identified (Fig 5A). The first cluster has generally high levels of AP1 subunits whereas the second cluster has high level expression of PEA3 family subunits (including ETV1). Together these clusters represented the majority of the samples but 29% formed a third cluster which is characterised by high level expression of individual ETS and AP1 family subunits rather than coordinated upregulation of several family members. Overall patients with OAC therefore exhibited either coordinated upregulation of AP1 subunits or PEA3 subunits or combinations of individual subunits from the two transcription factor families.
To verify these findings, we also interrogated a published microarray study on OAC-derived samples which also included data from the pre-cancerous Barrett’s oesophagus stage [4]. Again, we observed generally low level expression of both PEA3 and AP1 family members in normal tissue with the exception of FOSL2 which is higher (Fig 5B). Low level expression of all of the genes encoding these transcription factors was generally observed in samples from patients with Barrett’s oesophagus. In contrast, there are groups of patients with OAC which had either high level expression of one or more genes encoding PEA3 family members or subunits of the AP1 complex, and others that had mixed upregulation of both types of transcription factors. Interestingly when looking at average expression of individual transcription factors across all samples, the patterns we observe by comparing the entire families are less obvious although ETV4 and FOSL1 are generally overexpressed in the context of OAC (Fig 5C; S10 Fig). Collectively, these results are broadly in keeping with what we observed from our own patient cohort and indicate that upregulation of PEA3 family members or AP1 subunit components is characteristic of a large number of cancers from OAC patients.
Finally we wanted to understand the contribution of individual transcription factors to driving the activity of the open chromatin regions in OAC cells. We therefore correlated the mean gene expression changes for individual transcription factors in patient-derived OAC samples with the probability of observing their recognition motif in the regions showing increased accessibility in cancer-derived cell lines. This approach would enable us to determine whether any particular PEA3 family members or AP1 subunit components might be more important in this context. This analysis identified the FOS sub-family members, FOSL1, FOS and more markedly FOSB as the likely drivers of enhanced AP1 activity in patient samples (Fig 5D). We also identified FOXA2 and FOXA3 as potential regulators of OAC-specific gene expression. Conversely, FOSL2 was identified as a possible contributor to the changes in activity resulting in more open chromatin in normal cells (S11 Fig).
Overall, the expression data from patient-derived tumour material support a role for AP1 and PEA3 family members in contributing to OAC-specific gene regulatory events and in particular are strongly indicative of a role for members of the FOS-subfamily in this process. However, none of the clusters identified using AP1 and PEA3 family gene expression profiles associated with any particular patient clinicopathological characteristics, suggesting that overexpression of one or more of these factors is a general property of OAC.
Our data using cancer-derived cell lines, combined with gene expression data from cancer patients provides a compelling model which points to AP1 and PEA3 family members in driving OAC-specific gene expression. To further support this model, we used ATAC-seq to interrogate the regulatory open chromatin landscape of OAC-derived samples (3 normal samples and 6 OAC samples) and searched for evidence for regulation by these transcription factors. First we merged the reads from all the tissue samples both normal and cancer. We then identified the top 50,000 peaks from these merged reads and performed principal component analysis (PCA). The normal samples (blue) cluster together whereas the tumour samples (red) separated from the normal samples (Fig 6A). Furthermore on PC2 the tumours appear to separate into two groups with ATAC-001T and ATAC-004T being closer to the normal tissue samples and the others representing a distinct group which show similarity predominantly on PC2. To extend these results we identified the regions which are differentially accessible in cancer cells, by combining all of the reads from either the six OAC-derived samples or the three normal samples and recalled the peaks of open chromatin which are characteristic of the cancer or normal samples. This identified 1015 regions that showed significant differential accessibility (5 fold difference in tag counts; p-value <0.05) between the normal and the cancer samples. 964 regions are more open in the cancer cells and 52 more open in the normal cells indicating that the cancer cell genome is generally in a more open state (S7 Table). The genomic distribution of these regions is virtually identical to that observed in cell line models with the majority in putative intronic and intergenic regulatory regions with the biggest enrichments in the intergenic regions (Fig 6B). Of the 956 regions which exhibit increased accessibility in tumour tissue, 130 (14%) are also identified in the context of enhanced opening commonly found in OAC-derived cell lines. We then generated a heatmap of these regions in individual samples and probed their relationships using hierarchical clustering (Fig 6C). The normal samples all clustered together and the tumour samples were split into two groups with four tumours (cluster 2; ATAC-002T, ATAC-003T, ATAC-005T and ATAC-006T) appearing to have a similar pattern of increased accessibility across the samples whereas the other two tumour samples ATAC-001T and ATAC-004T (Cluster 1) cluster separately and appear to be closer to the normal tissue samples. Thus both clustering and PCA indicates the presence of two cancer subtypes. There are no obvious clinicopathological variables that segregate with the two clusters, but due to the small sample size more samples are needed to make any definitive conclusions.
The differentially accessible regions also point to potential differences in the underlying biology of OAC, as after associating these regions with the nearest gene, GO analysis identified various cancers and abnormalities in intestinal morphology as the most enriched categories (S12 Fig). By focusing in on the associated genes we are also able to pinpoint the location of potential regulatory elements such as the promoter proximal, intronic and distal elements associated in cancer cells with the IHH locus (Fig 6D, top). The activation of the IHH locus is consistent with recent discovery of IHH overexpression in OAC [26]. Reciprocally we can observe promoter elements being extinguished in cancer cells as exemplified for ZNF471 and ZFP28 (Fig 6D, bottom). Importantly, the appearance and loss of these open regions corresponds to changes in gene expression of the associated loci. IHH expression is upregulated in the cancer cells from cluster 2 tumours where regulatory chromatin accessibility increases (S13A Fig). However ZNF471 behaves in the opposite manner and is expressed at lower levels in cancer cells compared to matched normal tissue (S13B Fig). To provide evidence that the differentially accessible regions we identified are representative of active chromatin regions, we took advantage of a published ChIP-seq dataset for H3K27ac (an active chromatin mark) in normal oesophageal tissue. We then analysed the tag counts for this mark in the regions which are more open in either normal or cancer cells. As predicted, the regions which are more accessible in normal cells and hence are likely more active, exhibit substantially higher H3K27ac signal than those that are newly accessible in cancer cells (Fig 6E). We then asked whether the motifs recognized by AP-1 and PEA3 family members (ETS motifs) are over-represented in the regions becoming more accessible in cancer cells and found that both AP-1 and ETS motifs are highly enriched in these regions (Fig 6F). Importantly, footprinting analysis in the differentially open chromatin regions demonstrates potential occupancy of these motifs in cluster 2 OAC samples by both AP1 (S14A Fig) and ETS family members (S14B Fig), but provides little evidence for binding at these motifs in the normal cells. Finally, given the prevalence of ETS motifs we asked whether there was evidence for enhanced levels of open chromatin in cancer samples at regions occupied by ETV1 in the OE33 OAC cell line. We focused on the matched samples and found that elevated levels of open chromatin could be found around the ETV1 binding regions in two of the cancer samples; 006-T (Fig 6G) and 005-T (S15A Fig), with a much reduced effect in 003T where the expression of ETV1 is much lower (Fig 6G; S15A and S15B Fig).
Together these data therefore provide independent verification of our model that AP1 and PEA3 family members act combinatorically to control the regulatory chromatin network underlying the phenotypic gene expression patterns found in OAC cells.
Phenotypic changes in cancer cells are ultimately elicited by changes in their gene expression profiles. Alterations to the underlying regulatory chromatin landscape of cancer cells facilitate these changes [reviewed in 27]. Here, we have uncovered the regulatory open chromatin landscape of OAC cells, and used this to infer the transcriptional regulators that are responsible for driving OAC-specific gene expression programmes. We integrated this data with ChIP-seq analysis of the ETS Transcription factor ETV1. ATAC-seq analysis in OAC derived cell lines and samples from patients with OAC identified AP1 and ETS factors as potential regulators. Reciprocally, ETV1 ChIP-seq implicated AP1 as a potential regulatory factor in OAC. Our results therefore point to two different transcription factor families in directing these programmes; AP1 and members of the PEA3 subfamily of ETS proteins. This is suggestive of a model where AP1 plays a leading role in controlling OAC-specific gene expression and part of this activity is mediated through working alongside PEA3 subfamily members to control gene activation (Fig 1G). However, although the model suggests a close molecular association between these transcription factors, their binding sites show no obvious distance constraints that would imply a distinct binding mode that would drive cooperative DNA binding. Further studies are needed to establish whether there is a binding hierarchy between these factors. Loss of function experiments confirms the regulatory activities of both ETV1 and AP1 in this regulatory module. Importantly, consistent with this regulatory role, either AP1 subunits, PEA3 subfamily members or both are over-expressed in the majority of OACs. Although we identified these transcription factors through their association with cancer-specific open chromatin regions, it is currently unclear whether one of both of these transcription factors is responsible for driving chromatin opening. Our results indicate that AP1 is not required for maintaining “active” histone acetylation levels but further studies are required to determine the roles of these factors in establishing and long term maintenance of the regulatory chromatin landscape.
Previous studies have inferred regulatory interactions between AP1 and ETS transcription factors and extrapolated this in the context of different cancers. For example, studies on the viral enhancers and the stromelysin and collagenase promoters initially identified functional cooperativity between these transcription factor families [28–30]. Since then, genome-wide studies have inferred potential regulatory cooperativity between AP1 and other ETS factors such as ELK1 [31] and ERG [18] and in the latter case this association is thought to be important for the context of prostate cancer. Regulatory interactions between the PEA3 subfamily member ETV5 and the AP1 component JUN have been identified [32] and in the case of ETV1 and ETV4, the interactions with AP1 have been expanded into a genome-wide view of the global consequences of these interactions [18]. It is thought that overexpression of ETS and AP1 factors can mimic upstream RAS/ERK signalling events and drive tumourigenesis in this context. It is therefore interesting to note that we previously showed co-upregulation of ERK signalling and ETV4 expression in OAC [16] and that genes encoding members of the RAS/ERK signalling cascade such as EGFR and KRAS are often amplified in the transition from Barrett’s oesophagus to adenocarcinoma [9]. Here we identify a novel association between AP1 and ETS factors in oesophageal cancer, and in particular with PEA3 subfamily members. In the case of AP1, we generally see overexpression of many different subunits in OAC with the exception of FOSL2 whose expression decreases. This observation is intriguing given the fact that we also see a set of AP1-bound regions as potentially inactivated in cancer cells and might point to FOSL2 playing a regulatory role at these regions in the context of normal oesophageal cells.
Our results indicate that assessing AP1 and PEA3 transcription factor status might be a good indicator of OAC status. However, we could not detect any associations with disease stage or patient treatment regime. This suggests that the PEA3-AP1 regulatory module more likely contributes more generally to the cancer phenotype. In keeping with this observation, depletion of ETV1 and/or ETV4 causes an OAC cell growth defect ([16]; S1 Fig) and interfering with AP1 function also affects genes involved in growth and survival of OAC cells (Fig 3G). The PEA3-AP1 module and its regulatory network might therefore represent a target for therapeutic intervention. The PEA3-AP1 regulatory module had not been identified in these cancers before, most likely due to the complex nature of the AP1 and PEA3 family members involved in constituting this activity. Thus overexpression of any particular pairs of proteins would not necessarily have resulted in sufficient predictive power. This demonstrates the power of using ATAC-seq which reports on regulatory activities of transcription factor families rather than individual transcription factor activity. Moreover, the ability to apply ATAC-seq to low numbers of cells, makes this a particularly attractive method to use on the limited cell numbers available from patient-derived samples. The application of ATAC-seq to other cancers would therefore be a powerful approach to revealing new transcriptional regulatory events that contribute to tumourigenesis in these contexts. Moreover, ATAC-seq can be subjected to clustering analysis in a similar manner to RNA-seq to sub-partition cancers into different subtypes (Fig 6) and hence can be used in stratification strategies. By combining with single cell approaches [33], ATAC-seq has the potential to be able to uncover cancer heterogeneity and follow cancer evolution. This approach therefore has the potential to be used in both cancer diagnosis and stratification in addition to uncovering novel regulatory mechanisms that may prove amenable to therapeutic intervention.
Ethical approval for collection of oesophageal tissue samples from patients at the Royal Albert Edward Infirmary, Wigan and the Salford Royal Hospital were granted by the ethics committees at Wrightington, Wigan and Leigh NHS Foundation Trust (2007) and Salford Royal NHS Foundation Trust (2010) respectively (04/Q1410/57). Patient consent was obtained in written form and signed by the patient and doctor.
The OE33 and OE19 cells were cultured in RPMI media supplemented with 10% foetal bovine serum. The FLO1, HET1A and 293T cells were cultured in DMEM culture media supplemented with 10% FBS. HEEPIC cells were grown on poly-l-lysine coated plates (2 μg/cm2) in epithelial cell media supplemented with 1% epithelial growth supplement. Biopsy tissue samples (~4 mm) were processed as described previously (Wiseman et al., 2015).
Total cellular RNA was isolated from cell line and clinical tissue samples as described previously [16]. When required, short interfering (si) RNAs directed against human ETV1 or ETV4 (SMARTpools; Dharmacon), or a non-targeting pool (Dharmacon) were used in 24 hr transfections prior to RNA extraction as described previously [13]. For the validation experiments with either the SMARTpool or the single duplex siRNA construct against ETV1 (siGENOME human ETV1 siRNA, D-003801-04; Dharmacon), cells were left for 48 hrs after transfection. RT-qPCR was carried out as described previously [16] with the appropriate primer pairs (S8 Table). For analysis of the human tissue RNA samples, nanolitre volume RT-qPCR was performed using the Fluidigm Biomark HD system using EvaGreen chemistry according to the manufacturer’s instructions. The final relative expression was calculated by normalising to the geometric mean of the house-keeping genes SDHA, ALAS1, GAPDH, and HMBS using the delta CT method.
In knockdown experiments statistical significance was calculated using an unpaired two-tailed Student’s T test with a two sample equal variance. Gene expression data comparing expression in different groups of samples are represented with boxplots generated using Gene-e software. Outliers are not shown but represent values >1.5 interquartile ranges from the 25th or the 75th percentile (e.g 75th percentile + (1.5 x IQR) and 25th percentile–(1.5x IQR)). Statistical significance was assessed using a 2 tailed t-test calculated using StatPlus Microsoft Excel unless otherwise specified.
For expression microarray analysis of HET1A, HEEPIC, and OE33 cells (plus/minus siETV1/ETV4), biological triplicate RNA samples for each condition were processed. The raw intensity files (CEL) were generated by processing 500 ng of total RNA on Affymetrix HTA 2.0 arrays, according to the manufacturer's instructions (Affymetrix, Santa Clara, CA). The arrays were scanned through GENECHIP Scanner-7G (Affymetrix, CA). The CEL files generated by these arrays were converted into rma-gene-ful.chp and rma-alt-splice-dabg.chp files through Affymetrix Expression Console Software (version 1.3). The CHP files were analyzed through the Transcriptome Analysis Console v3.0 (TAC). Within the TAC software an excel spreadsheet for all conditions with the mean relative expression (n = 3) of all genes was generated. From this TAC also calculated fold change with between conditions. For all downstream analysis involving fold change genes this spreadsheet was used to generate gene lists for further analyses. Data are deposited in ArrayExpress (Accession number: E-MTAB-5163).
To generate samples for RNA-seq analysis, we transduced OE33 cells (3 biological replicates) with lentiviruses expressing a dominant-negative (DN) FOS (DN-FOS) construct (pInducer-DN-FOS) or lacking an insert (control)(pInducer) and grew for a further 12 hours. pInducer-DN-FOS was constructed by inserting a BamHI/EcoRI fragment from pBABE-puro-a-FOS/pAS2804 [21] into the same sites in pENTR1A (Addgene plasmid # 17398). The insert was then transferred to pInducer20 (a gift from Stephen Elledge; Addgene plasmid #44012; [34]) by Gateway cloning. Following transfection cells were treated with doxycycline (100 ng/ml) for 48 hours to induce DN-FOS expression and GFP positive cells were selected before harvesting RNA. The RNA-seq libraries are generated using the TruSeq stranded mRNA sample prep kit and 16 samples run on a NextSeq platform generating approximately 25 million reads per sample. Reads were first trimmed to remove Illumina adapter sequence using trimmomatic 0.3 [35]. Trimmed reads were aligned to the ensemble transcription (release 72) human genome 19 (hg19) using the RNA-Star aligner (version 2.3.0e) [36]. Differential expression analysis was carried out using Cuffdiff [37] in default settings. Data are deposited in ArrayExpress (Accession number: E-MTAB-5175).
ChIP-qPCR and ChIP-seq were carried out as described previously [13]. For ChIP-qPCR for AP1, a JUN antibody (Abcam (ab31419)) was used and for histone acetylation a H3K27ac antibody (Abcam (ab4729)) was used. For ChIP-seq, 3x107 cells, 3 μg antibody (ETV1; Abcam (ab81086)) and 30 μl Dynabeads were used per experiment. Parallel control experiments were run with ChIP-qPCR using rabbit IgG; Millipore (12–370). Library preparation was performed using the TruSeq ChIP Sample Preparation Protocol (Illumina) and DNA libraries were sequenced using the HiSeq 2500 (Illumina).
Sequencing tags/reads from the ETV1 ChIP-seq experiment in OE33 cells were aligned to the NBCI Build hg19 of the human genome with Bowtie v2.2.3 [38]. Up to two mismatches were allowed. Only reads with a mapping quality >q30 were retained. Peak calling was performed on individual replicates and merged datasets with MACS v2.1.0 software [39] using default parameters. Data are deposited in ArrayExpress (Accession number: E-MTAB-5168)
ATAC-seq data generation and analysis on OE19 and HEEPIC cell lines was performed as described previously for the HET1A, OE33 and FLO1 cell lines (Accession number E-MTAB-4209; [20]). To produce ATAC-seq libraries from human tissue samples fresh tissue was transported from the endoscopy department to the lab and processed within 1 hour of sampling. Alternatively, frozen tissue samples were used (surgical or endoscopic resections were immediately frozen down upon removal at -80°C, initially in liquid nitrogen, and then transferred to -80°C freezer). Frozen or fresh human tissue was first washed with 1x Dulbecco’s PBS. The tissue was then minced using scalpel and scissors. Minced tissue was lysed in 10 ml of fresh cold ATAC lysis buffer as described in [40] on ice. This suspension was briefly (5-10secs) vortexed every 10 mins during the lysis period and placed straight back onto ice. After 30 mins of lysis the lysed minced tissue was filtered first through 100 μm membrane then 20 μm membrane using low pressure vacuum driven filter units (Steriflipf filter units; Merck).The nuclei were then pelleted at 500g 4°C for 10 mins. The supernatant was removed and the pellet resuspended in 500 μl PBS and transferred to 1.5 ml microcentrifuge tube. The sample is again centrifuged at 500g 4°C for 10 mins to pellet the nuclei. The supernatant is removed and the nuclear pellet is resuspended in 25 μl nuclease free water before being quantified and then taken forward to the transposition reaction and processed as per the cell lines. Quantification of the nuclei is carried out by removing 2.5 μl of the resuspended nuclei and diluting to a total volume of 10 μl with 1x trypan blue and analysing on a haemocytometer.
Differentially accessible regions in cancer versus normal cells were identified by merging bam files of all of the conditions to be compared and recalling peaks using MACS2 [39]. To study higher confidence regions, a 500 bp window around the summit of the top 50,000 regions identified were then analysed for differential accessibility between different cell types using Cufflinks [41]. Differentially accessible regions were deemed to be those with a linear fold change of >5 fold and a p-value of <0.05 as determined by Cuffllinks. The differentially accessible regions were then taken forward for further analysis using a 500bp window centred on the summit of the region. Data are deposited in ArrayExpress (Accession number: E-MTAB-5169).
To visualise ATAC-seq data, normalised cleavage events across the differentially accessible regions were counted using HOMER [42] to produce heatmaps drawn using GENE-E (http://www.broadinstitute.org/cancer/software/GENE-E/). Hierachical clustering was carried out using this software, and all clustering was using one minus Pearson’s correlation unless otherwise specified. De novo motif discovery in ATAC-seq and ChIP-seq was carried out using HOMER [42] with the background normalised using–cpg parameter.
Gene annotation was performed using HOMER [42] to identify the closest gene to the ATAC-seq or ChIP-seq peak summits using the co-ordinates from the Refseq Hg19 v.37 protein coding list. The nearest gene was ascribed to the binding peak when the summit of the peak occurred within 100 kb upstream of the transcription start site (TSS). Gene ontology (GO) analysis was performed using the GREAT web application (http://bejerano.stanford.edu/great/public/html/) [43] using NBCI Build 37/hg 19 of the human genome. P-values for GO Terms shown are log10 of the binomial p-value generated using GREAT software [43].
Tag density heatmaps and profiles were generated using HOMER [42] using default settings and visualised using JavaTreeView 3.0 [44].
|
10.1371/journal.pntd.0006808 | Development and validation of a scale to assess attitudes of health care providers towards persons affected by leprosy in southern India | Assessment of attitudes of health care professionals is important as negative attitude could constitute a major deterrent to care-seeking by persons affected by neglected tropical diseases (NTDs) such as leprosy. Leprosy continues to pose a major disease burden in India with an annual new case detection rate of 10.17 per 100,000 population. This paper reports on the development and validation of a culturally appropriate scale to measure attitude of health care providers (HCPs) towards persons affected by leprosy in Tamil Nadu, India.
The Affective, Behavioural and Cognitive (ABC) model of attitudes guided the development of the scale. Steps in scale development included qualitative interviews and focus group discussions with medical officers and paramedical staff selected from high prevalence districts in Tamil Nadu, India which informed the development of the draft scale. Reviews of existing attitude questionnaires in related areas further contributed to scale development and together helped to generate a large pool of items which was then subjected to Thurston’s scaling method for selection of items from this pool. Face and content validity were obtained, following which internal consistency and test, re-test reliability were assessed. Scaling exercise resulted in 11 items being discarded from an initial pool of 38, owing to the poor agreement among experts regarding relevance. Face and content validity were good with experts endorsing relevance and applicability of items. The intra-class correlation coefficient (ICC) for test re-test reliability of the 27 item scale was 0.6 (95% CI: 0.20–0.78) indicating marginal intra-class correlation. The overall Cronbach’s alpha was 0.85 while the alphas for each of the affective and behavioural components was good at 0.78 and 0.69 respectively indicating a good degree of consistency and homogeneity between items but the alpha for the cognitive component was low at 0.53.
The ABC model of attitudes guided the development of the scale, ensured a mix of 27 items tapping into the three domains of Affect, Behaviour and Cognition which best explained the attitude construct. With good validity and alphas for each of the affective, behavioural components and overall alpha estimates, this scale can be a valuable tool to provide accurate estimates of the true attitudes held by HCPs. This, in turn, would be useful to obtain insights for appropriate intervention programmes that would help change negative attitudes of HCPs towards persons affected by leprosy. With some adaptations, the scales can be validated for other NTDs as well.
| Leprosy is an infectious disease caused by the Mycobacterium leprae and is one of the major causes of preventable disability. Early diagnosis and prompt treatment of all new cases of leprosy remain the key strategies for leprosy control as it would prevent nerve damage, disability and reduce the transmission of the disease. People affected by leprosy often experience severe stigmatization because of an adverse social judgment about the disease or its disabling consequences. This neglected tropical disease continues to pose a major disease burden in India. Despite the availability of health facilities there continue to be barriers towards leprosy diagnosis and early treatment. Assessment of attitudes of health care professionals is important as negative attitude could constitute a major deterrent to care-seeking by persons affected by leprosy. Researchers developed and validated a culturally appropriate scale to measure attitudes of health care providers towards persons affected by leprosy in Tamil Nadu, India. The scale would be useful to obtain insights of attitudes of health care professionals to plan appropriate programmes that would help to promote positive attitudes of healthcare providers towards persons affected by leprosy.
| Leprosy continues to pose a major disease burden in India with an annual new case detection rate of 10.17 per 100,000 populations. An average of 0.13 million new leprosy cases are detected every year in the country [1]. Despite the availability of health facilities there continue to be barriers towards leprosy diagnosis and early treatment. People affected by leprosy often experience severe stigmatization as a result of an adverse social judgment about the disease or its disabling consequences [2]. Their family members in turn experience negative attitudes, face social isolation and other discriminatory practices [3–5]. As a consequence, patients and their families often feel forced into stopping treatment for fear of being exposed, thereby contributing to increased morbidity and risk of disability [6]. The problem can be further compounded by negative attitudes of Health Care Providers (HCPs) which could constitute a major deterrent to care-seeking by patients. In a study carried out in Mumbai, it was found that 75% of general physicians who were untrained in the management of leprosy believed that isolation and treatment of leprosy patients was necessary, while 59% were opposed to social integration of leprosy patients even after complete cure [7].
Attitudes and beliefs about leprosy, are shaped by factors like knowledge about the disease, opportunities to interact with persons having the condition, influence of cultural stereotypes about leprosy, influence of media, familiarity with institutional practices, religion and past restrictions [8]. Positive attitudes held by HCPs towards persons affected by leprosy facilitate better access and utilization of health services [9]. Furthermore, if HCPs are seen as being willing and are comfortable treating persons affected by leprosy, it could lead to improved quality of care and enhance the value of these services [10]. By their training and knowledge about the disease, HCPs are usually expected to bear positive attitudes towards persons affected by leprosy [9]. Unfortunately, this may not always be the case. Croft and Croft report that the negative attitudes of health care workers acted as a block to the delivery of holistic health care for persons affected by leprosy and also contributed to reinforcing harmful traditional beliefs about leprosy in the community [11]. Reports also suggest that medical doctors too have been known to avoid providing care to persons affected by leprosy to the extent of even refusing to treat them [12]. The resultant sense of discrimination and shame that persons affected by leprosy face, invariably, causes them to avoid seeking care.
Persons affected by neglected tropical diseases (NTDs), such as leprosy, can experience stigma. In addition to the direct effects of societal stigma, persons suffering from NTDs face several forms of structural stigma, such as lack of resources allocated to this neglected group and low training levels and negative attitudes amongst health care staff. This, in turn, affects the availability, quality and uptake of treatments offered, resulting in poor treatment outcomes and persistent stigma. Negative attitudes amongst HCPs lead to low training levels, poor management resulting in poor prognosis & treatment outcomes, high visibility of illness [13,14].
Given that, persons affected by leprosy could also develop serious disabilities; that there are barriers towards diagnosis and treatment of leprosy attributable to stigma, the need to develop/modify and validate a culturally specific scale to measure the attitudes of HCPs towards persons affected by leprosy assumes significance. Such a validated scale will provide critical information on how HCPs perceive persons affected by leprosy thereby enabling appropriate interventions aimed at changing negative attitudes.
The Affective, Behavioural and Cognitive (ABC) model of attitudes guided the development and selection of items for this scale [15]. According to this model, attitude is comprised of 3 components- affective, behavioural and cognition. The affective component refers to emotional reactions individuals have towards the attitude object; the behavioural component refers to the way individuals behave when exposed to the attitude object and the cognitive component refers to an individual’s beliefs and knowledge about the attitude object.
A web search revealed 11 studies that assessed the knowledge, attitudes and practices of health care providers towards persons affected by leprosy. Most of these studies had been carried out in African countries [8,9,12,16–20] with three that had been undertaken in India [21–23]. Based on our review, we discovered that barring a few items, the majority of the items in the different scales were more representative of knowledge and practice rather than attitude thereby justifying the development of such a scale. The attitude items from these different scales used in the 11 studies were listed and served as the item pool from which- following information generated from the qualitative work- we made decisions on the ones to retain. The initial pool of items from these studies were examined for equivalence. Beaton et al. (2000) guidelines on cross cultural adaptation (to look for semantic, conceptual, experiential and idiomatic equivalence) of self-report measure guided us in this process. For example, one of the items that we did not include was “cause of leprosy is because of witches”. This was because the concept of ‘witches’ is not integral to the Indian culture [24].
Permission to conduct the study was obtained from the Directorate of Public Health & Preventive Medicine (DPH&PM) and the research protocol was approved by the Institutional Review Board of GLRA, India. Willing participants were required to sign an informed consent form. The study was carried out during the period April 2015 –March 2016.
A list of districts in the state of Tamil Nadu, which showed a high number of cases of leprosy were obtained from the DPH&PM,Chennai. The districts of Villupuram, Kancheepuram, Tiruvannamalai and Tiruvallur were some of the districts located relatively close to Chennai that was selected from this list. For the year 2013–2014, a total of 353 new cases of leprosy were reported in the district of Villupuram. Similarly, for the districts of Kancheepuram, Tiruvannamalai and Tiruvallur it was 160, 176 and 140 new cases respectively for the same time period. A list of primary health centers (PHCs) in each of these selected districts was also obtained from the DPH&PM and three PHCs were purposively chosen for the study. The Medical Officers (MOs), Health Inspectors (HIs) and Village Health Nurses (VHNs) in each of these PHCs were then contacted and their permission sought to participate in the study. Each of these cadres of HCPs, play important roles in the management of persons affected by leprosy. Thus, while both the HIs and VHNs assist in case detection in the community and in educating people about leprosy, etc., the MOs are involved in providing direct clinical services to the persons affected by leprosy. We also carried out interviews with persons affected by leprosy in order to understand their experiences about accessing care and any discriminatory behaviors they may have faced from HCPs.
For the qualitative component we included three PHCs from two districts namely Villupuram and Kancheepuram while for questionnaire validation we contacted as many HCPs from the above list who were willing to participate in the study. The scale development process was divided into two parts. In part 1, we carried out the qualitative Semi Structured Interviews (SSIs) and Focus Group Discussions (FGDs) for the purpose of item generation, followed by the scaling exercise and in part 2 we undertook the validation of the questionnaire.
Five MOs and five HIs participated in the qualitative interviews. While three MOs and three HIs were from Villupuram, two MOs and two HIs were from Kancheepuram. One MO and three HIs were women; all the rest were men. The MOs had all completed their MBBS degree. While four of the HIs held post-graduate degrees,one had only completed schooling (12 years). Two FGDs were carried out with the VHNs, one in Kancheepuram and one in Villupuram. The VHNs were all women who had completed their schooling and were trained in community health services.
As far as the MOs were concerned, they regarded leprosy like any other disease and were knowledgeable about its modes of transmission and management. They spoke of setting an example to others by being good role models so that other cadres of HCPs would learn from them and behave accordingly. The VHNs and HIs also had good knowledge about the disease but expressed the need for better understanding of its modes of spread and the side effects of treatment. They mostly spoke of the need to care for these patients and categorically denied any discriminatory behavior by any category of HCPs. Some said that such behavior had been present earlier but was no longer evident while others spoke of harboring some fears in the beginning which they got over once they started caring for patients. All the HCPs however, believed that stigma towards leprosy patients was very much present in the community. This was evident by the fact that patients did not want to be seen taking treatment and tried to hide their condition. They would specifically request the VHN and HIs to keep their disease a secret as they feared being isolated by the community.
The general opinion was that as the disease was identified early both because of better awareness of its symptoms by the community and because of proactive efforts made by VHNs and HIs, the disease rarely progressed to the stage of severe ulceration. Consequently, patients were diagnosed and treated early. Very few patients with severe ulceration and disabilities—which could give rise to both revulsion and fear—were seen by HCPs. A few VHNs opined that sometimes steps taken by them towards protecting others were misinterpreted and led to people affected by leprosy feeling offended and believing that they were being discriminated against.
Several issues emerging from the qualitative interviews were reflected in the items drawn from the scales used in the 11 studies. For example, one MO said that when he initially started working with leprosy patients he feared contracting the disease but over time he overcame the fear, “Earlier I used to fear whether I too would develop leprosy. Now one year, I have been working in leprosy. Now I feel confident. In the beginning, I had this fear but later it was not there” (PHC MO- Kancheepuram). The corresponding attitude item we selected was, ‘I am concerned with getting infection from patients with leprosy when I treat them’. Similarly, an HI indicated that seeing patients present with pus and ulcerative wounds was distressing, “if they are having pus then we would feel like a kind of something. But what I can do about that? I would just tell them how to prevent from getting pus” (PHC HI- Kancheepuram). The corresponding item we selected was, ‘There is a sense of revulsion when seeing a leprosy patient with ulcers’. The VHNs during the FGD said, “As we treat any other patient, we also treat the leprosy patients in the same way. We don’t isolate leprosy patients and treat them separately” (FGD VHN-Kancheepuram). One MO echoed the same issue when he said, “I am seeing the leprosy patient as normal as how I see other patients. We practice precautionary measures, like wearing gloves (PHC MO 3, Villupuram). The corresponding item we selected was. ‘It is possible to manage leprosy like any other disease in the general health service’. An HI described feeling a sense of satisfaction working and helping persons affected by leprosy, “I feel good because when we identify cases and provide them treatment, it gives a good feeling until now, there were many people, who had no knowledge about this disease at all, and they had remained without taking any kind of treatment. So, when we identify cases and put them on treatment, it gives a satisfactory sort of feeling” (HI Kancheepuram). The corresponding item we selected was, ‘I get a sense of satisfaction when I treat patients with leprosy’. Thus the information derived from these qualitative interviews and FGDs served to corroborate the items we selected to constitute this attitude scale. We also took care to bring in both positive and negatively worded statements.
All three persons affected by leprosy said that hospital staff treated them well, were supportive and had guided them to seek care early. They all said that the stigma they faced was mostly from the community and from family members as people feared contracting the disease. They therefore tended to isolate themselves to avoid being hurt and humiliated.
The experts’ group judged the 27 items to measure attitudes towards leprosy at face value thereby establishing face validity. In terms of the content validity, the researchers found that the items in each of the affective, behavioural and cognitive domains were an adequate representation of the construct of attitude. As regards reliability a total of 54 HCPs participated in the reliability exercise (Table 4).
While 54 persons participated in the test, thirty-eight persons (70%) completed the re-test. Unfortunately, despite scheduling a date and time for the re-test in consultation with the HCPs, many of them either did not come or else were pre-occupied and could not fill in the questionnaire. The ICC for test re-test reliability of the 27 items scale was 0.6 (95% CI 0.20–0.78) indicating marginal intra-class correlation while the overall Cronbach’s alpha was 0.85 The alphas for each of the affective and behavioural components was good at 0.78 and 0.69 respectively, but the alpha for the cognitive component was low at 0.53.
Understanding attitudes of health care providers towards persons affected by leprosy is an important step towards their better treatment, improved quality of life and mental health. Hence the development of an appropriate measurement of attitude becomes essential. Our use of the ABC model that served as the conceptual framework in guiding the development of the questionnaire was an added strength as evident by the fact that several scales in the field of mental health, disability and prejudice have been developed based on this framework [28–30]. As per the model, only a proper representation of affect, behavior and cognition best explains the construct of “attitude”. Therefore, a good attitude questionnaire needs to have a healthy mix of items that tap into all these three domains. This scale with a total of 27 items (10-Affect, 8-Behaviour, 9-Cognition) has a good balance of these different domains.
While there are other instruments that have been developed to measure the attitudes of HCPs towards leprosy patients, some had been developed in a totally different cultural context, some were lacking in a comprehensive assessment of the “attitude construct” and others were assessing knowledge rather than attitude. For example, Rao et al assessed knowledge, attitude and practice of MOs towards leprosy using a questionnaire that they had developed [22]. Their 58 items questionnaire had only 6 questions on attitudes with inadequate inclusion of affective, behavioural and cognitive components. Furthermore, the response scale, they used included only ‘yes’ or ‘no’ options. Use of Likert type of response scales on the other hand, would have been more appropriate and would have provided a diverse range of options in keeping with the measurement of attitudes. Attitude response scales are ideally required to indicate the direction (positive or negative) and the intensity (very likely to somewhat likely, for example) of those attitudes [31].
Cultural validation plays an important role in providing accurate estimates of the true attitudes held by individuals in a certain cultural context [32]. This in turn would provide cues as to the nature of remedial measures that would be needed to inculcate a more positive outlook among HCPs towards persons affected by leprosy. Content validity of the scale was enhanced through a process of triangulation, which included reviews of existing scales that measured attitudes of HCPs towards various disabilities and qualitative interviews with different categories of HCPs. The qualitative interviews helped in understanding provider perceptions with regard to treating leprosy patients in the Indian care setting and informed the framing of items that were included in the scale.
Thurstone scaling technique which is based on the consensus scale approach was used for selecting items [26]. Having a panel of experts evaluate each item in terms of whether it is relevant to the topic area and unambiguous in implication helped to enhance the comprehensiveness of the scale. It has been argued that the values assigned to various statements by experts may in effect reflect their own attitudes, and therefore is not entirely objective. We attempted to overcome this by including experts from diverse fields, namely, community medicine, psychology, psychiatry and social science who brought in diverse perceptions based on their respective experiences.
Stigma results in delayed diagnosis and might have an effect on health seeking behavior [33]. Inequalities due to the social, cultural, and economic context in these vulnerable segments act as barriers to access health services [34]. Hence, it is important to be able to assess this type of stigma for the prevention and management of NTDs like leprosy. Assessing stigma is not an easy task. There are several instruments available, but these were developed with different aims or tested in different settings [35–37]. Where negative attitudes prevail, the importance of carrying out interventions that will educate and sensitize HCPs against the harms that such attitudes can cause could be one strategy that could be implemented [38]. Hovland et al. have shown that when there is recurrent persuasive communication, it could result in change of attitudes and opinions. Further the study has shown that the characteristics of the person who receives the communication, the person who delivers the communication, the characteristic of the messages, source of the messages etc. are factors that influence attitude change [39]. Integration of leprosy programmes into general health care, Information Education & Communication programmes and socio-economic rehabilitation strategies have been found to have some effect in stigma reduction in leprosy [40]. But what needs to be kept in mind is that educating civil society needs to be a continuous process. Simply running a few education sessions tends to have a low success rate. Interventions like social marketing have been proved by Brown et al, to be an effective co-adjunct to reducing leprosy related stigma [41]. In Indonesia, contact intervention was found effective in increasing knowledge about leprosy and the negative attitudes reduced. [42]. These interventions target the health system, and community, including persons affected by leprosy and thus provide evidence of strategies that could be tested in the Indian context.
In terms of limitations, the test re-test reliability of the instrument was only marginal. The test was carried out during the monthly review meetings when all the different cadres of HCPs usually gathered at the block PHC. A total of 54 HCPs thus completed the test. The re-test had to be done between 12–14 days of the test and many HCPs who had participated in the test were otherwise preoccupied and indicated that they lacked the time to complete the scale again. The inadequate sample size could have contributed to the marginal test re-test reliability estimates. The overall Cronbach’s alpha for the scale was very good at 0.85 with both the affective and behavioural components showing good internal consistency. However, the alpha for the cognitive component was poor at 0.53. Tavakol and Dennick (2011) have described that one of the reasons for poor alphas in a scale can be attributable to poor inter relatedness between items. Future research will need to re-examine the items under the cognitive component and make suitable modifications such that it is more homogenous [43].
In terms of future work, administering the scale to a larger number of HCPs working in primary care settings will be helpful. One other limitation was our inability to conduct a factor analysis because of the small sample size. A larger sample would have allowed us to discover the underlying factor structure of the questionnaire and determine whether it indeed conforms to the domains we developed based on the ABC model. Factor analysis would also be an effective method for further scale reduction, thereby making it more clinic friendly.
Despite some limitations, this scale developed to measure the attitudes of HCPs towards people affected by leprosy presents a first step in this direction. It could be used to evaluate provider attitudes and could aid in identifying positive or negative attitudes held by providers towards persons affected by leprosy because negative attitudes may impede leprosy control activities. It may also serve as a viable tool to assess changes in the attitudes of HCPs following an intervention. However, further research is required, in terms of using the scale on a much larger sample of HCPs across different parts of India so as to fully substantiate its relevance and cultural appropriateness. With some adaptations, the scales can also be validated for other NTDs which are endemic in India.
|
10.1371/journal.ppat.1000695 | A Multivalent and Cross-Protective Vaccine Strategy against Arenaviruses Associated with Human Disease | Arenaviruses are the causative pathogens of severe hemorrhagic fever and aseptic meningitis in humans, for which no licensed vaccines are currently available. Pathogen heterogeneity within the Arenaviridae family poses a significant challenge for vaccine development. The main hypothesis we tested in the present study was whether it is possible to design a universal vaccine strategy capable of inducing simultaneous HLA-restricted CD8+ T cell responses against 7 pathogenic arenaviruses (including the lymphocytic choriomeningitis, Lassa, Guanarito, Junin, Machupo, Sabia, and Whitewater Arroyo viruses), either through the identification of widely conserved epitopes, or by the identification of a collection of epitopes derived from multiple arenavirus species. By inoculating HLA transgenic mice with a panel of recombinant vaccinia viruses (rVACVs) expressing the different arenavirus proteins, we identified 10 HLA-A02 and 10 HLA-A03-restricted epitopes that are naturally processed in human antigen-presenting cells. For some of these epitopes we were able to demonstrate cross-reactive CD8+ T cell responses, further increasing the coverage afforded by the epitope set against each different arenavirus species. Importantly, we showed that immunization of HLA transgenic mice with an epitope cocktail generated simultaneous CD8+ T cell responses against all 7 arenaviruses, and protected mice against challenge with rVACVs expressing either Old or New World arenavirus glycoproteins. In conclusion, the set of identified epitopes allows broad, non-ethnically biased coverage of all 7 viral species targeted by our studies.
| Arenaviruses cause significant morbidity and mortality worldwide and are also regarded as a potential bioterrorist threat. CD8+ T cells restricted by class I MHC molecules clearly play a protective role in murine models of arenavirus infection, yet little is known about the epitopes recognized in the context of human class I MHC (HLA). Here, we defined 20 CD8+ T cell epitopes restricted by HLA class I molecules, derived from 7 different species of arenaviruses associated with human disease. To accomplish this task, we utilized epitope predictions, in vitro HLA binding assays, and HLA transgenic mice inoculated with recombinant vaccinia viruses (rVACV) expressing arenavirus antigens. Because our analysis targeted two of the most common HLA types worldwide, we project that the CD8+ T cell epitope set provides broad coverage against diverse ethnic groups within the human population. Furthermore, we show that immunization with a cocktail of these epitopes protects HLA transgenic mice from challenge with rVACV expressing antigens from different arenavirus species. Our findings suggest that a cell-mediated vaccine strategy might be able to protect against infection mediated by multiple arenavirus species.
| Pathogen heterogeneity is a commonly encountered challenge for vaccine design. A considerable fraction of unmet vaccine needs for infectious disease is associated with pathogens naturally displaying significant levels of genetic diversity. In particular, RNA viruses pose a substantial challenge because of their propensity for rapid mutation and recombination. A prominent example includes arenaviruses, which are categorized in different complexes with distinct geographical locations, and continue to provide a formidable challenge for vaccine development.
The Arenaviridae family, consisting of a single genus (Arenavirus), contains 22 known viral species, which are classified phylogenetically into Old World and New World complexes [1]. The latter complex is further divided into three Clades (A, B, and C). Old World viruses include Lassa virus (LASV) and the prototypic arenavirus family member, lymphocytic choriomeningitis virus (LCMV). The New World Clade B contains some of the most pathogenic agents, including Guanarito virus (GTOV), Junin virus (JUNV), Machupo virus (MACV), Sabia virus (SABV), and the recombinant Clade A/B arenavirus, Whitewater Arroyo virus (WWAV). Arenaviruses are enveloped viruses, with a ∼10.7 kb RNA genome that encodes four viral proteins, including the glycoprotein precursor (GPC), nucleocapsid protein (NP), RNA-dependent RNA polymerase (L), and the zinc-finger binding protein (Z) [2].
Human infection with arenaviruses typically occurs through direct contact with infected rodents or by inhalation of infectious rodent excretions and secretions, and is associated with dehabilitating and sometimes life-threatening human disease, including central nervous system damage, aseptic meningitis [3], congenital deformities [4],[5], and severe hemorrhagic fever syndrome (reviewed in [6]). Despite the pathogenicity of arenaviruses, there are no licensed vaccines available. Therefore, it is important to develop novel prophylactic vaccine strategies to combat these viruses. CD8+ T cell responses have clearly been associated with reduced pathology and protection against Old World arenavirus infection in both murine models of infection [7],[8],[9],[10] and in humans [11],[12]. Previous reports also suggest a beneficial role for CD8+ T cell-mediated immunity in countering New World arenavirus infections in humans [13],[14]. Thus, vaccination strategies aimed at generating CD8+ T cell responses against both Old and New World arenaviruses should be considered.
The goals of the present study were to identify HLA-restricted CD8+ T cell epitopes from 7 different arenaviruses associated with disease in humans, including GTOV, JUNV, LASV, LCMV, MACV, SABV, and WWAV, and to develop a universal vaccination strategy designed to elicit a T cell-mediated immune response that would provide broad coverage against a variety of arenaviruses and across different ethnic populations. Because of the heterogeneity observed amongst different species of arenaviruses, two non-mutually exclusive concepts were explored. First, we wanted to address whether it was possible to identify CD8+ T cell epitopes that are conserved and/or cross-reactive amongst the different arenavirus species, and second, to determine whether it was possible to combine in the same vaccine, epitopes derived from each of the different arenavirus species, thus providing effective multivalent protection. Here, we report the identification of HLA-restricted CD8+ T cell epitopes that were either cross-reactive or species-specific, and demonstrate that immunization with these epitopes protected HLA transgenic mice from challenge with rVACV expressing antigens from different arenavirus species.
Because the majority of arenaviruses studied require biosafety level-4 (BSL-4) containment, rVACV constructs were designed to express the GPC, NP, L, or Z proteins from the 7 different arenaviruses as a tool to induce and evaluate CD8+ T cell responses. The gene sequences encoding the different arenavirus proteins were derived from prototypic virus strains (Table S1), and the constructs were engineered into the Western Reserve (WR) strain of VACV.
In total, individual rVACV constructs expressing 24 of the 28 arenavirus antigens of interest were generated. Arenavirus protein expression was confirmed for each rVACV through Western blot analysis with infected cell lysates. We verified that arenavirus proteins of the correct size were reproducibly detected from each construct (Figure S1 and data not shown). Although there were different levels of expression for the GPC, L, NP, and Z proteins within the same virus (as shown for GTOV in Figure S1), there was relatively little variation in the level of expression of the same viral protein across the 7 different arenavirus species. Protein expression remained relatively consistent across similar viral proteins as the large majority of arenavirus genes were under the control of a synthetic early/late promoter, PSYN. Overall, these results demonstrate that the 4 viral proteins from different arenavirus species can be ectopically expressed within a rVACV. We did not express WWAV L or Z protein because, at the initiation of this study, sequences for these viral antigens were not available. Technical difficulties also prevented the generation of JUNV Z and SABV L constructs.
To identify candidate arenavirus-derived CD8+ T cell epitopes, we screened the GPC, L, NP, and Z protein sequences from GTOV, JUNV, LASV, LCMV, MACV, SABV, and WWAV using bioinformatic algorithms described elsewhere [15]. The HLA-A02 and HLA-A03 supertype specificities were selected because together they should allow coverage of ∼75% of the human worldwide population [16], and because HLA transgenic mice are available for these two HLA specificities [17],[18]. Because high HLA binding affinity is associated with immunogenicity in vivo [19], peptides with a predicted affinity of ≤100 nM were selected. To limit the number of candidates to a manageable number we chose to study up to 30 different peptides for each HLA/virus/antigen combination. In total, 481 HLA-A02 and 527 HLA-A03 unique candidate peptide sequences, consisting of nonamers and decamers, were selected and synthesized.
Because human PBMC samples from arenavirus-exposed individuals were difficult to obtain, we used HLA transgenic mice to identify human arenavirus epitopes. To determine the in vivo antigenicity of the 481 arenavirus peptides predicted to bind HLA-A02, HLA-A*0201 transgenic mice were infected with rVACV expressing one of the arenavirus proteins (see Materials and Methods for details). Twelve different CD8+ T cell epitopes from 6 arenavirus species were identified utilizing IFN-γ ELISPOT assays: 2 from JUNV (GPC7–15 and GPC18–26), 1 from LASV (GPC111–120), 2 from LCMV (GPC34–43 and Z49–58), 3 from MACV (GPC18–26, NP19–27, and NP432–440), 2 from SABV (GPC142–150 and NP547–556), and 2 from WWAV (GPC42–50 and NP274–282) (Figure 1). The JUNV GPC18–26 epitope is 100% conserved in MACV, and was independently identified as an epitope in both viruses. A CD8+ T cell epitope was not identified from GTOV. None of the epitopes were recognized by CD8+ T cells derived HLA-A*0201 transgenic mice infected with wild type (wt) VACV-WR.
As these 12 HLA-A*0201 epitopes were identified following rVACV infection in transgenic mice, these experiments indicated that they can be generated by processing in mouse antigen-presenting cells (APCs). To assess processing in human APCs, HLA-A*0201 transgenic mice were immunized with the 12 arenavirus epitopes individually, and 11 to 14 days post-immunization, splenic CD8+ T cells were assayed for recognition of human target cells infected with the rVACV expressing the appropriate arenavirus antigen. Recognition of rVACV-infected target cells was robust for 7 of the 12 peptide-primed CD8+ T cells (LASV GPC111–120, LCMV Z49–58, MACV NP19–27, SABV GPC142–150, SABV NP547–556, WWAV GPC42–50, and WWAV NP274–282; Figure 2). Three epitopes (JUNV GPC18–26, MACV GPC18–26, and MACV NP432–440) were weakly recognized in replicate experiments, while for JUNV GPC7–15 and LCMV GPC34–43 processing in human APCs could not be demonstrated. None of the arenavirus peptide-primed CD8+ T cells had a detectable response to target cells infected with wt VACV-WR (data not shown). We also observed minor recognition of 2 epitopes by murine H-2bxd as LASV GPC111–120 and LCMV GPC34–43 were presented by one of the endogenous mouse MHC class I molecules co-expressed in the HLA-A*0201 transgenic mice (data not shown). Table 1 provides an overall summary of the properties of the arenavirus-derived HLA-A*0201-restricted CD8+ T cell epitopes, including which epitopes were endogenously processed in human APCs.
Peptides containing positively charged amino acids at the carboxyl-termini associated with HLA-A03 supertype binding are not effectively processed in mice [20]. To overcome this potential problem, and enable the use of mice expressing the A03 supertype prototype allele HLA-A*1101, the immunogenicity of the 527 predicted HLA-A03 binding peptides was determined by immunization of HLA-A*1101 transgenic mice with pools of the predicted arenavirus peptides. Using this strategy, we identified 165 HLA-A03 supertype peptides that were immunogenic in HLA-A*1101 transgenic mice (data not shown).
To assess processing of the immunogenic HLA-A*1101 peptides, HLA-A*1101 transgenic mice were immunized with individual arenavirus peptides (see Materials and Methods for details). At 11 to 14 days post-immunization, splenic CD8+ T cells were screened for recognition of HLA-A*1101-expressing human target cells that had been infected with an appropriate rVACV. It was found that 14 of the 165 HLA-A*1101 peptides tested were endogenously processed from native arenavirus antigens by human APCs and recognized by arenavirus peptide-primed CD8+ T cells. Of these peptides, 4 were overlapping with the 14 processed epitopes and had similar peptide-specific CD8+ T cell responses. Thus, in total, 10 unique epitopes were endogenously processed by human APCs. These peptides represented all 4 viral antigens and 5 different arenaviruses: 1 from GTOV (L1977–1985), 3 from LCMV (GPC46–55, GPC112–120 and Z24–33), 2 from MACV (NP82–90 and Z27–36), 3 from SABV (GPC90–98, NP82–90 and Z64–72), and 1 from WWAV (NP439–447) (Figure 3). A homologous epitope (NP82–90) was identified in both MACV and SABV. None of the arenavirus peptide-primed CD8+ T cells had a detectable response to target cells infected with wt VACV-WR (data not shown). In addition, a single determinant (LCMV GPC46–55) was immunogenic when presented by either HLA-A*1101-expressing or H-2bxd-expressing APCs (data not shown). An overall summary of the properties of the arenavirus-derived HLA-A*1101-restricted CD8+ T cell epitopes is provided in Table 2.
The identified epitopes were 89 to 100% conserved within a given arenavirus species (Tables 1 and 2). Next, we wanted to determine whether the CD8+ T cell responses generated by the identified HLA-restricted epitopes could cross-reactively recognize orthologous sequences derived from different arenavirus species. Such cross-recognition might further increase the coverage afforded by the epitope set against each different arenavirus species.
To identify cross-reactive peptides, we compared the primary amino acid sequence of the 12 HLA-A02 epitopes, and the 10 HLA-A03 epitopes identified herein, as well as the 3 LASV (GPC42–50, GPC60–68, and GPC441–449) and 3 LCMV (GPC10–18, GPC447–455, and NP69–77) HLA-A*0201-restricted epitopes identified in previous studies [9],[10] to the orthologous regions within the other arenavirus protein sequences used to generate the rVACV constructs. Peptides containing 2 or more identical amino acids as the epitope sequence and found in ≥20% of the isolates in the arenavirus protein sequence database (http://epitope.liai.org:8080/projects/arena) were chosen for further analysis. Using this approach, 144 candidate cross-reactive peptides were selected.
To test cross-reactivity, HLA transgenic mice were immunized with individual arenavirus epitopes we had identified. Splenic CD8+ T cells were screened for IFN-γ secretion in response to APCs pulsed with either the immunizing epitope or the potentially cross-reactive arenavirus peptides. Immunization with 5 HLA-A*0201-restricted epitopes (JUNV GPC18–26, MACV GPC18–26, LCMV GPC447–455, MACV NP19–27, and MACV NP432–440) and 2 HLA-A*1101-restricted epitopes (LCMV GPC46–55 and MACV NP82–90) generated one or more cross-reactive CD8+ T cell response(s). It should be noted that the JUNV GPC18–26 and MACV GPC18–26 epitopes have an identical amino acid sequence. The remaining 19 HLA-restricted arenavirus epitopes did not induce any detectable cross-reactive CD8+ T cell responses. Table 3 summarizes the results of these experiments.
The amino acid sequence differences between the epitope sequence and the cross-reactive peptides mainly represented conservative or semi-conservative amino acid substitutions occurring at either the position 2 primary MHC interacting anchor residue or the prominent secondary anchor residues at positions 1, 3, and 7 (Table 3). Thus, in general, the residues involved in TCR interaction were preserved in the cross-reactive peptides. We also observed that non-cross-reactive peptides contained 3 or more amino acid differences with the epitope sequence. As detailed in Table 3, in terms of response magnitude, CD8+ T cell responses to the cross-reactive peptides ranged from 20–100% of the epitope-specific response.
In most cases, the cross-reactivity detected was limited to New World arenaviruses, and in some cases, remained detectable even with low doses (0.1 µg/ml) of peptide (Figure 4). However, the LCMV GPC447–455 cross-reactivity spanned both Old (LASV) and New (WWAV) World arenaviruses (data not shown). A cross-reactive T cell response induced by LCMV GPC447–455 peptide immunization has also been reported in an independent study (J. Botten and M. J. Buchmeier, unpublished work). A total of 11 cross-reactive peptides, including 3 previously defined epitopes and 8 newly identified peptides, were recognized in repeat experiments by HLA-restricted epitope-specific CD8+ T cells. Summed together, 36 arenavirus-derived peptides, which include 28 epitopes and 8 cross-reactive peptides, generate arenavirus-specific CD8+ T cell responses that recognize the 7 different arenavirus species (Table 4). Thus, cross-reactive recognition of peptides significantly increases epitope coverage of the various HLA-virus combinations.
As expected, the majority of peptides bound to their restricting allele with high affinity (IC50≤500 nM; Tables 1 and 2). Thus, our predictive approach effectively identified high affinity HLA binding peptides. As HLA-A02 and A03 supertypes are composed of a group of alleles that share largely overlapping peptide binding specificities [21], we subsequently tested the binding affinity of the identified arenavirus determinants to the remaining HLA molecules found within each supertype. All of the HLA-A*0201 determinants, except LASV GPC111–120, bound 3 or more additional HLA-A02 supertype molecules (HLA-A*0202, -A*0203, -A*0206, and -A*6802), and all HLA-A*1101 determinants bound 3 or more additional HLA-A03 supertype molecules (HLA-A*0301, -A*3001, -A*3101, -A*3301, and -A*6801) (data not shown).
Based on these results, we calculated the theoretical population coverage afforded by the epitopes and cross-reactive peptides for each of the 7 arenaviruses (using the Population Coverage Calculation Tool available through the Immune Epitope Database and Analysis Resource [22]). These calculations were based on peptide binding data (shown in Tables 1, 2, and 3 and data not shown) and the reported frequencies of each HLA allele in different ethnic populations. For the present analysis, biologically relevant binding was defined as an IC50≤500 nM.
The total coverage of the general population provided by the corresponding set of epitopes for each of the 7 different arenaviruses is shown in Figure 5A. The epitopes identified from LCMV and SABV provided the broadest population coverage of 69.9% of the overall population, while JUNV epitopes provided the least amount of coverage of 36.3%. The remaining viruses, GTOV, LASV, MACV, and WWAV, provided 59.8%, 43.6%, 67.1%, and 62.7% population coverage, respectively. When averaged over the 7 different viruses, the defined epitopes provided population coverage of 58.5% of the entire population. In most cases, coverage entails recognition of multiple HLA-epitope combinations. For example, from Figure 5A, it can be seen that in an average population, about 35% of individuals can be expected to recognize 3 or more epitope-HLA combinations.
Coverage was fairly balanced throughout the major ethnic groups. A representative graph (Figure 5B) shows the coverage afforded by the MACV peptides across several major population groups. As shown, the MACV epitopes provide coverage ranging from about 50.0% of Australians, to 88.5% for South Americans, with an average coverage across all populations of 67.1%. These results indicate that the set of arenavirus epitopes described in the present study can provide broad coverage to the different viruses in different ethnic populations.
Lastly, we examined whether priming of HLA-A*0201 transgenic mice with a pool of the HLA-A02 arenavirus peptides would confer protection against subsequent viral challenge. Because challenge with 6 of the 7 different arenavirus species considered would require BSL-4 containment, which was unavailable to us, we utilized challenge with rVACVs expressing the various arenavirus antigens as an alternative surrogate system.
Groups of HLA-A*0201 transgenic mice were immunized with a pool of 14 CD8+ T cell peptides plus a single T helper cell epitope unrelated to arenaviruses, or mock-immunized (see Materials and Methods for details). To ensure protection was HLA-restricted, arenavirus peptides that were also restricted by murine MHC (LASV111–120 and LCMV GPC34–43) and not endogenously processed in human APCs (JUNV GPC7–15) were excluded. Peptide-immunized mice were challenged with a rVACV construct that expressed either LCMV GPC, LASV GPC, or SABV GPC. On day 5 post-challenge, we observed that peptide pool-immunized mice had significantly reduced mean viral titers following challenge compared to the mock controls (Figure 6A) with rVACV-LCMV GPC (2.0 log10 reduction; P = 0.0031), rVACV-LASV GPC (2.3 log10 reduction; P = 0.0398), and rVACV-SABV GPC (3.5 log10 reduction; P = 0.0498).
Although the results obtained from mice challenged with rVACV-LCMV GPC were quite consistent, we observed varying viral titers in both mock and peptide-immunized mice challenged with either rVACV-LASV GPC or rVACV-SABV GPC. This could be due to technical failure in delivering the proper infectious dose of these rVACV constructs during the i.p. inoculation. Despite this variability, there still remains a significant difference (P<0.05) between viral titers of the mock and peptide-immunized mice that were challenged with rVACVs expressing either LASV GPC or SABV GPC. We also found that the 3 different groups of mock-immunized mice had varying viral titer levels. All rVACVs utilized are capable of high titer replication both in vitro and in vivo. We have, however, observed that certain rVACVs (i.e. rVACV-SABV GPC) grow more efficiently compared to others (i.e. rVACV-LCMV GPC). This might help explain the differences observed in viral titers in mock-immunized mice challenged with the 3 rVACVs (Figure 6A).
To determine whether an expansion of epitope-specific CD8+ T cells after rVACV challenge would correlate with viral titer reduction, the frequency of epitope-specific CD8+ T cells following peptide pool immunization before and after rVACV challenge was measured. At 18 days post-peptide pool immunization, HLA-A*0201 transgenic mice that were not challenged displayed significant numbers of CD8+ T cells to all immunizing epitopes (Figure 6B). However, mice that were peptide pool immunized and challenged with an rVACV-GPC demonstrated a clear expansion of CD8+ T cells specific for the challenge virus. In some instances, cross-reactive CD8+ T cell expansion was also observed. For example, LASV GPC441–449-specific CD8+ T cells expanded following challenge with rVACV-LCMV GPC. Thus, these results demonstrate that CD8+ T cell-mediated immunity can protect against challenge with rVACVs expressing Old and New World arenavirus antigens in an epitope-specific manner.
Hemorrhagic fever and aseptic meningitis caused by arenavirus infection represent serious human public health problems that lead to dehabilitating and sometimes fatal disease. Furthermore, arenaviruses are regarded as a potential bioterrorism threat, and as such are classified as Class A pathogens. Development of a vaccine against arenavirus infection needs to address the genetic diversity observed within and between different viral species of the Arenaviridae family. Indeed, pathogen heterogeneity is also a prominent consideration among other families of RNA viruses. For instance, the genetic variability found amongst HIV clades circulating worldwide, and the high mutation rate of HIV that allows for evasion of the adaptive immune response, continue to provide a daunting challenge for vaccine development [23]. Likewise, HCV vaccine design faces the obstacle of different genotypes that are found with different prevalence in distinct locations [24]. The four dengue virus serotypes provide a unique challenge, as heterologous re-infection is associated with dengue hemorrhagic fever and dengue shock syndrome [25],[26]. The variability associated with influenza virus is reflected in different strains and subtypes, forcing the development of yearly updated vaccines, and is also the cause of concern in the context of new influenza pandemics [27],[28].
As it is unlikely that separate vaccines for each viral species within a family will be developed, the two most viable, and not necessarily exclusive, approaches might be the development of either a multivalent and/or a cross-protective vaccine. Herein, we performed proof of concept studies to evaluate these approaches, utilizing arenaviruses as a model system. Our results indicate that multiple conserved (89 to 100%) epitopes can be defined from each arenavirus. While broadly cross-reactive epitopes between arenavirus species were not identified, occasional cross-reactivites were demonstrated, leading to an increased viral coverage by the defined epitope set. Notably, it was possible to induce simultaneous responses against all epitopes by a peptide pool vaccination, demonstrating the feasibility of a multivalent vaccination. This shows that a single vaccination strategy with both multivalent and cross-protective CD8+ T cell epitopes was able to engender protection from recombinant viruses expressing antigens derived from a subset of different arenavirus species. Because our study was focused on demonstrating universal coverage against multiple arenavirus species, we chose to conduct protection studies utilizing rVACVs expressing antigens from 3 representative arenavirus species. Protection from virus challenge was demonstrated through reduced viral titers in peptide-immunized animals instead of survival, as the rVACVs used in this study are attenuated viruses. Previous studies utilizing animal models of HIV and influenza infection have demonstrated that a reduction in viremia correlates with survival [29],[30]. Importantly, we have also shown in an independent study that immunization with LCMV GPC447–455 protected HLA-A*0201 transgenic mice against lethal intracranial challenge with LCMV [31].
To identify HLA-restricted CD8+ T cell epitopes derived from pathogenic arenaviruses, certain technical challenges needed to be circumvented, including the need for BSL-4 containment for most species considered, and the lack of availability of samples from exposed human donors. We bypassed the requirement for BSL-4 containment by developing a panel of 24 rVACV vectors that expressed the different arenavirus antigens of interest. To our knowledge, this is the first instance in which this approach has been implemented on a large scale to study T cell responses directed against human pathogens. In addition, utilizing HLA transgenic mice enabled the identification of epitopes relevant for humans. HLA transgenic mice have been useful in identifying human T cell epitopes from small viruses, such as influenza [32], and more complex pathogens, such as VACV [33]. In the future, we plan to examine CD8+ T cell responses in arenavirus-immune human donors in order to assess the degree of overlap between arenavirus-specific responses recognized in HLA transgenic mice and humans.
In total, 16 HLA-A*0201, and 10 HLA-A*1101, arenavirus-specific CD8+ T cell epitopes that are naturally processed by human APCs have been identified so far. These numbers are similar to those found in other virus infection models. For example, we previously identified 14 HLA-A*0201, 4 HLA-A*1101, and 3 HLA-B*0702-restricted VACV-specific CD8+ T cell epitopes using HLA transgenic mice [34]. In additional studies in H-2b mice, 19 H-2Kb and 9 H-2Db-restricted LCMV-specific epitopes were identified [8]. In the VACV system, 27 H-2Kb and 22 H-Db-restricted epitopes were defined [35]. It is important to point out that our current study was not designed to identify the totality of the response, but rather focused on epitopes relatively well conserved within a given arenavirus species.
Widely conserved CD8+ T cell epitopes among the 7 different arenaviruses species were not identified. Given that amino acid sequence identities of homologous proteins of the 7 arenavirus species range from 44 to 63% [36], it is not surprising that only a few conserved epitopes were identified. However, we found that immunization of HLA transgenic mice with 7 of the arenavirus peptides induced cross-reactive CD8+ T cell responses, which significantly increased the coverage afforded by the epitope set against different arenavirus species. Importantly, T cell cross-reactivity might also extend to newly identified arenaviruses. To examine this further, we compared the amino acid sequence of the epitopes defined here to the orthologous regions within Chapare [37] and Lujo [38] viruses. We found that 6 of the identified epitopes had ≥70% amino acid sequence identity to the Chapare and Lujo viruses. Thus, cross-reactive T cells might also extend to peptides derived from newly discovered arenaviruses.
Previous studies have demonstrated T cell cross-reactivity for both Old and New World arenaviruses. It was shown in a guinea pig model that adoptive transfer of LCMV or Moepia virus (MOPV)-immune CD8+ T cells could confer protection against challenge with the highly virulent LASV [39]. Likewise, in a murine model, cell transfer of LCMV-immune CD4+ and CD8+ T cells protected mice from Pichinde virus infection [40]. Furthermore, in humans, T cell clones from patients that have recovered from acute LASV infection showed cross-reactive recognition of MOPV peptides [41]. Inter-clade cross-reactivity has also been demonstrated for HIV-specific CD8+ T cell responses and is considered of particular relevance in vaccines that are to be used in geographically and genetically distinct HIV epidemics [42],[43],[44].
We demonstrated that the arenavirus epitope set afforded redundant coverage of the different viruses by epitopes of different restriction, as an average about 50% of individuals are predicted to recognize more than one MHC-peptide combinations. Coverage of approximately 60% of the general population is projected on the basis of HLA frequencies and binding data. As only 2 HLA supertypes were investigated, it should be possible to achieve essentially 100% coverage by identifying epitopes restricted by additional supertypes, such as A01, A24, B07, and B44 [16],[21].
Herein, we implemented a multivalent vaccine strategy to protect against rVACVs expressing antigens from different arenavirus species. This approach has proven itself in the case of the currently licensed vaccine against HPV, which contains capsid proteins derived from 4 different serotypes prominently associated with disease [45]. We demonstrated that immunization of HLA-A*0201 transgenic mice with a cocktail containing 14 arenavirus-derived peptides and a T helper cell epitope resulted in a detectable CD8+ T cell response to each of the viral epitopes. The magnitude of epitope-specific CD8+ T cell responses following peptide pool immunization was diminished compared to the responses observed after individual peptide immunization (see Figures 2 and 6B for comparison). Competition for HLA-A*0201 binding might play a role in determining the response magnitude of epitopes when immunized as a pool. However, the magnitude of the CD8+ T cell responses was sufficient to significantly reduce viral titers following challenge with rVACVs expressing both Old and New World arenavirus antigens.
Furthermore, comparing our data to a previous study, we found that the coverage achieved with the breadth of epitopes does not compromise immunity to a single epitope. HLA-A*0201 transgenic mice immunized with the pool of the 14 arenavirus-specific determinants and challenged with rVACV-LASV GPC demonstrated a 2.3-log reduction in viral titers compared to the mock control. Botten et al. demonstrated that single peptide immunization of HLA-A*0201 transgenic mice with either LASV GPC42–50 or GPC60–68, two epitopes included in our peptide pool immunization, followed by challenge with rVACV-LASV GPC resulted in a 2.8 and 2.2-log reduction in viral titers, respectively [9]. Thus, viral titer reduction is very similar whether mice are immunized with a single epitope or an epitope cocktail.
None of the defined arenavirus epitopes demonstrated cross-reactivity with VACV-WR, strongly suggesting that the reduction in viral titers in peptide-immunized mice is mediated by arenavirus-specific CD8+ T cell responses (see Figure 1 for representative data). To provide further evidence for arenavirus epitope-specific protection, we observed a substantial expansion of CD8+ T cell responses specific for the challenge virus (relative to unchallenged mice). Furthermore, Botten et al. demonstrated that immunization of HLA-A*0201 transgenic mice with either LASV GPC42–50 or GPC60–68 did not result in protection against challenge with the irrelevant rVACV-LASV NP construct, confirming the protection following rVACV-LASV GPC challenge was LASV epitope-specific [9].
We also found cross-reactive CD8+ T cell responses were significantly boosted following rVACV challenge of peptide pool-immunized mice. As expected, the LCMV GPC447–455-specific CD8+ T cell response was boosted following rVACV-LASV GPC challenge, while challenge with rVACV-LCMV GPC boosted the CD8+ T cell response against LASV GPC441–449. The LCMV GPC10–18-specific CD8+ T cell response expanded approximately 40-fold following rVACV-LASV GPC challenge, while only 3-fold after challenge with rVACV-LCMV GPC. One possible explanation for this discrepancy is that the expanded CD8+ T cell subset generated following rVACV-LASV GPC challenge has a higher avidity for the LCMV peptide. Finally, mice immunized with the peptide pool and challenged with the rVACV expressing SABV GPC showed a significant expansion of CD8+ T cells with the SABV NP547–556 peptide. It is plausible that infection with rVACV-SABV GPC led to the generation of a cross-reactive CD8+ T cell subset that was capable of recognizing the SABV NP547–556 peptide.
In conclusion, our studies suggest that simultaneous induction of a multivalent and cross-protective CD8+ T cell response is a feasible approach to vaccination against multiple arenavirus species. As a proof of concept, protection studies with rVACVs expressing the GPC from LCMV, LASV, and SABV were performed. Additional studies are still required to investigate whether peptide pool immunization would reduce viral titers in transgenic mice challenged with rVACVs expressing other arenavirus GPCs (GTOV, JUNV, MACV, and WWAV), or other arenavirus proteins (NP and/or Z). The identification of epitopes restricted by additional HLA types, their validation in humans exposed to arenavirus infections, and their formulation in a multivalent construct would also be the logical next steps in the further exploration of this concept.
All mouse studies followed guidelines set by the National Institutes of Health and the Institutional Animal Care and Use Committee-approved animal protocols (Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC) and Office of Laboratory Animal Welfare (OLAW)).
The arenavirus open reading frames (ORFs) utilized in this study were a total of 333 unique sequences from one or more isolates of GTOV, JUNV, LASV, LCMV, MACV, SABV, and WWAV taken from the arenavirus protein sequence database (http://epitope.liai.org:8080/projects/arena) [36]. Candidate HLA-A02 and A03 supertype epitopes were identified using a previously described algorithm [15]. Peptides with a predicted IC50≤100 nM were selected from each ORF, and then ranked based on inter-virus conservancy (% identity matches). The top ranked peptides (up to 30) per supertype per virus per antigen were selected.
Peptides were synthesized as crude material by Pepscan Systems (Lelystad, The Netherlands). Candidate epitopes were resynthesized by A and A Labs (San Diego, CA) and purified to 95% or greater homogeneity by reverse-phase HPLC. Hepatitis C virus core 132 (DLMGYIPLV), and human MAGE 369 (SSLPTTMNY) were used as control HLA-A*0201 and HLA-A*1101-restricted peptides, respectively. IEDB submission identification numbers for HLA-A*0201 and HLA-A*1101-restricted epitopes are 1000381 and 1000374, respectively.
Quantitative assays to measure the binding affinity of peptides to purified HLA-A02 (A*0201, A*0202, A*0203, A*0206, A*6802) and HLA-A03 (A*1101, A*0301, A*3001, A*3101, A*3301, A*6801) supertype molecules were based on the inhibition of binding of a radiolabeled standard peptide, and were performed essentially as described elsewhere [46],[47],[48]. Briefly, after a 2-day incubation, binding of the radiolabeled peptide to the corresponding MHC class I molecule was determined by capturing MHC/peptide complexes on Greiner Lumitrac 600 microplates (Greiner Bio-One, Monroe, NC) coated with the W6/32 Ab (anti-HLA class I), and measuring bound cpm using the Topcount microscintillation counter (Packard Instrument). The concentration of peptide yielding 50% inhibition of the binding of the radiolabeled probe peptide (IC50) was then calculated. Peptides were typically tested at six different concentrations covering a 100,000-fold dose range, and in three or more independent assays. Under the conditions utilized, where [label] < [MHC] and IC50 ≥ [MHC], the measured IC50 values are reasonable approximations of the KD values.
HLA-A*0201/Kb and HLA-A*1101/Kb (referred to as HLA-A*0201 and HLA-A*1101) transgenic mice were bred and maintained in the animal facilities at the La Jolla Institute for Allergy and Immunology (La Jolla, CA). These mice were the F1 generation derived from a cross between BALB/c mice (The Jackson Laboratory, Bar Harbor, ME) and HLA transgenic mice (expressing a chimeric gene consisting of the α1 and α2 domains of the human HLA and the α3 domains of murine H-2Kb) created on the C57BL/6J background [17],[18]. CB6F1/J mice were the F1 generation derived from a cross between C57BL/6J and BALB/c mice (The Jackson Laboratory). C57BL/6J mice deficient in Kb and Db (KbDb−/−) mice were purchased from Taconic Farms (Hudson, NY).
Target cells used in the IFN-γ ELISPOT assays were the human Jurkat cells (JA2.1) that express the HLA-A*0201/Kb chimeric gene [17], human EBV B cells (BVR) that express HLA-A*1101, and LPS-stimulated B lymphoblasts prepared as previously described [34] from HLA-A*0201 and HLA-A*1101 transgenic mice, and CB6F1/J (H-2bxd) mice, which do not express the HLA transgene. The target cells were either pulsed for 1 h with peptide (10, 1, or 0.1 µg/ml), or infected with 107 PFU rVACV, or 2×106 PFU VACV-WR 24 h prior to the ELISPOT assay. CV-1 (CCL-70), Vero E6 (CRL-1586), BHK-21 (CCL-10), and BSC-40 (CRL-2761) cells were purchased from ATCC and grown as recommended by the supplier.
The rVACV-LCMV GPC and NP (derived from LCMV Armstrong Clone 53b) constructs were engineered using the WR strain of VACV as previously described [49]. LCMV GPC and NP gene expression were under the control of the VACV P7.5K early/late promoter [49]. The remaining rVACV arenavirus constructs were also generated on the WR background using ORF sequences from prototypic arenavirus strains (as outlined in Table S1), according to the protocols established by Blasco and Moss [50]. Arenavirus target gene expression was under the control of a synthetic early/late promoter, PSYN [50]. In brief, the arenavirus GPC, NP, L, and Z genes were subcloned individually into the pRB21 transfer vector. CV-1 cells were then infected with the VACV strain vRB12 at an MOI of 2, followed by transfection with 10 µg of the transfer vector containing a single arenavirus ORF. Viruses that underwent homologous recombination with the transfer vector were selected for their plaque-forming ability. Arenavirus protein expression was confirmed for each rVACV through immunoflourescence assay to verify that all plaques expressed the correct antigen, as well as by Western blot analysis to verify the correct processing and size of each arenavirus protein. Virus titers were determined through plaque assays on BSC-40 cells.
For epitope identification studies, HLA-A*0201 and HLA-A*1101 transgenic mice were peptide-immunized s.c. at the base of the tail with a pool of 8 to 11 peptides (15 µg/peptide) or 50 µg of a single peptide, along with 140 µg of the helper I-Ab-restricted epitope hepatitis B virus (HBV) core128–140 (TPPAYRPPNAPIL) in PBS emulsified in IFA. In general, effector CD8+ responses were analyzed in the spleens of mice 11 to 14 days after peptide immunization. One exception was the assessment of processing of the HLA-A*1101 peptides. In this case, effector CD8+ responses were analyzed by utilizing splenocytes from single peptide-immunized mice that were cultured in vitro for an additional 6 days with the immunizing peptide. For VACV infection, HLA-A*0201 transgenic mice were injected i.p. with either 2×106 PFU VACV-WR strain as a control, or 107 PFU rVACV containing one of the arenavirus genes. A higher dose of 107 PFU rVACVs was used because these viruses are attenuated, while wt VACV-WR is not. The infectious doses of 2×106 PFU VACV-WR and 107 PFU rVACV have been routinely used in previous studies [9],[31],[34]. On day 7 post-infection, mice were sacrificed and purified splenic CD8+ T cells were analyzed by ex vivo ELISPOT assays for IFN-γ.
For challenge studies, HLA-A*0201 transgenic mice immunized with a pool of 15 peptides, including 14 arenavirus epitopes (50 µg/peptide) and 140 µg of the HBV core128–140 helper epitope, in PBS emulsified in IFA, or mock-immunized with an equivalent volume of DMSO as the peptide-immunized group as well as 140 µg of the helper epitope in PBS emulsified in IFA. Mice were challenged 13 days later through i.p. inoculation with 107 PFU rVACV-LCMV GPC, rVACV-LASV GPC, or rVACV-SABV GPC. Five days after virus challenge, spleens were harvested for CD8+ T cells and ovaries were harvested for rVACV titer determination. To quantitate viral titers, ovaries were homogenized and sonicated prior to plating serial 10-fold dilutions of the homogenates on BSC-40 cells.
The mouse IFN-γ ELISPOT assay was performed as previously described [51]. In brief, 4×105 splenocytes or 2×105 splenic CD8+ T cells (purified by anti-CD8 magnetic beads [Miltenyi Biotec, Auburn, CA]) were cultured with either 104 or 105 peptide-pulsed or rVACV-infected target cells. For peptide pulsing, target cells were incubated with peptide for at least 1 h at 37°C, followed by 3 washes to remove excess peptide. For rVACV infection, JA2.1 or BVR cells were infected at a MOI of 10 with a rVACV 18 to 20 h prior to the assay. Each assay was performed in triplicate wells. After a 20 h incubation at 37°C, plates were developed, and responses calculated as described [34]. Criteria for positivity were net spot-forming cells (SFC)/106 cells ≥20, stimulation index (SI) ≥1.4 or 2.0, and p-value ≤0.05 using a Student's t test in at least 2 out of 3 experiments.
|
10.1371/journal.ppat.1003923 | Lundep, a Sand Fly Salivary Endonuclease Increases Leishmania Parasite Survival in Neutrophils and Inhibits XIIa Contact Activation in Human Plasma | Neutrophils are the host's first line of defense against infections, and their extracellular traps (NET) were recently shown to kill Leishmania parasites. Here we report a NET-destroying molecule (Lundep) from the salivary glands of Lutzomyia longipalpis. Previous analysis of the sialotranscriptome of Lu. longipalpis showed the potential presence of an endonuclease. Indeed, not only was the cloned cDNA (Lundep) shown to encode a highly active ss- and dsDNAse, but also the same activity was demonstrated to be secreted by salivary glands of female Lu. longipalpis. Lundep hydrolyzes both ss- and dsDNA with little sequence specificity with a calculated DNase activity of 300000 Kunitz units per mg of protein. Disruption of PMA (phorbol 12 myristate 13 acetate)- or parasite-induced NETs by treatment with recombinant Lundep or salivary gland homogenates increases parasite survival in neutrophils. Furthermore, co-injection of recombinant Lundep with metacyclic promastigotes significantly exacerbates Leishmania infection in mice when compared with PBS alone or inactive (mutagenized) Lundep. We hypothesize that Lundep helps the parasite to establish an infection by allowing it to escape from the leishmanicidal activity of NETs early after inoculation. Lundep may also assist blood meal intake by lowering the local viscosity caused by the release of host DNA and as an anticoagulant by inhibiting the intrinsic pathway of coagulation.
| Salivary components from disease vectors help the arthropod to acquire blood. Here we show that an arthropod vector salivary enzyme affects the innate immune system of the host—mainly the destruction of neutrophil traps—allowing the Leishmania parasite to evade the host immune response and to cause an infection. This work highlights the relevance of vector salivary components in parasite transmission and further suggests the inclusion of these proteins as components for an anti-Leishmania vaccine. Importantly, because salivary proteins are always present at the site of natural transmission, this work further encourages the testing of vaccine candidates using the natural route of transmission—the bites of an arthropod vector—instead of current practices based solely on needle injection of parasites.
| Leishmaniasis comprises human and animal diseases caused by parasites of the genus Leishmania that are transmitted by the bite of infected sand flies [1]. Leishmania transmission occurs when an infected sand fly probes the host's skin in search of a blood meal. During probing and feeding, sand flies salivate into the host's skin. Saliva contains powerful pharmacologic components that mediate blood-feeding success and facilitate Leishmania infection, first shown when Lutzomyia longipalpis salivary glands (SGs) were reported to enhance Leishmania major infection in mice [2], [3]. In the last two decades, SG and recombinant salivary proteins were investigated for their effect in enhancing pathogen transmission in different model systems (reviewed in [4]). The powerful Lu. longipalpis vasodilator maxadilan along with hyaluronidase were shown to facilitate transmission and establishment of L. major parasites [5], [6]; however, as we show here, these salivary compounds are not the only active components of sand fly saliva that exacerbate parasite infection.
Neutrophils are considered the host's first line of defense against infections and have been implicated in the immunopathogenesis of leishmaniasis [7]–[10]. Leishmania parasites evade killing by neutrophils by blocking oxidative burst and entering a nonlytic compartment unable to fuse with lysosomes or by resisting the microbicidal activity of neutrophil extracellular traps (NETs) [11]. The mechanism of NET formation (NETosis) in response to Leishmania sp. is still under investigation [11], [12]; however, recent studies have shown the direct effect of Lu. longipalpis SG extract (SGE) in L. major parasite survival inside host neutrophils [13]. This effect was abrogated by pretreatment of SGE with proteases as well as preincubation with antisaliva antibodies, supporting the hypothesis that Lu. longipalpis salivary protein(s) help Leishmania survival inside neutrophils. The negative effect of NETosis to Leishmania was recently documented [12].
In this report, we present direct experimental evidence that Lundep (Lutzomyia NET destroying protein), a secreted salivary endonuclease, is responsible for the NET-destroying activity of Lu. longipalpis and that this activity enhances parasite infectivity both in vitro and in vivo. Furthermore, Lundep may assist blood-meal intake by lowering the local viscosity caused by the release of host DNA and as an anticoagulant and anti-inflammatory by inhibiting the intrinsic pathway of coagulation.
Bioinformatic survey of the Lu. longipalpis sialotranscriptome [14] identified a transcript (AY455916; Lundep) containing the NUC-motif (prokaryotic and eukaryotic double (ds) and single (ss) stranded DNA and RNA endonucleases also present in phosphodiesterases) indicative of nonspecific DNA/RNA endonuclease. Alignment of the Lundep putative active center with other proteins of the same family shows the presence of the conserved RGH triad found in most DNases characterized so far (Figure S1A). The importance of these residues for catalysis has been previously studied in detail [15], [16]. Putative endonucleases retrieved using Lundep as query in the NCBI database, grouped into well supported clade, indicating that they are orthologs (Figure S1B). Visual inspection of sand fly sequences revealed the presence of signal peptide and the putative active site triade RGH necessary for DNA hydrolysis. These putative secreted salivary endonucleases may have the same biological role as Lundep in other sand fly species. The expressed sequence tag of Lundep has a predicted signal peptide of 24 aa, indicative of secretion. Accordingly, endonuclease activity was confirmed in SGEs of female Lu. longipalpis (Figure 1A). No endonuclease activity was detected in the SGs of males, which are not blood feeders (Figure 1B). Moreover, this activity is present in secretions of probing Lu. longipalpis (Figure 1C). Rabbit polyclonal antibodies against rLundep blocked the DNase activity of rLundep and SGE, indicating that Lundep is the major endonuclease in Lu. longipalpis SGs (Figure S2).
Recombinant Lundep (rLundep) was cloned from a SG cDNA library using standard PCR procedures and subcloned into VR2001 expression vector. rLundep was expressed in HEK293 cells and purified by affinity and size exclusion chromatography (Figure S3A). rLundep has a strict requirement of divalent metal ions for endonuclease activity (Figure S3B,C) with broad pH optimum (5.0–8.0). Purified rLundep hydrolyzes both single-stranded (ss)- and double-stranded (ds)DNA with little sequence specificity (Figure S4). No significant RNase activity was detected (Figure S5). Lundep has a specific activity of 300,000 Kunitz U/mg as determined by a hyperchromicity assay on salmon sperm genomic DNA, where one Kunitz unit causes 0.001 change of absorbance at 260 nm per minute.
Because the scaffold of NETs is DNA, we hypothesized that SGE or rLundep could help L. major parasites escape from the microbicidal activity of NETs. To test this, we analyzed the ability of Lu. longipalpis SGE and rLundep to destroy human NETs induced by phorbol 12-myristate 13-acetate (PMA) and L. major metacyclic promastigotes.
Human neutrophils from healthy subjects were activated by PMA or L. major parasites for 4 h at 37°C. DNA—the major structural component of NETs—and neutrophil elastase (HNE) were detected by immunofluorescence (Figure 2A–F). Released human neutrophil elastase (HNE) was quantitated using a fluorogenic substrate (Figure 3A). First, we analyzed the ability of Lu. longipalpis SGE and rLundep to affect the integrity of PMA-induced NETs. PMA-activated neutrophils were incubated with culture medium (negative control), rLundep, Lu. longipalpis SG, and commercial bovine DNase-I (positive control; Figure 4). After 30 minutes of incubation, supernatants were collected for HNE quantification and neutrophils fixed and stained for DNA (blue) and HNE (green). NETs remained intact in cells treated with culture medium but were disintegrated by SG or rLundep (Figure 2B, C). Mutagenesis of Lundep active site (mLundep, RGH197AAA) abrogated its DNase activity and did not affect NETs' integrity (Figure 4). We also looked at the HNE released from NETs as an indicator of NET destruction. HNE is normally bound to NETs and found at low concentrations in culture supernatants [17]. PMA-activated neutrophils showed that treatment with SGE or rLundep significantly increases the concentration of HNE compared with control samples (Figure 3A).
The effect of rLundep and Lu. longipalpis SGE on Leishmania-neutrophil interaction was analyzed in vitro. Neutrophils were activated with 106 Leishmania major expressing red-fluorescent protein (Lm-RFP) promastigotes for 4 h before treatment with Lu. longipalpis SG or rLundep. The NET-destroying activity of Lu. longipalpis SGE and rLundep hydrolyzed the parasite-induced NETs (Figure 2E,F) and significantly increased the concentration of HNE in the supernatants when compared with medium alone (Figure 3B). Furthermore, Lu. longipalpis SGE and rLundep significantly increased L. major survival, indicating that parasites can escape from the leishmanicidal activity of NETs (Figure 3C). Our results show that PMA- or L. major -induced NET are disrupted by treatment with commercial bovine DNase-I, Lu. longipalpis SGE, or rLundep (Figure 2A–F; Figure 3). These results indicate that the effect of Lu. longipalpis SG and rLundep in helping L. major parasites escape from NETs is exclusively due the catalytic activity of the salivary endonuclease.
The infection model we used in this work is Lu. longipalpis-L. major. Although Lu. longipalpis is not the natural vector of L. major this specie of sand fly is permissive to L. major infections in laboratory conditions. Furthermore, our laboratory has a well-established murine model of infection for this pair.
Sand-fly bites and needle injection have been previously shown to induce neutrophil recruitment to the parasite inoculation site [18]. These neutrophils capture L. major parasites early after inoculation and efficiently initiate L. major infection. To investigate whether rLundep had any effect in exacerbating parasite infectivity in vivo, a model of L. major infection in C57BL/6 mice was utilized. Four- to five-week-old mice (five animals per group, three independent experiments) were intradermally infected with 103 L. major promastigotes (control) or parasites admixed with rLundep. Co-inoculation of rLundep with L. major parasites resulted in a significantly increased cutaneous lesion, averaging 2-fold larger than those observed in the control group (Figure 5A). By week 9, control mice had their lesions significantly reduced, whereas cutaneous lesions in mice inoculated with the parasite-rLundep mixture had not healed. Furthermore, the presence of rLundep in L. major inoculum resulted in a markedly higher parasite burden in their lesions (15-fold) when compared with parasite alone or in the presence of mLundep (Figure 5B and Figure 6).
The plasma contact system consists of five plasma proteins that assemble when blood comes into contact with negatively charged surfaces. It has been previously shown that soluble DNA and NETs allow the assembly and activation of the contact system [19]. The effect of Lundep on the intrinsic coagulation pathway activation was based on the generation of human factor XIIa by soluble DNA or aPTT reagent. One hundred nM of Lundep or TBS was preincubated at 37°C with 100 µg of salmon sperm DNA or 10 µl of aPTT reagent in the presence of the chromogenic substrate S2302. After 20 minutes, the reaction was initiated by adding 50 µl of human normal reference plasma, and the amidolytic activity of FXIIa was measured at 405 nm. Figure 7A shows that pretreatment of soluble DNA with Lundep markedly inhibited activation of FXIIa in human normal reference plasma while no effect was observed when aPTT reagent was utilized. Oehmcke et al. [19] demonstrated that NETs and activated PMN cells can initiate contact activation and promote thrombus formation in the arterial and venous systems [20]. Consequently, our results indicate that the DNase activity of Lundep may contribute to the antithrombotic and anti-inflammatory functions of Lu. longipalpis saliva.
The feeding success of Lu. longipalpis on mice passively immunized with anti-Lundep or pre-immune IgG (control) was carried out on anesthetized mice. Starved female flies, 2 to 4 days old (never previously fed on blood), were placed in meshed vials, and groups of 10 flies were applied to the surface of both ears of mice passively immunized with either anti-Lundep or pre-immune (naïve) IgG. Flies were allowed to feed for 10 minutes and scored by visual inspection as fully fed, partially fed, or unfed. Figure 7B shows that flies fed on mice passively immunized with anti-Lundep antibodies were significantly less successful in obtaining a blood meal, while flies feeding on passively immunize mice with naïve IgG fed significantly better in the 10-minute period (p = 0.0001, χ2 test). Tripet et al. [21] highlighted the benefits of feeding aggregations in Lu. longipalpis in particular when feeding on hosts pre-exposed to sand flies bites, suggesting that group feeding maximizes the effect of the salivary component injected at the biting site. Accordingly, abrogating or reducing the salivary nuclease activity of the flies may result in a more viscous blood pool affecting the dispersion of other salivary components involved in blood feeding.
Although the killing mechanism(s) of pathogens trapped by NETs is poorly understood, the relevance of secreted endonuclease as a mechanism of evading the microorganism-killing activity of NETs has been highlighted by the presence of endonuclease activity in bacteria-evading, NET-dependent killing [17], [22]. The mechanism by which Leishmania promastigotes evade killing by neutrophils may be related to their ability to block oxidative burst and to enter a nonlytic compartment unable to fuse with lysosomes [23] or by resisting the microbicidal activity of NETs [11]. Munafo et al. [24] demonstrated that disrupting NETs with DNase-I attenuates extracellular production of reactive oxygen species (ROS) by neutrophils stimulated with bacteria. Moreover, this reduction in ROS production is independent of actin depolymerization and phagocytosis.
Our results demonstrate that Lundep, a female-specific secreted endonuclease, is an important factor contributing to establishment of Leishmania infection. Our observations are also in agreement with previous reports showing that parasite-induced NETs have leishmanicidal activity, thought to be mediated by histones, one of the NETs' structural components [12], [25]. NETs may also play a role in entrapment of parasites, hence interfering with their ability to enter host cells. Accordingly, by disrupting NETs, Lundep can effectively facilitate the survival of L. major parasites in neutrophils and, ultimately, in infecting macrophages and dendritic cells.
Together, these in vitro and in vivo experiments demonstrate that rLundep and Lu. longipalpis SG degrade the DNA scaffold of NETs, destroying their functional integrity. Furthermore, Lundep protected L. major parasites from the leishmanicidal activity of NETs, increasing promastigote survival and exacerbating L. major infection. However, we cannot exclude the possibility of an induced pathology arise from anti-NET DNAse activity or some other untested mechanism. Because Lundep exacerbated infection with L. major in vivo and anti-Lundep antibodies abrogate the enzyme's function, Lundep may be considered a potential vaccine target in an anti-Leishmania vaccine cocktail.
With regards to the role of salivary endonucleases in blood feeding arthropods, Calvo and Ribeiro [26] proposed that a salivary endonuclease from the mosquito Culex quinquefasciatus could act as a spreading factor for other salivary activities by reducing the local viscosity at the biting site and hence decreasing the time taken to obtain a blood meal. We also found that anti-Lundep antibodies significantly decreased the feeding success of female Lu. longipalpis flies in passively immunized mice. These results may have epidemiologic relevance in the potential use of Lundep in vaccine, as longer probing and feeding times may trigger defensive behavior of the host, resulting in a disruption of blood feeding or even killing of the sand fly. Evidence of cross-talking between inflammation and coagulation is mounting in the literature. Fuchs et al. [27] demonstrated that NETs are a unique link between inflammation and thrombosis-promoting thrombus organization and stability. Moreover, NETs provide a negatively charged surface that allows the binding and activation of a contact activation system. Activation of the plasma contact system triggers several cascade systems such as the kallikrein-kinin system, the intrinsic pathway of coagulation, the classical complement cascade, and the fibrinolytic system [28] rendering an unfavorable environment for blood feeding arthropods. Accordingly, hydrolyzing the DNA scaffold of NETs at the biting site may also reduce local inflammation and prevent propagation of blood clotting facilitating the intake of a blood meal.
In conclusion, we provide experimental evidence that a secreted salivary endonuclease in Lu. longipalpis is capable of destroying NETs produced by activated human neutrophils and that this enzymatic activity exacerbates L. major infection in vivo. Furthermore, Lundep can assist the flies in blood feeding by reducing local inflammation elicited by the vertebrate host. We believe that our findings are of broad interest to the scientific community. Besides providing new insight into the basic biology of sand-fly blood feeding, the discovery of an endonuclease in SGs of Lu. longipalpis may also have broad implications for understanding the biologic function of secreted endonucleases in other arthropods and the pathogens they transmit.
Unless otherwise indicated, the protocols followed standard procedures, and all the experiments were performed at room temperature (25±1°C). All water used was of 18 MΩ quality, produced by a Milli-Q Synthesis water purification apparatus (Millipore, Billerica, MA). Hoechst 33258 dye (bis-benzamidine) was from Molecular Probes (Eugene, OR). Rabbit anti-human neutrophil elastase (HNE) was purchased from Calbiochem (San Diego, CA), and goat anti-rabbit Alexa Fluor 488 was from Invitrogen (Carlsbad, CA).
This project was approved by the Ethics Committee of the National Institutes of Health. Human neutrophils from healthy subjects were obtained under written informed consent, under NIH Clinical Center IRB-approved protocols from the NIH Clinical Center Department of Transfusion Medicine. Blood was taken from five healthy adults (aged 23–42 years).
Lu. longipalpis (Jacobina strain) were reared in the Medical Entomology Section, NIAID, NIH as described previously [14]. Sand flies were anesthetized with CO2 and transferred to a chilled plate until dissection. SGs were dissected under a stereomicroscope in 20 mM phosphate buffered saline, and SG were prepared according to Calvo and Ribeiro [29].
Public Health Service Animal Welfare Assurance #A4149-01 guidelines were followed according to the National Institute of Allergy and Infectious Diseases (NIAID), National Institutes of Health (NIH) Animal Office of Animal Care and Use (OACU). These studies were carried out according to the NIAID-NIH animal study protocol (ASP) approved by the NIH Office of Animal Care and User Committee (OACUC), with approval IDs ASP-LMVR3 and ASP-LMVR4E.
Blast-p analysis was performed with Lundep (AY455916) against the non-redundant database. Sequences were cleaned up to obtain a non-redundant set (proteins with >95% identity in the core domain were treated as identical) and aligned with ClustalX [30]. Alignments were manually checked, adjusted, and trimmed to include the conserved active site core. Phylogenetic analysis was performed using neighbor-joining analysis [31]. Gapped positions were treated by pairwise deletion. Poisson correction was used as a substitution model to determine pairwise distances. Confidence was determined using bootstrap analysis (10000 replicates) with 346 informative sites.
Endonuclease reactions contained 50 mM Tris, 150 mM NaCl, 5 mM MgCl2, pH 8.0 (TBS-M) and 200 ng double stranded (ds) circular plasmid DNA (VR2001; Vical Inc., San Diego, CA) in a final volume of 15 µl. Reaction mixtures were incubated with different dilutions of SG and recombinant Lundep. After 10 minutes at 37°C, samples were electrophoresed in 1.2% precast agarose gel (egel; Invitrogen, San Diego, CA) and visualized under ultraviolet (UV) light. RNase activity was carried out as described by Calvo and Ribeiro [29].
To demonstrate salivary secretion of endonuclease (Lundep) by Lu. longipalpis adult females, an ex vivo assay was designed. Ten female sand flies (4 days old, non-blood fed) were starved of sugar for 24 h before the test. Starved sand flies were allowed to probe for 30 minutes in a 1% agarose gel containing 50 mM Tris, 150 mM NaCl, 10 mM NaH2CO3, 1 mM MgCl2, pH 8.0, and 200 ng/ml ds circular plasmid DNA. The probing assay was carried out at room temperature and the probing gel kept in a slide warmer at 30°C (Precision Scientific, Chicago, IL). The combination of CO2 released from the bicarbonate buffer plus the temperature stimulated the sand flies to probe the slide [29]. After probing, the agarose gel was further incubated at 37°C for 30 minutes and stained with ethidium bromide. DNA hydrolysis in the gel was visualized under UV light.
PCR fragments coding for Lundep (AY455916) were amplified (Platinum Supermix; Invitrogen) from Lu. longipalpis SG cDNA using gene-specific primers designed to amplify the mature peptide and added a 6x-His tag before the stop codon. PCR-amplified product was cloned into a VR2001-TOPO vector (modified version of the VR1020 vector; Vical Incorporated) and the sequence and orientation verified by DNA sequencing. Plasmid DNA (5 mg; VR2001-Lundep construct) was obtained using EndoFree plasmid MEGA prep kit (Qiagen, Valencia, CA) and filter sterilized through a 0.22-µm filter. Recombinant Lundep was produced by SAIC Advanced Research Facility (Frederick, MD) transfecting FreeStyle 293-F cells. Transfected cell cultures were harvested after 72 h and the supernatant shipped frozen to our laboratory until further processing. Recombinant protein expression was carried out by affinity and size-exclusion chromatography as described elsewhere [32]. Protein identity and purity was determined by Edman degradation and mass spectrometry.
rLundep (10 ng) was separated by 4–20% NuPAGE in MES buffer (Invitrogen). After electrophoresis, samples were electrotransferred onto nitrocellulose membrane using an iBlot gel transfer system (Invitrogen). The membrane was incubated overnight at 4°C with TBS (25 mM Tris, 150 mM NaCl, pH 7.4) containing 5% (w/v) powdered nonfat milk (blocking buffer), followed by incubation for 90 min at room temperature with anti-rLundep rabbit sera diluted 1∶1000 in blocking buffer. The membrane was washed 4x with TBS-T (TBS, 0.05% Tween 20) and incubated with goat anti-rabbit alkaline phosphatase conjugated (Sigma, St. Louis, MO) diluted 1∶10000 in blocking buffer. The immunoblot was developed by addition of 1 ml of Western Blue-stabilized substrate for alkaline phosphatase (Promega, Madison, WI).
To determine the cation dependency of rLundep activity, reaction buffers (25 mM Tris, 150 mM NaCl) at pH 8.0 containing different divalent cations (MgCl2, CaCl2, ZnCl2, NiCl2 or CoCl2), 5 mM final concentration, were assayed. Reaction mixtures containing 10 nM rLundep in the appropriate buffer were incubated for 10 minutes at 37°C with 200 ng ds plasmid DNA. A negative control using 10 nM of rLundep without any ions was carried out. This negative control was also supplemented with 5 mM of EDTA to ensure no free divalent cations were present in this reaction. rLundep products were resolved in a 1.2% egel (Invitrogen) and visualized under UV light. For optimum pH, the following buffers were utilized: 50 mM sodium citrate (pH 3.0), 50 mM sodium acetate (pH 4.0 and 5.0), 50 mM MES (pH 6.0), 50 mM HEPES (pH 7 and 7.4), 50 mM Tris (pH 8.0 and 9.0), and 50 mM CAPS (pH 10.0). All reaction buffers contained 250 ng of plasmid DNA, 150 mM NaCl, and 5 mM MgCl2. Enzymatic reactions and DNA visualization were carried out as described above.
To determine substrate (ds or ss) specificity of rLundep, 1 nM of enzyme was incubated for 20 minutes at 37°C with different combinations of ds circular plasmid DNA or polynucleotides (ss and ds) as described [29]. Reactions were performed in TBS-M (20 µl final volume) and 400 ng of plasmid DNA or 2 µg of polynucleotides. Synthetic ds polynucleotides (1∶1 molar ratio) were produced by incubating ss-poly nucleotides for one cycle of 20 minutes at 95°C; 20 minutes at 85°C; 10 minutes at 72°C; 10 minutes at 60°C; and 10 minutes at 50°C. The pattern of nuclease activity of rLundep was determined by ion-exchange high-pressure liquid chromatography as described by Calvo and Ribeiro [29].
The DNA-hydrolytic activity assay was based on DNA hyperchromicity. A 1-ml quartz cuvette was loaded with 50 µg/ml DNA sodium salt from salmon testes (Sigma) dissolved in TBS-M buffer. Absorbance of the non-hydrolyzed DNA was measured before adding rLundep at three different concentrations (10, 20, and 50 nM) to the reaction cuvette. The increase of absorption of the sample at 260 nm was monitored by UV-Vis spectroscopy every 10 sec for 10 minutes. Initial velocity was calculated from slope of the linear phase of the progress curve. One Kunitz unit causes 0.001 change of absorbance at 260 nm per minute. Spectra were measured in quadruplicate in a spectrophotometer at 37°C. Identical solutions without rLundep were utilized for the blank in all cases. Spectroscopic hydrolysis analysis was performed at 37°C on a Varian Cary 100 Bio dual-beam spectrophotometer equipped with a Cary Cell Peltier temperature controller (Varian, Inc., Palo Alto, CA).
Mutation for Lundep was carried out using the QuikChange I site-directed mutagenesis kit (Agilent Technologies, Santa Clara, CA) following the manufacturer's recommendations. Briefly, a high-pressure liquid chromatography-purified 40-mer complementary primer set (Lundep-RGH194AAA Forward: 5′-CTCAATTTTCTATCAGCCGCAGCTTTAAGCCCC GAAGTGG-3′ and Lundep-RGH194AAA Reverse: 3′-GGAGTTAAAAGATAGTCGGCG TCGAAATTCGGGGCTTCAC-5′) was designed to mutate R194, G195, and H196 in the catalytic center of Lundep for AAA (RGH194AAA, mLundep). The primers were designed to carry the triple amino-acid mutation in the central region of the oligonucleotide, flanked by 13–15 nucleotides (3′ and 5′ ends). Lundep-VR2001 plasmid (10 ng) was utilized as a template in the PCR reaction. Amplification cycles were 95°C for 1 minute followed by 18 cycles of 95°C for 50 sec, 63°C for 50 s, and 68°C for 7 minutes. After a final extension step of 10 minutes at 68°C, the PCR product was digested with 1 U of DnpI to digest the parental supercoiled dsDNA. Two µl of the DpnI-treated PCR product was used to transform XL10 Gold ultracompetent cells. mLundep-VR2001 plasmid DNA was isolated from transformed cells and the sequences verified by DNA sequencing. A positive plasmid construct containing the mutation was selected for expression as described above. Protein purification and DNase activity were carried out as described above.
Polymorphonuclear cells were isolated from heparinized whole venous blood using Mono-Poly-Resolving Medium (MP Biomedicals, Solon, OH) according to the manufacturer's recommendations. Fresh heparinized blood was obtained from the NIH Clinical Center Department of Transfusion Medicine. The isolated fraction contained approximately 90–95% neutrophils as estimated by Trypan blue stain. The cells were counted and used immediately for NET production and visualization.
NET production and visualization was carried out as described elsewhere [17]. Freshly purified human neutrophils (106; 200-µl volume) were seeded in a Lab-Tek chamber slide (Thermo Scientific, San José, CA) for 1 h at 37°C. Seeded neutrophils were activated with 100 nM of PMA or 5×106 L. major expressing red-fluorescent protein (Lm-RFP) metacyclic parasites for 4 h at 37°C. Activated neutrophils were individually treated with 3 nM of rLundep, 1 Lu. longipalpis SG pair, 1 U of bovine DNase-I (positive control), and 20 nM of mLundep or medium alone (negative control). After 1 h at 37°C, supernatants were carefully removed for HNE quantification and the cells fixed and stained for DNA and HNE as described in [17]. Images were obtained using a DMIRE2 SP2 confocal microscope (Leica, Solms, Germany). All experiments were carried out in triplicate.
HNE concentration was measured in 50 µl of supernatant using the fluorogenic substrate N-methoxysuccinyl-Ala-Ala-Pro-Val-7-amido-4-methylcoumarin (Sigma) at a final concentration of 0.25 mM (100 µl final reaction volume). The assay buffer was 50 mM HEPES buffer, pH 7.4, 100 mM NaCl, 0.01% Triton X-100. After 1 h incubation at 37°C, the substrate hydrolysis was measured in a Spectramax Gemini XPS fluorescence microplate reader (Molecular Devices, Menlo Park, CA) with 365/450 nm excitation/emission wavelengths. HNE concentration was determined by using a standard curve of serial dilutions of purified HNE (Elastin Products Company, Inc., Owensville, MO).
L. major promastigotes (WR 2885 strain) were cultured in Schneider's medium supplemented with 10% heat-inactivated fetal bovine serum, 2 mM l-glutamine, 100 U/ml penicillin, and 100 µl/ml streptomycin. WR 2885 strain was typed at the Walter Reed Army Institute of Research [33]. Infective-stage metacyclic promastigotes of L. major were isolated from stationary cultures (4–5 days old) by negative selection using peanut agglutinin (Vector Laboratories, Inc., Burlingame, CT). Metacyclic promastigotes (1000) with or without 10 ng of rLundep in 10 µl of PBS buffer (supplemented with 5 mM of MgCl2) were inoculated intradermally into both ears' dermis using a 29-gauge needle. Evolution of the lesion was monitored weekly by measuring ear thickness using a vernier caliper (Mitutoya America Corporation, Aurora, IL).
Human neutrophil from five healthy donors (2×106) were infected with 107 Lm-RFP metacyclic parasites. After 4 h of incubation at 37°C, samples were treated with rLundep (100 nM), 1 SG pair of Lu. longipalpis, 1 U of bovine DNase-I (positive control) or medium alone (negative control). After 3 days of incubation at 23°C, cells were harvested and the supernatants spun down to evaluate the viable parasites. Live parasites were stained with Giemsa. All experiments were carried out in triplicates.
A region containing the DsRed gene flanked by the 5′ and 3′ untranslated regions of the Leishmania donovani A2 gene was amplified by PCR using the following forward A2F and reverse A2R primers and the pKSNEO-DsRed plasmid as template [34]. The A2F primer 5′-TGGCATATGCGTCGACCGCTGCTTGCGTTC-3′ contains a SalI restriction site. The A2R primer 5′-ACGCGTGGATCCTGAATTCGAGCTCTGGAGAGA-3′ contains a BamHI restriction site. The resulting ∼3.7-kb PCR fragment was cloned and sequenced. It contained an internal BamHI site that was mutated using standard PCR techniques. The approximately 3.7-kb SalI/BamHI fragment with the mutated internal BamHI site was subsequently cloned into the SalI and BamHI sites of the pF4X1.4hyg plasmid (Jena Biosciences, GmbH, Jena, Germany), resulting in the pA2RFPhyg plasmid. This plasmid was digested with SwaI to generate a linear fragment containing the RFP/Hyg expression cassette flanked by the 5′ and 3′ ssu sequences used for homologous recombination into the parasite 18S rRNA gene locus (ssu locus) as described in the original pFX1.4hyg plasmid (Jena Biosciences).
Promastigotes of the L. major Friedlin strain, NIH/FV1 (MHOM/IL/80/FN) were transfected by electroporation with 20 µg of SwaI-digested pA2RFPhyg plasmid as described previously [35] and plated onto 1% noble agar plates prepared in M199 Leishmania culture medium [36] supplemented with 4 µg/ml 6-biopterin (Calbiochem) and 100 µg/ml hygromycin B (Roche). Integration of the RFP/Hyg cassette into the ssu locus of hygromycin B-resistant clones was confirmed by PCR.
Parasite load was determined using a limiting dilution assay as described elsewhere [37]. Briefly, ear tissue was excised and homogenized in RPMI medium. The homogenate was serially diluted on Schneider medium 10% heat-inactivated fetal bovine serum, 2 mM l-glutamine, 100 U/ml penicillin, 100 µl/ml streptomycin, and 40 mM HEPES and seeded in 96-well plates containing biphasic blood agar (Novy-Nicolle-McNeal). The number of viable parasites was determined from the highest dilution at which promastigotes could be found after 21 days of culture at 23°C.
Polyclonal antibodies against rLundep were raised in rabbits by Spring Valley Laboratories, Inc. (Woodline, MD) using a standard protocol. Briefly, rabbits were immunized three times with 125 µg of rLundep every 21 days and the serum collected at day 72. A 10-ml aliquot of rabbit serum (immunized or naïve) was diluted to 50 ml in phosphate buffer, pH 6.5, and loaded onto a 5-ml HiTrap protein A HP column (GE Healthcare, Piscataway, NJ) and the IgG eluted using a linear gradient of citric acid (100 mM, pH 3.4) on an Akta purifier system (GE Healthcare). Fractions containing purified IgG were pooled and dialyzed against 1× PBS for 16 h at 4°C. IgG quantification was based on 1 absorbance unit at 280 nm equals 0.7 mg/ml.
To investigate the neutralizing activity of anti-rLundep antibodies on its enzymatic activity, an in vitro assay was developed. Purified anti-rLundep or naïve antibodies (0 and 5 µg/ml) were mixed with rLundep (10 nM) or SGEs (1 pair) and incubated for 30 minutes at 37°C. Plasmid DNA hydrolysis and visualization was carried out as described above.
The effect of rLundep on the intrinsic coagulation pathway was based on the generation of human factor XIIa by DNA or aPTT reagent. Ten µl of rLundep (100 nM) or TBS was preincubated at 37°C with 100 µg of salmon sperm DNA (Sigma) or 10 µl of aPTT reagent (Helena Laboratories, Beaumont, TX) in the presence of 5 µl of the chromogenic substrate S2302 (Diapharma, West Chester, OH). TBS alone was utilized as the reaction blank. After 20 minutes, the reaction was initiated by adding 50 µl of citrated-human normal reference plasma (Diagnostica Stago, Inc. Parsippany, NJ) and the amidolytic activity of fXIIa measured at 405 nm in a plate reader (Molecular Devices). All reactions were supplemented with 5 mM MgCl2. Final concentrations of rLundep and S2302 in the assay reaction were 13 nM and 300 µM, respectively.
Purified rabbit anti-Lundep or naïve antibodies were given to the recipient mice via intraperitoneal inoculation of 100 µg (100 µl volume) 15 minutes before exposing five to six CL57B/6 mice to sand flies (10 flies in each ear). Feeding success of sand flies in passively immunized mice was measured on anesthetized animals (four per group) as described Belkaid et al. [38]. Briefly, groups of 10 female flies (3 to 4 day old) were caged in polystyrene tube the day before the experiments was carried out and deprived of sugar. Ready-to-use vials containing starved flies were applied to the surface of anesthetized mouse ear that was previously given either anti-Lundep antibodies or purified naïve IgG. Flies were allowed to feed for 10 min, removed, and scored as either fed or unfed. Flies with their entire abdomen fully engorged or with visible blood were considered as fed. Lu. longipalpis flies used in this experiment were not blood fed previously and had no eggs in their abdomen, enabling assessment of a blood meal by visual inspection.
Data were analyzed using GraphPad Prism v 5 software (GraphPad Software, Inc., San Diego, CA) and plotted as bar graphs or scatter plots. Comparisons were made with the 2-tailed t test with 95% confidence interval and 2-way analysis of variance. P<0.05 was considered significant (*, p<0.05; **, p<0.01; ***, p<0.001).). For feeding success, a χ2 test with 95% confidence interval was used.
|
10.1371/journal.pntd.0005500 | Acute juvenile Paracoccidioidomycosis: A 9-year cohort study in the endemic area of Rio de Janeiro, Brazil | Paracoccidioidomycosis (PCM) is a systemic mycosis caused by pathogenic dimorphic fungi of the genus Paracoccidioides. It is the most important systemic mycosis in Latin America and the leading cause of hospitalizations and death among them in Brazil. Acute PCM is less frequent but relevant because vulnerable young patients are affected and the severity is usually higher than that of the chronic type.
The authors performed a retrospective cohort study from 2001 to 2009 including acute juvenile PCM patients from a reference center in Rio de Janeiro, Brazil. Clinical, epidemiological, diagnostic, therapeutic, and prognostic data were reported.
Twenty-nine patients were included. The average age was 23 years old and the male to female ratio was 1:1.07. All cases were referred from 3 of 9 existing health areas in the state of Rio de Janeiro, predominantly from urban areas (96.5%). Lymph nodes were the most affected organs (100%), followed by the skin and the spleen (31% each). Twenty-eight patients completed treatment (median 25 months) and progressed to clinical and serological cure; 1 death occurred. Twenty-four patients completed 48-month median follow-up. Four patients abandoned follow-up after the end of treatment. The most frequent sequela was low adrenal reserve. Paracoccidioides brasiliensis S1 was identified by partial sequencing of the arf and gp43 genes from 4 patients who presented a viable fungal culture.
Acute juvenile PCM is a severe disease with a high rate of complications. There are few cohort clinical studies of acute PCM in the literature. More studies should be developed to promote improvement in patients’ healthcare.
| Paracoccidioidomycosis (PCM) is a neglected systemic mycosis caused by fungi of the genus Paracoccidioides present in the soil and is endemic to Latin America. The acute clinical form of this disease is less frequent than the chronic type of presentation. However, it is a more severe clinical condition, potentially life-threatening, affecting many important organs of the immune defense such as the lymph nodes, liver, and spleen. It can lead to serious complications as well as many sequelae in young vulnerable patients and children. There are few works published in the literature concerning clinical, epidemiological, prognostic and therapeutic features of acute juvenile PCM. This study aims to describe a 9-year cohort study of patients with acute juvenile PCM from a reference center in the endemic area of Rio de Janeiro, Brazil. Results demonstrate that early diagnosis can prevent poor outcomes and that a specialized medical structure is required to promote proper healthcare for these patients. Severe outcomes such as low adrenal reserve and a death occurred in 4 and 1 patients, respectively. The authors expect that these results can contribute to a better understanding of this severe fungal disease.
| Paracoccidioidomycosis (PCM) is a severe systemic mycosis endemic to Latin America [1]. In Brazil, it is the leading cause of hospitalizations and death among all systemic mycoses in immunocompetent patients and an important cause of morbidity [2–4]. Primary pathogenic dimorphic fungi of the genus Paracoccidioides are the etiological agents of this disease and, according to the literature; the remarkable genetic diversity between phylogenetic species seems to cause variations in clinical presentation, therapeutic response, diagnosis, and prognosis [5–7]. These hypotheses are based on observations of a few case reports with molecular identification of the fungus [8–11]. Acute PCM, also known as juvenile-type PCM, corresponds to 3% to 5% of all PCM cases but is considered the most severe clinical form of this mycosis because it affects vulnerable young or, less frequently, immunocompromised people, usually presenting as a disseminated disease involving the mononuclear phagocyte system including the lymph nodes, liver, spleen, and bone marrow [12]. It is frequently characterized by significant consumptive syndrome and massive cervical adenopathy that can be initially misdiagnosed as lymphoma or tuberculosis. Acute PCM usually develops with some complications and sequelae. The present study aims to perform a descriptive analysis of epidemiological, clinical, therapeutic, and prognostic data in a cohort of patients with acute PCM, evaluated in a reference center for this mycosis in Rio de Janeiro state, Brazil, an important endemic area. There are few detailed cohort clinical studies concerning this type of clinical form in the literature because of its low incidence and due to its environmental, geographical, and occupational exposure variability [13–15]. Furthermore, molecular identification of available Paracoccidioides strains is provided to contribute to a better understanding of this challenging subject.
Evandro Chagas National Institute of Infectious Diseases Research Ethics Committee has approved this study protocol under the register CAAE 42590515.0.0000.5262. The patients’ data were anonymized/de-identified to protect patients’ privacy/confidentiality.
This is a retrospective cohort study from a reference center for PCM clinical assistance in the state of Rio de Janeiro, an important endemic area for this mycosis in Brazil. Rio de Janeiro is a Brazilian state located in the southeast of the country, with the highest demographic density in Brazil. Rio de Janeiro is divided into 92 municipalities; 5 are predominantly rural, and 81 have habitants living in rural zones [16]. In 2013, the state was divided into 9 regions known as health areas based on demographic and socio-economic data to better understand and to plan strategic health policies (Fig 1).
All cases of acute PCM admitted to INI/Fiocruz from 2001 to 2009 were included in this study. Inclusion criteria were diagnosis of PCM by direct examination, culture or histopathology, and classification of the acute form by clinical findings based on a consensus in PCM [12]. Medical records of these patients were collected, and information concerning epidemiological, clinical, and therapeutic data was documented on a standardized anonymized clinical report form.
Direct examination of clinical samples was performed with KOH 10%. Cultures were performed on modified Sabouraud dextrose agar and mycobiotic agar at 25°C. Suspected Paracoccidioides cultures were subcultured in Fava-Netto agar at 37°C for dimorphism confirmation. Serology for PCM was performed using Ouchterlony double immunodiffusion (ID).
Histopathological examination of tissues was performed using hematoxylin-eosin (H&E) and other staining techniques such as Grocott's methenamine silver stain (GMS) or Periodic acid—Schiff (PAS) for a better visualization of fungal parasitic structures.
All patients were given a standard routine clinical evaluation including physical examination, blood tests (hematology and biochemistry), parasitological stool analysis, bacterial microbiologic analysis of sputum (3 samples for acid-fast bacilli and culture), chest radiograph, and other imaging exams when indicated (brain computerized tomography [CT], and abdominal CT or ultrasonography).
Adrenal function was evaluated using the ACTH (Cortrosyn) stimulation test. Low adrenal reserve was defined as a normal basal level without reaching at least 20 mg/dl after 30 and 60 minutes of stimulation. Glandular insufficiency was defined as reduced basal levels associated with clinical symptoms (extreme fatigue, skin hyperpigmentation, low blood pressure, fainting, hypoglycemia, nausea, diarrhea or vomiting, and abdominal pain).
The therapeutic regimen was based on consensus in PCM [12]. Sulfamethoxazole/trimethoprim (SMZ-TMP), itraconazole and amphotericin B were the standard drug therapy. Combination of drugs was administered in cases of refractory, poor outcome and severe clinical conditions such as neurological complications.
The grade of severity was based on a standard classification proposed by Mendes [17]. Cure criteria were clinical, serological and radiological according to those defined in the consensus as well as the periodicity of clinical and laboratorial evaluations [12]. The recommended time of follow-up was 24 to 60 months after the end of treatment.
Genomic DNA was extracted from the yeast phase of viable Paracoccidioides cultures obtained at the time of the patients’ diagnosis. Amplicon products of polymerase chain reaction (PCR) using 2 protein-encoding genes arf (ADP ribosylation factor) and gp43 (glucan 1,3-beta-glucosidase) primers were submitted to automated partial nucleotide sequencing in the Platform PDTIS/FIOCRUZ [10]. A BLAST (Basic Local Alignment Search Tool) analysis (www.ncbi.nlm.nih.gov/BLAST) was performed comparing these sequences to those from isolates belonging to the Paracoccidioides brasiliensis complex previously deposited by Matute et al.
Statistical analysis was conducted using Stata 12. The data were summarized as percentages for categorical variables and median for continuous variables.
Twenty-nine patients fulfilled the inclusion criteria. Clinical, demographic, and prognostic data of these patients are summarized in Table 1. Therapeutic and serological data are presented in Table 2.
The average age was 23 years (minimum 8, maximum 44). Students and general services assistants were the 2 occupational labors reported. Rural activity was reported in 1 case. All 29 cases from this study were referred from 3 of 9 health areas in the state of Rio de Janeiro, predominantly from the urban area (Fig 1). All these patients were born in Rio de Janeiro state except 2 patients who were from Bahia and Minas Gerais states (northeast and southeast, 1640 and 500 km from Rio de Janeiro, respectively). There was no report of travels to other regions before the symptoms began. The median time of symptoms’ onset until PCM diagnosis was 4 months (minimum 2, maximum 84). Diagnostic confirmation of PCM occurred most frequently (72.4%) through clinical specimen analysis obtained by invasive methods such as biopsies (Table 3).
The lymph nodes were the most affected organs (100%), followed by the skin (31%), the spleen (31%), and the liver (27.6%). Fig 2 shows 2 patients from this study presenting lymph nodes enlargement and skin lesions, before and after treatment. The adrenals and the central nervous system (CNS) were affected in 5 and 2 patients, respectively. The frequency of other organs involvement is detailed in Table 4.
The most common clinical complication was low adrenal reserve, while the most frequent laboratory abnormalities observed were anemia and hypoalbuminemia. Colon stenosis leading to intestinal obstruction occurred in 1 patient. Low adrenal reserve was the most frequent sequela requiring indefinite steroid replacement therapy. Hospitalization was necessary for 20 patients to promote intensive healthcare support and/or intravenous therapy with amphotericin B. Coinfections were detected in 8 patients: 3 cases of pulmonary tuberculosis (TB) from which 1 also presented hepatitis C (HCV), and 1 case of human immunodeficiency virus (HIV) infection. The other patients presented intestinal worm infections. Twenty-eight patients completed treatment (median 25 months) and progressed to clinical and serological cure; 1 death due to neurological PCM occurred. Itraconazole and SMZ-TMP in monotherapy prevailed followed by different types of drug associations (Fig 3).
Twenty-four patients completed follow-up after the treatment and 4 abandoned it. Paracoccidioides brasiliensis S1 was identified by the partial sequencing of the arf and gp43 gene regions from 4 patients from whom we could retrieve a viable culture. Two of these 4 patients presented a moderate clinical condition without complications or sequelae, while the other 2 cases were considered severe since the adrenals were affected, although the recovery of adrenal’s function occurred in 1 patient.
Acute juvenile PCM is a severe presentation of this neglected systemic mycosis that most frequently affects vulnerable young patients with low socioeconomic status and can lead to life-threatening clinical conditions, serious complications, hospitalization, and permanent sequelae. The infection is acquired via the inhalation of Paracoccidioides filamentous propagules present in the soil of endemic areas but also probably dispersed by the wind and influenced by other climatic features [18]. Thus, infection easily occurs in the countryside and is usually related to rural activities, although susceptible individuals can be found in the urban areas without this occupational profile as observed in classical acute PCM and confirmed by sociodemographic data of the present study. These patients present a specific cellular immunodeficiency against the fungal agent and the inability to develop a granuloma response leads to acute/subacute infection’s progression soon after hematogenous spread. The mononuclear phagocytic system is the main site of infection. In the literature, the frequency of organs involved is as follows: the lymph nodes, the digestive tract, the liver, the spleen, bones, joints and the skin [12,19]. In this study, the lymph nodes were affected in all cases, followed by skin lesions and hepatic/spleen involvement. Acute PCM cases without lymph nodes involvement are rare and challenging to diagnose [10]. Cases with bone and joints commitment were not observed, but important organs such as the adrenals and the CNS were affected in the patients from the casuistic studied. Lymph abdominal presentation was once thought to be related to Paracoccidioides lutzii infection [6], although this is based on clinical sporadic observation and a lack of studies with consistent statistical data does not allow inferences about species-specific clinical manifestations. In this work, although a few viable fungal cultures could be retrieved to allow a genetics evaluation, the results suggest that P. lutzii alone is not responsible for lymph abdominal, critical, and severe conditions, and perhaps the severity of the disease could be better explained by host-parasitic interaction [20,21], as proposed before in a published clinical and molecular severe case report in which the etiological agent involved was P. brasiliensis S1 [10]. The important immune profile in PCM physiopathology is applicable in this context. Acute and subacute PCM are characterized by high titers of secondary antibodies produced by lymphocytes B activation due to exacerbated Th2 cytokines responses such as IL4, IL5, and IL10 [22,23]. Thus, serology tests usually present high positivity in acute and severe cases [17]. The absence of antibody detection in Ouchterlony immunodiffusion tests can be explained by low titer production or differences in antigenic profiles obtained from distinct species to perform the test [24–28]. In this study, 3 strains identified as P. brasiliensis S1 were isolated from patients with positive serology tests. The other patient, whose strain was also identified as P. brasiliensis S1, presented negative results before, during, and after treatment despite the severity of the case. The small number of strains retrieved for molecular analysis is a limitation of this study although these data reveal that the immune response in PCM needs to be further explored. Epidemiological factors and immune response patterns can justify the presence of intestinal worm coinfection. The incidence of TB and PCM coinfection reported in the literature is about 5% to 19% [29,30], similar to the data detected in this study. However, pulmonary TB is mostly related to the chronic type of PCM. HIV and HCV infections are barely reported to coexist with PCM perhaps because of epidemiology aspects, since PCM is essentially a rural mycosis. The important superposition of other severe and endemic infections such as TB, HIV and HCV also highlights the vulnerability of these young patients.
Regarding prognostic data, the results presented here show high rates of hospitalization (almost 70% of cases) and the occurrence of a fatal outcome in 1 case. Therefore, the occurrence of severe sequelae such as low adrenal reserve requiring indefinite steroid replacement therapy is a worrisome problem [31]. Early diagnosis and treatment can prevent complications and poor outcomes. In the literature, the time from symptoms’ onset until health assistance access is about 1 to 3 months, while this study shows a 4-month period until diagnosis confirmation [12]. PCM diagnosis requires a specialized health multidisciplinary team and laboratory infrastructure. Invasive techniques are also required to obtain clinical samples for diagnostic confirmation. Almost 50% of the patients from this study were referred from heath areas with low access to specialized medical assistance (Fig 1). Complications related to diagnostic delay can be severe and even fatal, such as adrenal insufficiency, acute abdomen due to intestinal obstruction, seizures secondary to fungal brain tumors, and respiratory impairment [31–35]. These complications require a high-complexity multidisciplinary health assistance, including surgery and intensive care support. In this study, a satisfactory plan of treatment and follow-up accomplishments were obtained, since consensus recommends 6–24 months and 24 months, respectively [12]. This certainly contributed to reducing PCM morbimortality in the casuistic included in this study. Drug association is a good strategy for critical and neurological cases [36,37]. The authors encourage clinical research and more reports concerning acute PCM clinical experience to promote greater knowledge and assistance for this challenging, severe, and neglected infectious fungal disease.
All sequences generated in this study were deposited in GenBank® (accession numbers KX463647, KX463648, KX463649, KX463650, KX463651, KX463652, KX463653, and KX463654).
|
10.1371/journal.pgen.1003787 | Meiotic Recombination in Arabidopsis Is Catalysed by DMC1, with RAD51 Playing a Supporting Role | Recombination establishes the chiasmata that physically link pairs of homologous chromosomes in meiosis, ensuring their balanced segregation at the first meiotic division and generating genetic variation. The visible manifestation of genetic crossing-overs, chiasmata are the result of an intricate and tightly regulated process involving induction of DNA double-strand breaks and their repair through invasion of a homologous template DNA duplex, catalysed by RAD51 and DMC1 in most eukaryotes. We describe here a RAD51-GFP fusion protein that retains the ability to assemble at DNA breaks but has lost its DNA break repair capacity. This protein fully complements the meiotic chromosomal fragmentation and sterility of Arabidopsis rad51, but not rad51 dmc1 mutants. Even though DMC1 is the only active meiotic strand transfer protein in the absence of RAD51 catalytic activity, no effect on genetic map distance was observed in complemented rad51 plants. The presence of inactive RAD51 nucleofilaments is thus able to fully support meiotic DSB repair and normal levels of crossing-over by DMC1. Our data demonstrate that RAD51 plays a supporting role for DMC1 in meiotic recombination in the flowering plant, Arabidopsis.
| Recombination ensures coordinated disjunction of pairs of homologous chromosomes and generates genetic exchanges in meiosis and, with some exceptions, involves the co-operation of the RAD51 and DMC1 strand-exchange proteins. We describe here a RAD51-GFP fusion protein that has lost its DNA break repair capacity but retains the ability to assemble at DNA breaks in the plant, Arabidopsis - fully complementing the meiotic chromosomal fragmentation and sterility of rad51 mutants, and this depends upon DMC1. No effect on genetic map distance was observed in complemented rad51 plants even though DMC1 is the only active strand transfer protein. The inactive RAD51 nucleofilaments are thus able to fully support meiotic DSB repair and normal levels of crossing-over by DMC1 in Arabidopsis. The RAD51-GFP protein confers a dominant-negative inhibition of RAD51-dependent mitotic recombination, while remaining fully fertile - a novel and valuable tool for research in this domain. These phenotypes are equivalent to those of the recently reported yeast rad51-II3A mutant, (Cloud et al. 2012), carrying the implication of their probable generality in other eukaryotes and extending them to a species with a very different relation between numbers of meiotic DNA double-strand breaks and crossing-overs (∼2 DSB/CO in yeast; ∼25–30 DSB/CO in Arabidopsis; ∼15 DSB/CO in mice).
| Meiosis is the specialised cell division essential for sexual reproduction that halves the chromosome number in the production of gametes. It is characterised by one round of DNA replication followed by two successive divisions, resulting in the production of 4 haploid nuclei from a single mother cell. In contrast to the mitotic cell divisions of development and growth, meiosis necessitates the recognition and coordinated segregation of pairs of homologous chromosomes, a function ensured by meiotic recombination in the majority of studied eukaryotes (reviews by [1], [2]).
Meiotic recombination is initiated by programmed DNA double strand breaks (DSBs), which are resected to generate 3′ single-stranded DNA overhangs (ssDNAs) that are bound by specialised recombinases. The resulting nucleoprotein filaments catalyse the invasion of a homologous DNA template by the 3′-ended DNA strand(s) to form a joint recombination intermediate, which in turn can be processed to yield crossing-over (CO) or non-crossing-over (NCO) products. In most eukaryotic organisms, the crucial invasion step of meiotic recombination requires the co-operation of the RAD51 and DMC1 recombinases. Biochemical and structural analyses indicate that RAD51 and DMC1 have homologous DNA pairing and strand exchange activities and have similar properties [3]–[7]. However, DMC1 is only required in meiosis while RAD51 is essential for both mitotic and meiotic recombination [8]–[11].
Repair of mitotic DSB is believed to principally involve the invasion of the sister chromatid, while during meiosis both sister and non-sister chromatids serve as templates for repair [12]–[15]. The choice of template for repair of DSBs is a key and specific feature of meiosis and must be tightly regulated to favour interhomologue recombination and crossing-over that ensure coordinated chromosomal disjunction at the first meiotic anaphase [8], [13], [16]. The RAD51 and DMC1 recombinases play key roles in these events and DMC1 is specifically implicated in meiotic interhomologue crossing-over [16].
Budding yeast RAD51 and DMC1 proteins share both overlapping and distinct functions during meiotic recombination [8], [17]–[19]. Absence of RAD51 strongly affects meiotic recombination and results in failure to repair DSBs and cell cycle arrest. Lack of DMC1 leads to a similar phenotype, with dmc1 mutants producing some viable spores [8], [20] and these defects of dmc1 mutant cells can be partially complemented by overexpression of RAD51 [21]. DMC1 nucleofilament formation is altered in the rad51 mutant, but RAD51 localisation appears normal in dmc1 mutants [17]. Simultaneous mutation of both recombinases results in a more severe phenotype than either of the single mutants [18], [19]. Both RAD51 and DMC1 are indispensable for efficient meiotic recombination in yeast and given similar activities of the two proteins, it has been generally accepted that they play similar roles in catalysing the invasion of the template DNA duplex. This assumption has however been called into question by the recent characterisation of the catalytically inactive yeast rad51-II3A mutant, showing that it is the presence of the RAD51 protein and not its strand-exchange activity that is needed in meiosis [22], [23]. In contrast, the equivalent dmc1-II3A mutant protein is inactive and has the same meiotic prophase arrest and absence of joint-molecule formation as the dmc1Δ mutant.
Given the importance of DMC1 for meiotic crossing-over and the considerable variation in the ratios of numbers of meiotic DSB and CO in different organisms, it is possible that the situation is more complex in vertebrates and higher plants (about 15 or 25 DSB per CO in mouse or Arabidopsis versus 2 in budding yeast (reviewed by [24]). Mouse dmc1 knockout mutants are completely sterile, with defects in homologous chromosome pairing, synapsis and DSB repair [25], [26] but the lethality of the rad51 mutants in vertebrates has hampered the study of their meiotic phenotype. Arabidopsis rad51 and dmc1 mutants have strikingly different meiotic phenotypes [27]–[32]. The chromosomes of rad51 mutants fragment in late meiotic prophase I and the plants are completely sterile [28]. In contrast and notwithstanding the absence of chiasmata and bivalents, meiotic chromosomes of dmc1 mutant plants remain intact and the plants have some fertility (∼1.5%; Couteau et al., 1999). As for yeast, loading of DMC1 is strongly reduced in Arabidopsis rad51 mutants, however localisation of RAD51 on meiotic chromosomes appears not to depend upon DMC1 [31], [32].
These differing meiotic phenotypes of Arabidopsis rad51 and dmc1 mutants have been generally accepted to be a clear illustration of DMC1 driving meiotic DSB repair through non-sister chromatid donors, while RAD51-driven repair uses sister-chromatid donors (discussed by [2]). An interpretation called into question by this work. We present here the analysis of a novel Arabidopsis RAD51 separation-of-function mutant, showing that RAD51 plays an essential role in supporting the activity of DMC1, which alone is sufficient to promote full homologous pairing, crossing-over and DSB repair in Arabidopsis meiosis.
RAD51 plays a central role in homologous recombination (HR) in both mitotic and meiotic cells of eukaryotes, including plants. To further investigate the roles of this protein during homologous recombination in planta, we constructed a RAD51-GFP translational fusion (Figure 1A). The RAD51 genomic coding sequence, including introns but without the stop codon, and 1031 bp of upstream sequence was amplified by PCR from DNA of wild-type Arabidopsis (Columbia) and the eGFP coding sequence fused to the 3′ end of the RAD51 open reading frame (Figure 1A). The fusion construct was introduced into RAD51/rad51 heterozygote plants and transformants expressing the RAD51-GFP translational fusion protein were selected. PCR genotyping of the RAD51 locus of the 32 RAD51-GFP transformants showed that 5 were rad51/rad51, 19 RAD51/rad51 and 8 were RAD51/RAD51. All five rad51/rad51 plants expressing the RAD51-GFP translational fusion protein were fully fertile, confirming that the fusion protein is able to complement the sterility phenotype of the Arabidopsis rad51/rad51 mutant (Figure 1B). This complementation strictly cosegregated with the transgene in the following generation. The RAD51-GFP fusion protein is thus properly expressed and functional during meiosis in these plants.
The fertility of the rad51/rad51 mutant plants complemented by the fusion protein clearly establishes that RAD51-GFP is able to substitute for the RAD51 protein in its essential meiotic role. A detailed cytogenetic analysis of meiotic progression in pollen mother cells (PMC) of plants expressing the RAD51-GFP fusion protein confirmed this, with meiotic stages appearing indistinguishable from wild-type meiosis (Figure 2). In wild-type plants, meiotic chromosomes condense at leptotene (Figure 2A). Pairing and synapsis of homologues is seen as the synaptonemal complex at pachytene (Figure 2B). Chromosomes further condense and the expected five bivalents are observed at metaphase I (Figure 2C). Homologous chromosomes then segregate to opposite poles to give two sets of five chromosomes at metaphase II (Figure 2D). Meiosis II then proceeds and gives rise to 4 haploid nuclei (Figure 2E). In contrast, pairing and synapsis are strongly impaired in rad51/rad51 mutant (Figure 2G). Defects in DSB repair further lead to strong chromosome fragmentation, fusion and chromosome mis-segration producing unbalanced and fragmented polyads (Figure 2H–J). In rad51/rad51 RAD51-GFP plants, meiosis appears indistinguishable from wild-type resulting in the expected 4 haploid meiotic products (Figure 2K–O). Normal structure of the synaptonemal complex at pachytene of rad51/rad51 RAD51-GFP meiosis was confirmed by immunolocalisation of the synaptonemal complex (SC) axial element protein ASY1 [33] and the SC transverse filament protein ZYP1 [34] (Figure 2P).
As seen above, the RAD51-GFP fusion protein is properly expressed and functional during meiosis. In somatic tissues, the expected strong GFP expression is visible in nuclei of meristematic cells in primary and lateral roots and none detected in non-dividing root transition and elongation zones (Figure 3).
To confirm the function of the fusion protein in mitotic cells, we tested its capacity to complement the sensitivity of rad51 mutant plants to the DNA cross-linking agent Mitomycin C (MMC). Wild-type and rad51 mutant plants, carrying or not the RAD51-GFP fusion protein, were grown on solid media containing increasing concentrations of Mitomycin C and growth was scored after 2 weeks (Figure 4). As expected under these conditions, rad51 mutants are highly sensitive while wild-type plants show little sensitivity to MMC (Figure 4A and B). Unexpectedly given the complementation in meiosis and the expression patterns in somatic cells, the RAD51-GFP protein does not complement the MMC hypersensitivity of rad51 plants. It acts as a dominant negative with both wild-type and rad51 mutant plants expressing the RAD51-GFP protein clearly hypersensitive to MMC (Figure 4A and B). This dominant negative phenotype implies that the fusion protein interferes with the proper functioning of the native RAD51 protein.
The importance of homologous recombination (HR) in the repair of DNA cross-links has led to the use of MMC hypersensitivity as an indirect test for recombination capacity in a number of organisms. Given the dominant negative MMC hypersensitivity conferred by RAD51-GFP, we also directly tested somatic homologous recombination in these plants using the previously described IU.GUS in planta recombination tester locus, consisting of an interrupted ß-glucuronidase (GUS) gene and a template GUS sequence for repair [35]–[37]. The IU.GUS recombination reporter locus was crossed into rad51 mutant plants expressing the RAD51-GFP fusion protein and somatic HR frequencies (HRF) monitored in F3 progeny homozygous for IU.GUS. Figure 4C shows quantification of spontaneous somatic recombination in WT and rad51 mutants, expressing or not the RAD51-GFP protein. As expected HR is severely reduced in rad51/rad51 mutants (1 recombinant spot found in 48 plants) compared to wild-type plants, which have a mean of 5.9 recombination events per plant (SEM = 0.75; n = 51). The presence of RAD51-GFP in both WT and rad51/rad51 plants resulted in levels of homologous recombination similar to those observed in the rad51 mutant (Figure 4C). The RAD51-GFP protein is thus not functional in mitotic recombination and these results very clearly confirm the dominant-negative effect, with the presence of the RAD51-GFP protein reducing recombination of RAD51/RAD51 plants to the level of rad51/rad51 plants (100-fold reduction; Figure 4C).
Despite its ability to restore the fertility of rad51 mutant plants, the RAD51-GFP protein is clearly defective for somatic homologous recombination. Presence of the RAD51-GFP protein creates a separation-of-function phenotype.
The phenotypes conferred by the Arabidopsis RAD51-GFP fusion appear very similar to those recently described in yeast for the mutant rad51-II3A protein, which retains its ability to form nucleofilaments but has no joint molecule activity [22], [23]. Using immunocytology to detect RAD51 nucleofilaments as brightly staining nuclear foci, we tested whether the RAD51-GFP protein retains its ability to assemble at sites of DNA breaks induced by gamma-irradiation in rad51/rad51 RAD51-GFP plants. We have recently shown that gamma-ray induced RAD51 foci are easily visualised in Arabidopsis using a dose of 100 Gy [38]. As expected, no foci and only diffuse anti-RAD51 nuclear staining were observed in root tip nuclei of unirradiated control plantlets (Figure 5A). Numerous foci were detected in nuclei from rad51/rad51 RAD51-GFP root tips fixed one or two hours after 100 Gy of gamma-rays (Figure 5A), confirming that the ability of RAD51 to assemble at sites of DNA damage is retained in somatic cells. The radio-inducibility of the RAD51-GFP was confirmed by western blotting analyses after irradiation (Figure S1 and Protocol S1).
The ability of the RAD51-GFP fusion to assemble at DNA breaks in meiotic cells was confirmed by immunolocalisation of RAD51 and ASY1 in pollen mother cell nuclei of rad51/rad51 RAD51-GFP plants, which show the expected numerous meiotic RAD51 foci (Figure 5B). Immunostaining of DMC1 protein in these nuclei revealed the expected presence of abundant DMC1 foci in the rad51/rad51 RAD51-GFP plants (Figure S2).
The RAD51-GFP protein is thus present and forms nucleofilaments in rad51/rad51 RAD51-GFP plants. Meiosis is normal and the severe, prophase I chromosome fragmentation of rad51 mutants is fully complemented by the fusion protein. RAD51-GFP is properly expressed and forms the expected radio-induced foci in mitotic cells, but is catalytically non-functional in mitotic HR. The most straightforward explanation for these results is that DMC1 protein carries out meiotic DSB repair in rad51/rad51 RAD51-GFP plants, and that it requires the presence of the (catalytically non-functional) RAD51-GFP protein. Should this be so, in the absence of DMC1, the RAD51-GFP protein should no longer be able to complement the rad51 meiotic phenotype (chromosomal fragmentation).
We thus crossed rad51/rad51 RAD51-GFP and dmc1 mutant plants and identified the rad51/rad51, dmc1/dmc1, and rad51/rad51 dmc1/dmc1 mutants in the F2, with and without RAD51-GFP. Observation of meiosis in pollen mother cells of these plants showed the expected ten intact univalents in dmc1 (Figure 6A), and fragmented chromosomes in rad51 metaphase I (Figure 6B). Meiosis progresses normally in rad51 mutants expressing RAD51-GFP, with 5 bivalents visible at Metaphase I (Figure 6C). Extensive chromosome fragmentation and fusion were observed in 100% of the meiocytes (n = 16) of rad51/rad51 dmc1/dmc1 RAD51-GFP plants (Figure 6D to F), clearly confirming that the meiotic DSB repair observed in rad51/rad51 RAD51-GFP mutants is DMC1-dependent. In accord with the dominant negative mitotic phenotype, chromosome fragmentation was also observed in all meiocytes of dmc1/dmc1 RAD51/rad51 plants expressing the RAD51-GFP fusion protein (n = 30. Figure 6G). The dominance was however incomplete, with fragmentation observed in only 61% of dmc1/dmc1 RAD51/RAD51 meiocytes expressing the RAD51-GFP fusion protein (n = 31).
These data confirm that the complementation of the meiotic chromosome fragmentation and sterility of Arabidopsis rad51 mutant plants by the RAD51-GFP protein is fully dependent upon the presence of DMC1. DMC1 is thus able to repair all meiotic DSB in Arabidopsis and depends upon the presence, not the strand exchange activity, of RAD51 to do so.
The RAD51 and DMC1 recombinases play essential roles in the repair of SPO11-induced meiotic DSB. The strikingly different phenotypes of Arabidopsis rad51 and dmc1 mutants however provide a clear illustration of their differing roles: the intact univalents in dmc1 meiosis, and chromosome fragmentation in rad51 meiosis showing that although RAD51 is able to repair all DSBs in the absence of DMC1, DMC1 cannot do so in the absence of RAD51. The absence of chiasmata in dmc1 mutants furthermore confirming that interhomologue crossing-over is a DMC1-dependent process.
As shown above, meiotic DSB repair is carried out by the activity of DMC1 alone in rad51 RAD51-GFP plants. We thus checked whether this results in elevated levels of meiotic interhomologue crossing-over in these plants. WT plants and rad51/rad51 RAD51-GFP plants of the Columbia (Col) ecotype were crossed to a wild-type plant of the Landsberg erecta (Ler) ecotype, to yield RAD51/rad51 plants heterozygotes for the RAD51-GFP transgene, and wild-type RAD51/RAD51 Col/Ler hybrids in the F1 (the dominant negative effect of the transgene allows analysis in F1 plants, see above and Figure 6D to 6F). Meiotic recombination was evaluated by analysing the segregation of markers in F2 populations originating from at least two F1 hybrid parents of each genotype. We measured crossing-over rates in two genetic intervals defined by insertion/deletion (INDEL) DNA sequence markers on chromosomes 1 and 3 (Table S1).
As seen in Table 1, no effect was observed on meiotic crossing-over rates in the rad51 separation-of-function mutant for either of the two genetic intervals. This was confirmed through counting chiasmata in metaphase I of wild-type and rad51/rad51 RAD51-GFP male meiocytes, which show means of 9.6 (SD = 0.5; n = 23) and 9.5 (SD = 0.5; n = 16) chiasmata per meiosis respectively. These results concord with those reported for the rad51-II3A yeast mutant and strongly suggest, as in yeast, that DMC1 is the catalytically active strand-exchange protein in Arabidopsis meiosis, and that RAD51 plays a supporting role [23].
In most eukaryotes, meiotic recombination requires the co-operation of two strand-exchange proteins, RAD51 and DMC1. RAD51 is present and active in mitosis and meiosis while DMC1 is specific to meiosis. DMC1 is not absolutely necessary since several organisms do not possess a DMC1 orthologue (e.g. Drosophila, Caenorhabditis elegans, Neurospora crassa and Sordaria macrospora) [9]. Why meiosis necessitates two DNA strand-exchange proteins and what unique functions are accomplished by DMC1 remains however elusive. A key to this question comes perhaps from the recent description of the yeast rad51-II3A separation-of-function mutant, showing that the joint molecule forming activity of RAD51 is not needed for meiotic recombination [23]. In yeast, the strand-exchange activity of DMC1 alone is thus sufficient for meiotic recombination and the requirement for RAD51 is for the protein itself (as a nucleofilament) and not for its catalytic strand-exchange activity. We present here an analysis of an Arabidopsis RAD51-GFP fusion protein that produces analogous phenotypes to the yeast mutant, confirming the yeast results and extending them to the higher plant Arabidopsis thaliana, with the implication that these conclusions are potentially applicable in general to eukaryotes with a DMC1 homologue.
As a tool to further study the roles of RAD51 in plants, we tagged the Arabidopsis RAD51 protein with the Green Fluorescent Protein (GFP). A number of published studies of different organisms have made use of such tagged RAD51 proteins to analyse the in vivo localisation of RAD51 to chromatin (see Table S2 for a referenced list). These reports show that FP-tagged RAD51 proteins form foci in both mitotic and meiotic cells, that the kinetics of focus formation is similar to that of native RAD51 and accurately depicts the behaviour of the endogenous RAD51 proteins. The fusion of the fluorescent protein at the N- or C-termini of RAD51 does thus not affect the ability of the protein to assemble at DNA DSB. N-terminal fusions are able to complement (sometimes partially) the radiation sensitivity or inviability of yeast, human, chicken, Ustilago maydis and Magnaporthae oryzae rad51 mutant cells (see Table S2). In contrast fusion of fluorescent proteins to the Rad51 C-terminus does not rescue the rad51 mutant mitotic phenotype (tested in S. pombe, human and chicken cells - see Table S2). Furthermore a dominant negative effect was observed on repair of gamma-ray induced DSB when RAD51-GFP was expressed in human cells [39], although not for ultraviolet light hypersensitivity in S. pombe [40].
We show here that the Arabidopsis C-terminal RAD51-GFP fusion is properly expressed in dividing mitotic cells, that the fusion protein localises to the nucleus and forms the expected gamma-ray induced nuclear foci. This C-terminal fusion protein is not however functional in mitotic recombination and does not complement the MMC hypersensitivity of Arabidopsis rad51 mutants. Furthermore, RAD51-GFP expression confers a dominant negative, rad51-like, recombination and MMC sensitivity phenotype on wild-type plantlets. As mentioned above, a human RAD51-GFP fusion protein also acts as a dominant negative in DSB repair and this phenotype presumably reflects the formation of inactive, mixed RAD51/RAD51-GFP nucleofilaments in these cells [39]. Although we do not have direct biochemical evidence, we favour the hypothesis that the RAD51-GFP fusion protein lacks strand-exchange activity by analogy with the phenotypes of the yeast rad51-II3A protein [23]. In yeast Rad51, mutation of three amino acids (R188, K361, and K371) in the low affinity DNA binding site inhibits the strand-exchange activity of the RAD51 protein, while leaving intact its capacity to form nucleofilaments [23]. These amino acids are conserved in the Arabidopsis RAD51 and two of them (R306, K316) are located in the C-terminal part of the protein (Figure S3). It is therefore possible that the steric effect caused by addition of the GFP affects the conformation of the RAD51 protein and/or the access of these amino acids to other proteins or DNA.
Considerably less is known concerning the activity of RAD51-GFP fusion proteins in meiosis. C-terminal fusions of RAD51 to GFP or RFP permit visualisation of RAD51 in meiotic cells of Sordaria macrospora [41]–[43]. Meiosis is normal in a wild-type Sordaria strain expressing the RAD51-GFP, with however a slight defect in sporulation (90 to 95% of viable spores instead of 100% in WT). In the absence of a rad51 mutant, meiotic complementation by the fusion protein has not been tested directly 41,43. Given that no DMC1 orthologue has been identified in Sordaria, implying that both meiotic and mitotic recombination are catalysed by RAD51, this would argue that the fusion protein is not dominant negative, at least in meiosis.
Recombination and interhomologue crossing-over are responsible for the physical recognition and linking of homologous chromosome pairs required to ensure proper chromosomal disjunction at the anaphase of the first meiotic division. The invasion of a template DNA duplex for the repair of a DSB by recombination is catalysed by RAD51-like strand-transfer recombinases and as is the case for many eukaryotes [9]. Arabidopsis has two of these: RAD51 active in meiosis and mitosis, and DMC1 which is meiosis-specific [44], [45]. Many studies specifically implicate the meiosis-specific DMC1 protein in meiotic crossing-over recombination with the homologous chromosome [2], [9], [46], as illustrated by the different meiotic phenotypes of Arabidopsis rad51 and dmc1 mutants. Absence of RAD51 in Arabidopsis meiosis leads to defects in chromosome pairing and synapsis, and to extensive chromosome fragmentation at meiotic zygotene/pachytene [28]–[32], [47]. Arabidopsis dmc1 mutant plants have a strikingly different meiotic phenotype, with synapsis defects and absence of chiasmata leading to the random segregation of intact univalent chromosomes [27], [29]–[32], [47].
Notwithstanding our results in somatic cells showing that RAD51-GFP is inactive in mitotic recombination and cross-link repair, its presence fully complements the meiotic defects of rad51 mutant plants. This essential activity is thus presumably furnished by the DMC1 protein in meiosis in rad51 mutants expressing RAD51-GFP. Removal of DMC1 confirms this hypothesis, with rad51-like meiotic chromosome fragmentation observed in dmc1 rad51 mutant plants expressing RAD51-GFP. The meiotic complementation of rad51 mutant plants by the fusion protein is thus fully dependent on the presence of the DMC1 protein. Arabidopsis DMC1 is able to repair all meiotic DSB and to do so requires the presence of the RAD51 protein, not its activity.
DMC1 is thus the active strand-invasion enzyme in meiotic crossing-over recombination in both Arabidopsis and yeast. In addition to the results presented here, the role of RAD51 in supporting DMC1 is seen in the reduced numbers of meiotic DMC1 foci in rad51 knockouts and in the phenotype of the rad51-II3A mutant [17], [18], [23], [31], [32]. A reduction of the fidelity of meiotic chromosome synapsis in the hypomorph Arabidopsis rad51-2 mutant suggests a role for RAD51 supporting DMC1 function [29], as does the impaired RAD51 and DMC1 focus formation observed in Arabidopsis brca2 plants [48]. In the absence of DMC1 however, significant levels of homologous pairing are observed in yeast [20], [21], [49], [50]. Similarly, traces of the synaptonemal complex central element (ZYP1 staining) and (limited) homologous chromosome synapsis of centromere-proximal regions are observed in the Arabidopsis dmc1 mutant [29], [30], [32].
The RAD51 nucleofilament thus plays a crucial role in regulating DMC1 activity but its strand-exchange activity must be inhibited in meiosis. In S. cerevisiae, restriction of the activity of RAD51 involves the action of the meiosis-specific protein Hed1. Hed1 down-regulates the activity of RAD51 to disfavour the use of the sister chromatid and hence favour DMC1-dependent inter-homologue recombination [51]–[53]. No Hed1 homologue has been identified in plants, but it seems likely that such a regulator exists. Recent work shows the importance of the ATR (Mec1) in regulating DMC1 filament formation in Arabidopsis, with absence of ATR in the Arabidopsis rad51 mutant permitting DMC1 assembly and subsequent synapsis, meiotic DSB repair and crossing-over formation [31]. DMC1 inter-homologue recombination in Arabidopsis is also controlled by ASY1 and ASY3, the Arabidopsis homologues of Hop1 and Red1 [31], [54], [55]. Arabidopsis DMC1 is thus clearly able to catalyse meiotic recombination using both sister or non-sister chromatid templates. In the absence of RAD51 this is however relatively inefficient as, in contrast to the separation-of-function mutant described here, atr rad51 double mutants still show significant chromosome fragmentation in meiosis [31].
Does this mean that DMC1 catalyses all strand-invasion in wild-type meioses? Either all meiotic recombination is catalysed by DMC1 with support from the RAD51 nucleofilament, or both DMC1 and RAD51 act catalytically in strand-invasion. It is not possible to distinguish between these two possibilities at this time, however the absence of an effect on crossing-over in both yeast and Arabidopsis is intriguing, given the significant variation in relative numbers of meiotic DSB and crossing-overs, with a ratio of 1.8 in yeast and 25–30 fold more DSB than crossing-overs in Arabidopsis (a high ratio is also seen in mice, with 15-fold more DSB than crossing-overs - see review by [24]). This would favour the idea of a supporting role for RAD51 in promoting DMC1 activity, and this was confirmed directly by in vitro experiments showing that both yeast RAD51 and rad51-II3A proteins stimulate the D-loop forming activity of DMC1 in the presence of the Mei5-Sae3 [23]. That RAD51 strand-transfer activity does play at least a minor “fail-safe” role in yeast meiosis is however suggested by the observation of a delay in the appearance of joint molecules and a slight reduction in sporulation efficiency (from 99 to 87%) in rad51-II3A [23]. No orthologues of Mei5 or Sae3 have been identified as yet in Arabidopsis [31], but it seems probable that they, or proteins of equivalent function exist.
Working with Arabidopsis thaliana, we describe here a RAD51-GFP fusion protein that lacks DNA repair activity but retains the capacity to assemble at DNA breaks. This protein fully complements the meiotic chromosomal fragmentation and sterility of Arabidopsis rad51 mutants, and we show that this depends upon DMC1. Even though DMC1 is the only active recombinase in the absence of RAD51 catalytic activity, no effect on genetic map distance was observed in complemented rad51 plants. The presence of inactive RAD51 nucleofilaments is thus able to fully support meiotic DSB repair and normal levels of crossing-over by DMC1 in Arabidopsis.
The Arabidopsis thaliana rad51 (AT5G20850) and dmc1 (AT3G22880) mutants used in this work have been previously described [27], [28].
Plants were grown under standard conditions: seeds were stratified in water at 4°C for 2 days and grown on soil or in vitro on 0.8% agar plates, 1% sucrose and 0.5× Murashige and Skoog salts (M0255; Duchefa Biochemie). Plants were then cultivated in a greenhouse or growth chamber with a 16/8 hour light/dark cycle, temperature 23°C and 45% to 60% relative humidity.
For translational GFP fusions, the genomic region without stop codon and a 1036 bp 5′ upstream sequence of RAD51 was amplified (forward primer TGATTAGCATTTAGCGTCAAG and reverse primer ATCCTTGCAATCTGTTACACC), inserted into pDONR221 and verified by sequencing. The complete fragment was then cloned into the GATEWAY destination vector pB7FWG2 in which the 35S promoter was removed with a SacI/SpeI digest [56]. The plasmid was inserted in an Agrobacterium tumefaciens C58C1 strain and used to transform wild-type and rad51 mutant plants by the floral dip method [57].
For the Mitomycin C sensitivity assay, seeds were surface-sterilised and sown onto solid medium containing 0.5× Murashige and Skoog salts, 1% sucrose, 0.8% agar and 0, 10, 20 or 40 µM Mitomycin C (SIGMA). After stratification for 2 days at 4°C, plants were grown for two weeks and sensitivity analysed as previously described [58], [59]. Plants with more than three true leaves were considered as resistant [58], [59].
The IU.GUS in planta recombination tester locus consisting of an interrupted ß-glucuronidase (GUS) gene and an internal repair template GUS sequence [35]–[37] was used to determine the rate of spontaneous somatic homologous recombination. Seeds were surface-sterilised, stratified at 4°C for 2 days and grown in petri dishes on 0.8% w/v agar, 1% w/v sucrose and 0.5× Murashige and Skoog salts for 2 weeks. Seedlings were then harvested and incubated in staining buffer containing 50 mM sodium phosphate buffer (pH 7.2), 0.2% v/v Triton X100, and 2 mM X-Gluc dissolved in N,N-dimethylformamide. Plants were then infiltrated under vacuum for 15 min and incubated 24 hours at 37°C. Staining solution was replaced with 70% ethanol to remove chlorophyll and blue spots counted under a dissecting microscope.
Wild-type and rad51 RAD51-GFP mutant plants (Col ecotype) were crossed with wild-type Landsberg erecta ecotype (Ler) plants. RAD51/rad51 heterozygotes carrying the RAD51-GFP (also heterozygous), and RAD51/RAD51 wild-type F1, Col/Ler hybrids were selected. Meiotic recombination rates were monitored in the F2 segregating populations by INDEL marker genotyping. For genotyping, seeds from the F2 populations were surface-sterilised and grown in vitro on 0.5× MS, 1% sucrose, 0.8% agar for two-weeks. Individual seedlings were harvested, DNA extracted using NaCl method and samples genotyped by PCR followed by analysis on 2% agarose gels.
Meiotic chromosome spreads were prepared according to Ross [60] with the modifications introduced by Fransz [61]. Whole inflorescences were fixed in ice-cold ethanol/glacial acetic acid (3∶1) for 3×30 min and stored at −20°C until further use. Immature flower buds were rinsed twice at room temperature in distilled water for 5 min followed by two washes in 1× citrate buffer for 5 min. Buds of appropriate size were selected under a binocular microscope and incubated for 3.5 h on a slide in 100 µl of enzyme mixture (0.3% w/v cellulase (Sigma), 0.3% w/v pectolyase (Sigma) and 0.3% cytohelicase (Sigma) in a moist chamber at 37°C. Each bud was then softened for 1 minute in 15 µl 60% acetic acid on a microscope slide at 45°C, fixed with ice-cold ethanol/glacial acetic acid (3∶1) and air dried. Finally, slides were mounted in Vectashield mounting medium with DAPI (1.5 µg.ml−1; Vector Laboratories Inc., http://www.vectorlabs.com/).
Five day-old seedlings were irradiated with a dose of 100 Gy from a 137Cs source according to Charbonnel et al. (2010). Preparation and immunostaining of nuclei were performed as previously described [62], except that slides were incubated with primary antibody (1∶100) for 24 hours at 4°C. The RAD51 antibody used in this study has been previously described and was raised in rabbit [63].
Immunolocalisation of proteins in pollen mother cells was performed as described previously [33], with the modifications introduced by Kurzbauer et al [31]. The anti-ASY1 raised in Guinea-Pig [33], [64] was used at a dilution of 1∶250. The anti-ZYP1 raised in rat [34] was used at a dilution of 1∶250. The anti-RAD51 raised in rabbit [63] was used at a dilution of 1∶100, and the anti-DMC1 raised in rabbit [65] was used at a dilution of 1∶20.
All observations were made with a motorised Zeiss AxioImager.Z1 epifluorescence microscope (Carl Zeiss AG, Germany) using a PL Apochromat 100X/1.40 oil objective. Photographs were taken with an AxioCam Mrm camera (Carl Zeiss AG, Germany) and appropriate Zeiss filter sets adapted for the fluorochromes used: filter set 25HE (DAPI), filter set 38HE (Alexa 488), and filter set 43HE (Alexa 596). Image stacks were captured in three dimensions (x, y, z) and further deconvoluted with the deconvolution module (theoretical PSF, iterative algorithm) of AxioVision 4.6.2 software (Carl Zeiss AG, Germany). For presentation, the pictures are collapsed Z-stack projections obtained using the Extended-focus module (projection method) of the AxioVision 4.6.2 software.
|
10.1371/journal.pgen.0030184 | Alu Recombination-Mediated Structural Deletions in the Chimpanzee Genome | With more than 1.2 million copies, Alu elements are one of the most important sources of structural variation in primate genomes. Here, we compare the chimpanzee and human genomes to determine the extent of Alu recombination-mediated deletion (ARMD) in the chimpanzee genome since the divergence of the chimpanzee and human lineages (∼6 million y ago). Combining computational data analysis and experimental verification, we have identified 663 chimpanzee lineage-specific deletions (involving a total of ∼771 kb of genomic sequence) attributable to this process. The ARMD events essentially counteract the genomic expansion caused by chimpanzee-specific Alu inserts. The RefSeq databases indicate that 13 exons in six genes, annotated as either demonstrably or putatively functional in the human genome, and 299 intronic regions have been deleted through ARMDs in the chimpanzee lineage. Therefore, our data suggest that this process may contribute to the genomic and phenotypic diversity between chimpanzees and humans. In addition, we found four independent ARMD events at orthologous loci in the gorilla or orangutan genomes. This suggests that human orthologs of loci at which ARMD events have already occurred in other nonhuman primate genomes may be “at-risk” motifs for future deletions, which may subsequently contribute to human lineage-specific genetic rearrangements and disorders.
| The recent sequencing of a number of primate genomes shows that small segments of DNA known as Alu elements are found repeatedly along all chromosomes, and indeed comprise ∼10% of the human genome. Although older Alu elements that have been in the genome for a long time accumulate some random mutations, overall these elements retain high levels of sequence identity among themselves. The presence of many near-identical Alu elements located close to each other makes primate genomes prone to DNA recombination events that generate genomic deletions of varying sizes. Here, by scanning the chimpanzee genome for such deletions, we determined the role of the Alu recombination-mediated deletion process in creating structural differences between the chimpanzee and human genomes. Using a combination of computational and experimental techniques, we identified 663 deletions, involving the removal of ∼771 kb of genomic sequence. Interestingly, about half of these deletions were located within known or predicted genes, and in several cases, the deletions removed coding exons from chimpanzee genes as compared to their human counterparts. Alu recombination-mediated deletion shows signs of being a major sculptor of primate genomes and may be responsible for generating some of the genetic differences between humans and chimpanzees.
| Mobile elements are a major source of genetic diversity in mammals [1,2]. Alu elements, a family of short interspersed elements (SINEs), emerged ∼65 million y ago (Mya) and have successfully proliferated in primate genomes with >1.2 million copies [2–5]. Alu elements consist of a left monomer and a right monomer [2,6]. Each of these monomers independently evolved from 7SL-RNA [7] and subsequently fused into the dimeric Alu element in the primate lineage [6]. Alu elements are known to be associated with primate-specific genomic alterations by several mechanisms, including de novo insertion, insertion-mediated deletion, and unequal recombination between Alu elements [8–11]. The Alu family consists of a number of subfamilies, which maintain high sequence identity among themselves (70%–99.7%) [12–15].
Mispairing between two Alu elements has been shown to be a frequent cause of deletion or duplication in the host genome [10,11,16]. A recent study of human-specific Alu recombination-mediated deletion (ARMD) reported a significant number of events associated with Alu elements [10]. An ARMD may arise through either interchromosomal recombination by mismatch of sister or nonsister chromatids during meiosis [17] or by intrachromosomal recombination between two Alu elements on the same chromosome. Previously, Sen et al. [10] found 492 human-specific ARMD events responsible for ∼400 kb of deleted genomic sequence in the human lineage [10]. Here, we report 663 chimpanzee-specific ARMD events identified from comparative analysis of the chimpanzee and human genomes. The chimpanzee-specific ARMD events deleted a total of ∼771 kb of genomic sequence in chimpanzees, including exonic deletions in six genes, sometime after the divergence of the human and chimpanzee lineages (∼6 Mya). ARMD events in the chimpanzee genome have generated large deletions (up to ∼32 kb) relative to human-specific ARMD events. Taking deletions in both the human and chimpanzee lineages into account, we suggest that ARMD events may have contributed to genomic and phenotypic diversity between humans and chimpanzees.
To investigate chimpanzee-specific ARMD loci, we first computationally compared the chimpanzee (panTro1) and human (hg17) genome reference sequences. A total of 1,538 ARMD candidates were initially retrieved using panTro1. These loci were converted to panTro2 (March 2006), which, due to the better quality of the sequence assembly, allowed us to eliminate a number of loci that mimicked authentic ARMD loci. Through a comparison of panTro1 and panTro2, we discarded 258 of the 1,538 loci (Table 1). The remaining 1,280 loci were manually inspected using the repetitive DNA annotation utility RepeatMasker (http://www.repeatmasker.org/cgi-bin/WEBRepeatMasker). In terms of local sequence architecture, human-specific mobile element insertions between two preexisting adjacent Alu elements could be computationally confused with a chimpanzee-specific deletion. Because the consensus sequences of the human-specific mobile elements (e.g., AluYb8, AluYa5, SVA, and L1Hs) have been well established in RepeatMasker, we were able to identify and eliminate from our analysis 189 human-specific insertion loci, including processed pseudogenes. The remaining 1,091 candidate ARMD loci were inspected using triple alignments of human (hg18), chimpanzee (panTro2), and rhesus macaque (rheMac2) sequences at each locus, and also on the basis of their target site duplication (TSD) structures (see Materials and Methods). After manual inspection, 342 of the candidate ARMD loci were examined by PCR to verify their status as authentic ARMD loci. Finally, combining computational and experimental results, 663 loci were confirmed as bona fide chimpanzee-specific ARMD loci (Table 1 and Dataset S1).
In this study, we combined computational data mining and wet-bench experimental verification, an approach that is optimal for identifying lineage-specific insertions and deletions [10]. Whereas Sen et al. [10] computationally compared the human and chimpanzee genomes, in our analysis, the draft version of the rhesus macaque genome sequence was used as an outgroup when filtering computational output for false positives (see Materials and Methods). This allowed us to eliminate 215 candidate ARMD loci prior to wet-bench verification, minimizing the cost and time needed to confirm authentic chimpanzee-specific ARMD events, as compared with the previous human-specific ARMD study.
Since the human-chimpanzee divergence ∼6 Mya, chimpanzee-specific ARMD events have occurred 1.3 times as often as their human-specific counterparts (663 chimpanzee-specific versus 492 human-specific events). The total amount of genomic DNA deleted by ARMD events from the chimpanzee genome is estimated to be 771,497 bp. However, when we consider that the average indel divergence between the human and chimpanzee genomes has been estimated at 5.07% [18], the precise amount of DNA deleted through ARMDs in the chimpanzee genome could be anywhere between ∼733 and ∼811 kb (±5.07% of ∼771 kb). The size distribution of DNA sequences deleted through chimpanzee-specific ARMD events ranged from 111 to 31,861 bp, with 1,164 bp average and 615 bp median ARMD sizes. Similar to the pattern observed in human-specific ARMD events [10], a histogram of the size distribution of chimpanzee-specific ARMDs is skewed toward deletions of shorter size, with ∼68% (449 of 663) of the deletion events shorter than 1 kb (Figure 1). As expected, about 70% of the deleted genomic DNA sequences are composed of repetitive elements (Table 2), of which Alu element sequences account for ∼64% (338 kb of 528 kb). Interestingly, the amount of sequence deleted through the ARMD process from the chimpanzee genome is twice as much as that from the human genome during the same period of time. Ten chimpanzee-specific ARMD events were found to have each deleted >7.3 kb of sequence (Figure 1); ARMD sizes this large were not observed in the human-specific study. Among these, the largest deleted sequence is 31,861 bp in length, within which only the SLC9A3P2 pseudogene and two intergenic regions are found in the ancestral sequence (i.e., human ortholog).
To examine the possible effects of the removal of ancestral genomic sequences during the 663 chimpanzee lineage-specific ARMD events, we retrieved the pre-recombination sequences (i.e., unaltered orthologs) from the human genome. About 46% (305 of 663) of the ARMD events were located within known or predicted RefSeq genes (http://www.ncbi.nlm.nih.gov/mapview/map_search.cgi?taxid=9606), and five ARMD events generated 13 exonic deletions in six genes annotated as either demonstrably or putatively functional in the human genome. Among them, two ARMD events deleted exons from demonstrably functional genes in the NBR2 (neighbor for BRCA1 [breast cancer 1] gene 2) and HTR3D (5-hydroxytryptamine [serotonin] receptor 3 family member D) genes. While no alternative pre-mRNA spliced forms exist for the NBR2 gene, the HTR3D gene shows three alternative pre-mRNA spliced forms in the human according to the ECR Browser (http://ecrbrowser.dcode.org). Among them, one of the HTR3D isoforms does not contain exon 3, which was deleted from the chimpanzee genome. Thus, chimpanzees could produce a similar protein to the HTR3D isoform mentioned above, because the ARMD event deleted the entire exon 3 and portions of some introns in the chimpanzee genome. However, we cannot rule out that the ARMD event has produced cryptic splicing sites causing either nonfunctionalization or subfunctionalization of HTR3D. The remaining three chimpanzee ARMD events generated exonic deletions in four putative human genes of unknown function (LOC339766, LOC127295, LOC729351, and LOC645203).
To further analyze the genomic sequences lost due to the ARMD process in the chimpanzee genome, we used the National Center for Biotechnology Information's (NCBI) UniGene utility (http://www.ncbi.nlm.nih.gov/sites/entrez?db=unigene) to look at the orthologous loci in the human genome, which contained sequences that would have been present in the chimpanzee genome if the ARMD events had not occurred. UniGene indicated that 164 ARMD events had caused deletions of coding sequence on the basis of expressed sequence tags (ESTs), although this number decreased to 94 when a high threshold indicating protein similarities (≥98% ProtEST) was selected (Table S1). This number is much higher than the exonic deletions in six genes generated by ARMD events reported above when RefSeq annotation was used instead.
Ten different Alu subfamilies are associated with chimpanzee-specific ARMD events: AluJo, AluJb, AluSx, AluSq, AluSp, AluSg, AluSg1, AluSc, AluY, and AluYd8. Their composition and ratio in chimpanzee-specific ARMD events are remarkably similar to those in human-specific ARMD events (Figure 2). The Alu subfamily analysis shows that the number of elements from each Alu subfamily involved in the ARMD process is proportional to the genome-wide copy number of each Alu subfamily in the chimpanzee genome. For example, the AluS subfamily has contributed the most to chimpanzee-specific ARMD events because it is the most successful Alu subfamily in the primate genome in terms of copy number. However, we found one exception to this rule; the AluJ subfamily is more ubiquitous than the AluY subfamily in both the chimpanzee and human genomes (Figure 3), but more members of the AluY subfamily were found to be involved in the ARMD process. The major expansion of the AluJ subfamily in primate genomes occurred ∼60 Mya, whereas the AluY subfamily expanded only ∼24 Mya [14,19,20]. On the basis of these ages, the individual members of the AluJ subfamily have likely accumulated more point mutations than those of the AluY subfamily. As a result, AluY copies have more sequence identity among them than do the AluJ copies, which results in increased involvement in ARMD events. In addition, we investigated intra-Alu subfamily recombination-mediated deletions for both the AluJ and AluY subfamilies. Of the 103 events involving at least one AluJ element in the ARMD event, only 15 (14.6%) involved recombination between two AluJ elements. The AluY subfamily shows a higher rate of intra-subfamily recombination than the AluJ subfamily, with 219 loci in which at least one AluY element was involved in the recombination event, and 57 (26%) that were between two AluY elements. This suggests that the rate of recombination between AluY elements is 1.8 times higher than that between AluJ elements. Taken together, this suggests that, in addition to the copy number of each Alu subfamily, the level of sequence identity between the individual Alu elements in the genome is also an important variable influencing ARMD events.
From a mechanistic viewpoint, four different types of recombination may occur between two Alu elements. An Alu element consists of left and right monomers. In the first type, comprising about 88% (583 of 663) of the ARMD events in our study, the recombination occurred between the same monomers of the two Alu elements. A second type of recombination occurred between two Alu elements in which one had previously integrated into the middle of the other. Such insertions are commonly found in both the chimpanzee and human genomes because each Alu element bears two endonuclease cleavage sites (5′-TTTT/A-3′) between its two monomers. About 8% (51 of 663) of the ARMD events in the chimpanzee genome are products of this second type of recombination. The third type of recombination, seen in 25 of the 663 events (∼4%), involved recombination between the left and right monomers on two separate Alu elements. The last type occurred between oppositely oriented Alu elements. Instances of this type of ARMD are very rare, found only in four of the 663 cases (0.6%). This style of recombination is likely to be uncommon because the stretch of sequence identity between two Alu elements oriented in opposite directions to one another is too short to frequently generate unequal homologous recombination. Instead, these two Alu elements are more likely to cause Alu recombination-mediated inversions or A-to-I RNA editing through the posttranscriptional modification of RNA sequences [21].
To analyze the frequency of recombination at different positions along the length of the Alu elements (which we refer to as “recombination breakpoints”) at our ARMD loci, we aligned the two intact human Alu elements involved in each recombination event with the single chimeric Alu element from the chimpanzee genome (Figure S1). The windows between the two Alu elements range in size from 1 to 116 bp, with a mean of 20 bp and a mode of 22 bp. In general, the ARMD loci generated by intra-Alu subfamily recombination, as well as the recombination events between relatively young Alu elements, show longer stretches of sequence identity than others. Through this analysis, we identified a recombination “hotspot” on the Alu consensus sequence (5′-TGTAATCCCAGCACTTTGGGAGG-3′), located between positions 24 and 45 (Figure 4). This recombination hotspot is congruent with previous studies of gene rearrangements in the human LDL-receptor gene involving Alu elements [22], and with the pattern of recombination found in the 492 human-specific ARMD events [10]. Of these studies, the former suggested that the hotspot sequence (therein called the “core sequence”) might induce genetic recombination because it subsumes the prokaryotic chi sequence (the pentanucleotide motif CCAGC), which is known to stimulate recBC-dependent recombination [23]. We searched for and found the CCAGC motif at four places (positions 31–35, 85–89, 166–170, and 251–255) along the Alu consensus sequences. The percentages of breakpoints found at these positions are 0.00886%, 0.00336%, 0.00406%, and 0.00372%, respectively. Among these, the percentages of breakpoints found at the latter three positions are similar to the average percentage of breakpoints across the entire length of the Alu elements (0.0035%) in our ARMD events. The only spot where the motif is found that showed a substantially higher percentage of breakpoints is the one located at positions 31–35, which is within our proposed hotspot. Therefore, this motif may invoke, but does not seem to be essential for the generation of ARMD events.
Interestingly, the 22-bp hotspot sequence contains no CpG dinucleotides. These CpG dinucleotides have been shown to mutate approximately six times faster than other dinucleotides in Alu elements [24] due to cytosine methylation and subsequent deamination [25]. In addition, when we aligned the consensus sequences of the 10 different Alu subfamilies involved in ARMDs, we found that the hotspot sequence is located within the longest stretch of their conserved regions. Furthermore, using the software utility WebLogo [26], we confirmed that this 22-bp sequence is the most conserved region among Alu elements involved in ARMD events (Figure 4). Therefore, the recombination hotspot that we have identified, by virtue of having an increased level of conservation among the Alu subfamilies involved in the ARMDs in our study, has potentially allowed frequent recombination between Alu repeats from different Alu subfamilies to occur.
Most Alu elements located in the primate genomes that have been sequenced (e.g., human, chimpanzee, and rhesus macaque) exist in high-GC content regions [3–5], and also have high GC content (an average of ∼62.7%). Moreover, it has also been previously reported that human-specific ARMD events preferentially occur in areas of high GC content (∼45% GC content, on average) [10]. To analyze the genomic environment of chimpanzee-specific ARMD events, we estimated the GC content of 20 kb (±10 kb in either direction) of neighboring sequence for each ARMD locus. Our results indicate that the chimpanzee-specific ARMDs are similar to human-specific ARMDs in having a tendency to occur in GC rich regions (45.2% GC content, on average). This preference is correlated with the distribution of Alu elements involved in ARMDs (Figure 3) because the genomic distribution of ARMD events would in effect have an a priori dependence on the preferred locations of Alu elements after insertion of the different Alu subfamilies. About 74% of chimpanzee-specific ARMDs are associated with the older Alu subfamilies, AluJ and AluS. Although young Alu subfamilies are found in AT-rich, gene-poor regions, the older Alu subfamilies are most often found in GC-rich, gene-rich regions [3]. This could account for the preferential occurrence of ARMD events in GC-rich regions. Moreover, the local rate of genomic recombination has been shown to be positively correlated with GC content [27], which may further explain the observed distribution of ARMD events. About 44% of genomic DNA deleted through ARMD events were Alu sequences in the human ortholog. This could indicate that regions of high local Alu element density within chromosomes are more likely to provide increased opportunities for local recombination, a trend previously noticed during analysis of the global genomic distribution of human lineage-specific ARMD events [10].
To further characterize the genomic environment of chimpanzee-specific ARMD events, we estimated the gene density of the genomic regions flanking each chimeric Alu element resulting from the process by extracting 4 Mb of flanking genomic sequences (±2 Mb in either direction), and counting the number of known or predicted chimpanzee RefSeq genes. The gene density of the flanking regions of chimpanzee-specific ARMD events is estimated to be, on average, one gene per 60.7 kb, which is similar to that of human-specific ARMD events (one gene per 66 kb). This indicates that the global distribution of chimpanzee-specific ARMD events is biased towards gene-rich regions, since the global average gene density in the chimpanzee genome is approximately one gene per 112 kb. To test for any relationship between the size of an ARMD and its flanking gene density or GC content, we performed a correlation test. While the r-values for both tests were negative, as would be expected given the danger of large deletions in gene-rich areas, the low p-values indicate that no significant correlation exists between the two variables in either test (gene density: r = −0.028; p = 0.472; GC content: r = −0.065; p = 0.095).
In order to estimate the polymorphism rates in chimpanzees, we analyzed and amplified a total of 50 chimpanzee-specific ARMD loci on a panel composed of genomic DNA from 12 unrelated chimpanzee individuals (see Materials and Methods). Our results show that the polymorphism level of chimpanzee-specific ARMDs (28%) is about two times higher than the polymorphism rate of human-specific ARMD events (15%) [10], which is in general agreement with the polymorphism levels from previous studies of chimpanzee- or human-specific retrotransposons (e.g., Alu and L1 elements) [28,29].
About 32% of the ARMD candidates were found to have ambiguous TSD structures and a triple alignment that proved too complex to assign ARMD status to the locus solely on the basis of our computational output. These loci were verified experimentally using PCR (see Materials and Methods) to determine the authenticity of the chimpanzee-specific ARMDs and identify false positives in the computational data, which were usually caused by human-specific Alu insertions. However, 16 ambiguous loci were identified at which human-specific Alu insertions were not present. In 11 of these loci, the human and gorilla genomes appear to have two Alu elements, while the chimpanzee and orangutan genomes have only one element at the orthologous position. DNA sequence analysis of the PCR products classified five of these 11 loci as chimpanzee-specific ARMDs, with the second of the two recombining Alu elements having integrated into the host genome after the divergence of orangutan and the common ancestor of humans, chimpanzees, and gorillas (Figure 5A). Four out of the 11 loci show a pattern consistent with incomplete lineage sorting, in which the ARMD event occurred before the divergence of great apes and was still polymorphic at the time of speciation. Subsequently, the chimeric Alu elements produced by these ARMD events became fixed in the chimpanzee and orangutan lineages while the two original Alu elements involved in the ARMDs were fixed in the human and gorilla genomes (Figure 5B). Incomplete lineage sorting has been reported in cases of retrotransposon insertion polymorphism involving closely related species [28,30]. In cases where the time between any genomic event and a subsequent speciation is very short, incomplete lineage sorting can easily occur. The remaining two of the 11 ambiguous loci were identified as parallel independent ARMD events in separate primate genomes by aligning the pre-recombination sequence and chimeric Alu elements (Figure 5C). These events suggest that orthologous loci may experience two independent lineage-specific ARMDs at different times (i.e., chimpanzee-specific ARMDs and orangutan-specific ARMDs).
In contrast, PCR analysis of the remaining five ambiguous loci (from the 16 referred to above) showed that humans and orangutans have two Alu elements, whereas chimpanzees and gorillas have only one at the orthologous position. Of these five loci, three showed a pattern suggesting incomplete lineage sorting events, while the other two were parallel independent ARMDs. For one of the loci displaying a parallel independent ARMD event, the structural characteristics of the two chimeric Alu elements resulting from independent recombination events are clearly different between the chimpanzee and gorilla genomes. The 574-bp chimpanzee genomic deletion occurred between the left monomer on the first Alu and the right monomer on the second Alu, whereas the 708-bp genomic deletion in the gorilla happened between the two left monomers of the two Alu elements.
These results indicate that at least ∼0.9% of chimpanzee-specific ARMD loci (2 of 233 loci which were analyzed by PCR) are shared by the gorilla genome and another ∼0.9% are shared by the orangutan genome, due to parallel independent ARMDs at two different time points in two separate primate genomes. As such, the presence of independently occurring ARMD events in both the human and chimpanzee genomes could lead to false negative events being missed during the previous analysis done by Sen et al. [10], although the frequency of such false negatives is likely to be very low. In addition, we believe that the human orthologs of the chimpanzee-specific ARMD loci represent sites predisposed for potential future ARMDs in the human genome that could generate human lineage-specific rearrangements and genetic disorders. Identifying putative ARMD hotspot genomic regions is not surprising based upon the frequency of Alu-mediated recombination events that have given rise to mutations in a number of different loci, including the LDLR and MLL1 genes [11,31–33].
Despite the high level of overall similarity between their genomes, humans and chimpanzees have subtly different genomic landscapes because of alterations such as insertions, deletions, inversions, and duplications after their divergence from a common ancestral primate [8–11,34,35]. Although from a mechanistic viewpoint, the chimpanzee-specific ARMD events are similar to the human-specific ones, the total number and size of deletions are substantially different between the two lineages. One reason for the observed differences between these two lineage-specific ARMD patterns may be the increased genetic diversity of the chimpanzee population as compared to the human population, which is known to have experienced a significant reduction in its effective population size after the divergence of humans and chimpanzees [36], leading to a consequent reduction in genetic diversity. These results are supported by the higher polymorphism level for chimpanzee-specific ARMDs than human-specific ARMDs.
Alu elements as well as other retrotransposons can contribute to the size expansion of primate genomes by increasing their copy numbers and causing homology-mediated segmental duplications [37–39]. However, the retrotransposon-mediated increase in genome size is not unilateral, because several processes such as retrotransposon-mediated deletions and recombination-mediated deletions concurrently act in the opposite direction, causing reduction in genome size as well [8–10]. Retrotransposon-mediated negative control of genome size has been well documented in plants such as Arabidopsis and rice [40,41].
In this study, we analyzed the contribution of ARMDs to genome size regulation in the chimpanzee genome by estimating an Alu-mediated sequence turnover rate, which is the amount of sequence increase caused by chimpanzee-specific Alu insertions relative to the amount of reduction by the chimpanzee-specific ARMD process. The copy number of chimpanzee-specific Alu elements (i.e., those that inserted after the divergence of human and chimpanzee) is ∼2,340, accounting for ∼700 kb of inserted sequence in the chimpanzee lineage [3], while the amount of sequence deleted by chimpanzee-specific ARMDs is ∼771 kb. Therefore, within the past ∼6 million y, the genome size of chimpanzees has not expanded but rather has contracted by ∼71 kb, when considering the combined effects of Alu retrotransposition and recombination-mediated deletion (i.e., the Alu-mediated sequence turnover rate is more than 100% in the chimpanzee genome). This observation suggests that ARMD events efficiently counteract genomic expansion caused by novel Alu inserts in the chimpanzee genome when compared to the human genome. A previous analysis of human-specific ARMD events indicates that the Alu-mediated sequence turnover rate is ∼20% in the human genome [10]. This significantly different turnover rate between the two species could be explained by differences in the tempo of Alu amplification (i.e., higher Alu retrotransposition activity in the human genome) and rates of ARMD events (i.e., higher ARMD activity in the chimpanzee genome). Ultimately, it is worth noting that at least in the chimpanzee lineage, concurrent Alu insertion/ARMD mechanisms have balanced the gain and loss of sequences during Alu-mediated genomic alterations.
To investigate whether chimeric Alu elements are able to retrotranspose in the chimpanzee genome, we tried to find progeny of the 663 chimpanzee-specific chimeric Alu elements using the BLAST-Like Alignment Tool (BLAT) program (http://genome.ucsc.edu/cgi-bin/hgBlat). However, we failed to recover any such elements in the chimpanzee genome for one or more of a number of reasons. First, Alu elements involved in ARMD events are expected to be relatively old (i.e., more than 6 million y) because our comparative analysis detects only ARMD events involving Alu elements that were inserted into the genome before the divergence of humans and chimpanzees. Therefore, most of the ARMD-associated Alu elements probably lost their ability to retrotranspose before the Alu–Alu recombination process. In reality, the contribution of chimpanzee-specific young Alu elements to the ARMD process may be extremely limited due to their low copy number (∼2,000 copies) in the chimpanzee genome [3]. Indeed, ARMD events generated by the relatively young AluY subfamilies account for 0.19% of the total AluY elements in the chimpanzee genome. Second, only a few source genes are responsible for new Alu subfamily amplification through retrotransposition. Although some Alu subfamilies (e.g., AluYc1) are still active in the chimpanzee genome [3,29], it is improbable that their source gene(s) are involved in the Alu–Alu recombination events. Similarly during an earlier analysis [10], we investigated the retrotransposition ability of 492 human-specific ARMD-generated chimeric Alu elements and were unable to recover their progeny as well.
Recently, the genomic relationship and genetic divergence between the human and chimpanzee genomes have been the subjects of extensive comparative genomic analyses on the basis of their respective draft genome sequences [3,35,42–44]. However, these studies have not focused on Alu-mediated genomic deletions in the chimpanzee lineage, aside from the 14 Alu retrotransposition-mediated deletions reported previously [9].
Thus, our study forms the first comprehensive analysis of recombination-mediated genomic alteration by Alu elements in a nonhuman primate (chimpanzee) lineage. We found 305 chimpanzee-specific deletions within protein-coding genes as annotated by the RefSeq gene annotation database, 299 genes from which introns were deleted, and six genes in which thirteen exons were deleted. Remarkably, two chimpanzee-specific ARMD events deleted exons from genes demonstrably functional in the human lineage (NBR2 and HTR3D), providing direct proof that the ARMD process contributes to creating phenotypic differences between humans and chimpanzees. The NBR2 gene is located near the BRCA1 gene on Chromosome 17, which is responsible for tumor repressor activity in the human genome, and shares a common promoter for transcription, forming a bidirectional transcriptional unit with BRCA1. Although the complete NBR2 cDNA sequence is ∼1.3 kb, it has a short open reading frame (112 amino acids), and is subject to nonsense-mediated decay [45,46]. In humans, this gene is suppressed by a non–tissue-specific protein complex that binds to its first intron (i.e., the 18-bp repressor element) [47]. However, in the chimpanzee lineage, an ARMD event occurred between the third intron and the 3′ flanking region, causing an exonic deletion (Figure 6A). Thus, this ARMD event could potentially inhibit NBR2 gene expression in the chimpanzee genome, regardless of whether or not the repressor element is present. Although the exonic deletion of the NBR2 gene has been independently reported through a comparative analysis of cancer genes between the human and chimpanzee genomes, the previous analysis did not report what caused this genetic difference between human and chimpanzee genomes [48]. Our study of chimpanzee-specific ARMDs illuminates the underlying molecular mechanism for this deletion.
A chimpanzee-specific ARMD event also deleted the first coding exon of HTR3D, a functional gene in humans (Figure 6B). This gene belongs to the 5-HT3 serotonin receptor-like gene family, which has been recently characterized [49]. The 5-HT3D subunit is not a functional receptor on its own (i.e., a homomeric receptor), but when it binds to the 5-HT3A subunit to form the heteroligomeric receptor, 5-HT, maximum response is significantly increased as compared to the homomeric 5-HT3A receptor [50]. HTR3D is primarily expressed in the gastrointestinal tract [50], where serotonin is synthesized extensively [51]. We speculate that the exonic deletion in this gene caused by the chimpanzee-specific ARMD event may lead to a reduction in serotonin levels in the chimpanzee lineage, and thus have an impact on physiological variation between the human and chimpanzee lineages.
The analyses using the RefSeq and UniGene annotations (see Results) indicate that ARMD events could have affected the expression of many genes. Moreover, intronic or intergenic deletions caused by ARMD events may also affect the levels of gene expression in both the human and chimpanzee genomes through alteration of splicing patterns and loss of transcription factor binding sites, further contributing to the divergence of the human and chimpanzee lineages. Additional studies of the functional genomics of the genes altered in both human and chimpanzee ARMD events will be instructive and provide new insight into the genetic and phenotypic differences between the two species.
Retrotransposon-mediated genomic rearrangement could be one of the major factors responsible for the lineage-specific changes in genomes that ultimately lead to speciation. Comparative investigations of the ARMD events apparent between the human and chimpanzee genomes indicate that this process plays an important role in the biological differences between humans and chimpanzees, and provides a reliable record of lineage-specific evolutionary histories due to the nearly homoplasy-free nature of these mutations. Moreover, in the chimpanzee lineage, the chimpanzee-specific ARMD process has completely counteracted the genomic expansion caused by new Alu inserts since the divergence of the chimpanzee and human lineages. The existence of parallel independent ARMD events found at the orthologous loci of some of the 663 chimpanzee-specific ARMD events suggest that other chimpanzee-specific ARMD orthologs in humans may be predisposed to undergo recombination between the two Alu elements in the future. These ARMD orthologous loci may be sites of unstable structure in humans as well as other apes, because they still preserve the pre-recombination structure that has proven itself susceptible to unequal recombination in the chimpanzee lineage.
To computationally screen the chimpanzee genome for potential ARMD loci, we used a technique previously described by Sen et al. [10] in a study of human lineage-specific ARMD events, with the distinction that, for this analysis, the query and target genomes were reversed. In summary, we extracted 400 bp of 5′ and 3′ flanking sequence for all chimpanzee Alu elements (PanTro1; November 2003 freeze) and joined the two 400 bp sequences to form a single “query” sequence. A best match for each query sequence was determined by using BLAT [52] against the reference human genome (hg17; May 2004 freeze). Then, the sequence in the human genome (the “hit”) found between the orthologs of the two 400 bp stretches of the query was extracted and aligned with the chimpanzee Alu element sequence initially used to design the query (the “query Alu”) using a local installation of the NCBI bl2seq utility.
One hallmark of de novo Alu insertion is the presence of TSDs flanking each side of the Alu element, generated by the target-site primed reverse transcription process [1,53–55]. However, the single chimeric Alu element created by an ARMD event lacks matching TSD structures in the chimpanzee because it is comprised of fragments from a pair of Alu elements with mutually unique TSDs at the orthologous ancestral locus [10]. If a potential ARMD locus exhibited the structures of a valid ARMD as described by Sen et al. [10], we accepted the computational detection as an authentic ARMD locus. In addition, we used the BLAT software utility [52] to compare the human, chimpanzee, and rhesus macaque genomes at each potential ARMD locus. If the two Alu elements in the human genome that are considered to be the pre-recombination Alu elements for an ARMD locus are shared with the rhesus macaque genome at orthologous loci, despite the presence or absence of TSDs, the single Alu element remaining at the orthologous chimpanzee locus is most likely a chimeric element generated an ARMD event. On the basis of these features, we manually inspected 1,538 potential ARMD loci retrieved by the computational data analysis. However, some loci displayed ambiguous TSD structure or remained ambiguous after analysis using the triple alignment. These loci were subjected to PCR analysis and, if necessary, DNA sequencing in order to confirm or eliminate each as being products of bona fide ARMD events.
PCR analysis was performed using four different primate species as templates. The cell lines used to isolate DNA samples corresponding the primate species are as follows: human (Homo sapiens) HeLa (CCL2; American Type Culture Collection [ATCC], http://atcc.org), common chimpanzee “Clint” (Pan troglodytes; NS06006B), gorilla (Gorilla gorilla; AG05251) and orangutan (Pongo pygmaeus; AG05252A). To evaluate polymorphism rates, we amplified 50 randomly selected ARMD loci on a common chimpanzee population panel composed of 12 unrelated individuals of unknown geographic origin obtained from the Southwest Foundation for Biomedical Research (San Antonio, Texas, United States).
Oligonucleotide primers for the PCR amplification of ARMD events were designed using the Primer3 utility (http://www-genome.wi.mit.edu/cgi-bin/primer/primer3_www.cgi). The sequences of the oligonucleotide primers, annealing temperatures, and PCR product sizes are shown in Table S2. Each PCR amplification was performed in 25-μl reactions using 10–50 ng DNA, 200 nM of each oligonucleotide primer, 200 μM dNTPs in 50 mM KCl, 1.5 mM MgCl2, 10 mM Tris-HCl (pH 8.4), and 2.5 U Taq DNA polymerase. Each sample was subjected to an initial denaturation step of 5 min at 95 °C, followed by 35 cycles of PCR at 1 min of denaturation at 95 °C, 1 min at the annealing temperature, and 1 min of extension at 72 °C, followed by a final extension step of 10 min at 72 °C. PCR amplicons were loaded on 1%–2% agarose gels, depending on the amplicon sizes, stained with ethidium bromide, and visualized using UV fluorescence. In cases where the expected size of the PCR product was greater than 1.5 kb, iTaq (Bio-Rad, http://www.bio-rad.com) or Ex Taq polymerase (TaKaRa, http://www.takara-bio.com) were used, following the manufacturer's suggested protocols.
When necessary, individual PCR amplicons were gel purified using the Wizard gel purification kit (Promega, http://www.promega.com) and cloned into vectors using the TOPO-TA Cloning kit (Invitrogen, http://www.invitrogen.com) according to the manufacturer's instructions. DNA sequencing was performed using dideoxy chain-termination sequencing [56] on an Applied Biosystems ABI3130XL automated DNA sequencer (Applied Biosystems, http://www.appliedbiosystems.com). Raw sequence reads were assembled using DNASTAR's Seqman program in the Lasergene version 5.0 software package (http://www.dnastar.com).
For each chimpanzee-specific ARMD locus, 10 kb of flanking sequence upstream and downstream were collected using a combination of in-house Perl scripts and the nibFrag utility bundled with the BLAT software package. The GC content of the flanking regions of each ARMD locus was calculated by analyzing the combined 20 kb of flanking sequence using another in-house Perl script, which excluded Ns from the analysis. Gene density around individual ARMD loci was estimated using the NCBI Map Viewer utility, run on Build 2.1 of the Pan troglodytes genome (http://www.ncbi.nlm.nih.gov/mapview/map_search.cgi?taxid=9598). The neighboring 2 Mb of sequence 5′ and 3′ to each chimeric chimpanzee Alu element was analyzed, and the number of genes found within this combined 4 Mb were noted. All computer programs used are available from the authors upon request.
The gorilla and orangutan DNA sequences generated during the course of this study have been deposited in GenBank (http://www.ncbi.nlm.nih.gov/Genbank) under accession numbers EF682150–EF682182. The GenBank accession numbers for the three HTR3D isforms discussed in this article are NM_182537, BC101090, and AJ437318. |
10.1371/journal.pntd.0003811 | A Pre-clinical Animal Model of Trypanosoma brucei Infection Demonstrating Cardiac Dysfunction | African trypanosomiasis (AT), caused by Trypanosoma brucei species, results in both neurological and cardiac dysfunction and can be fatal if untreated. Research on the pathogenesis and treatment of the disease has centred to date on the characteristic neurological symptoms, whereas cardiac dysfunction (e.g. ventricular arrhythmias) in AT remains largely unstudied. Animal models of AT demonstrating cardiac dysfunction similar to that described in field cases of AT are critically required to transform our understanding of AT-induced cardiac pathophysiology and identify future treatment strategies. We have previously shown that T. brucei can interact with heart muscle cells (cardiomyocytes) to induce ventricular arrhythmias in ex vivo adult rat hearts. However, it is unknown whether the arrhythmias observed ex vivo are also present during in vivo infection in experimental animal models. Here we show for the first time the characterisation of ventricular arrhythmias in vivo in two animal models of AT infection using electrocardiographic (ECG) monitoring. The first model utilised a commonly used monomorphic laboratory strain, Trypanosoma brucei brucei Lister 427, whilst the second model used a pleomorphic laboratory strain, T. b. brucei TREU 927, which demonstrates a similar chronic infection profile to clinical cases. The frequency of ventricular arrhythmias and heart rate (HR) was significantly increased at the endpoint of infection in the TREU 927 infection model, but not in the Lister 427 infection model. At the end of infection, hearts from both models were isolated and Langendorff perfused ex vivo with increasing concentrations of the β-adrenergic agonist isoproterenol (ISO). Interestingly, the increased frequency of arrhythmias observed in vivo in the TREU 927 infection model was lost upon isolation of the heart ex vivo, but re-emerged with the addition of ISO. Our results demonstrate that TREU 927 infection modifies the substrate of the myocardium in such a way as to increase the propensity for ventricular arrhythmias in response to a circulating factor in vivo or β-adrenergic stimulation ex vivo. The TREU 927 infection model provides a new opportunity to accelerate our understanding of AT-related cardiac pathophysiology and importantly has the required sensitivity to monitor adverse cardiac-related electrical dysfunction when testing new therapeutic treatments for AT.
| African trypanosomiasis (AT) is a disease caused by the single-celled protozoan parasite Trypanosoma brucei. In humans, AT causes neurological problems including sleep disturbances, which give the disease its colloquial name of “sleeping sickness”. Much of the focus of AT research has been on the neurological deficits, but other major organs are also affected, including the heart. Previous studies in humans and animals with AT have identified heart abnormalities such as contractile dysfunction, ventricular arrhythmias and significant cardiac tissue inflammation post-mortem. The need for studies investigating AT-induced cardiac dysfunction is critically important. There is a growing cardiovascular disease (CVD) burden in the population of Sub-Saharan Africa and CVD is now considered the leading cause of death in those aged over 45. AT-induced cardiac dysfunction has the potential to compound poor outcome in patients with pre-existing CVD and those being treated with existing AT medications due to their inherent adverse cardiac effects. As a consequence, new treatments for AT require evaluation in appropriate pre-clinical animal models to ensure that they do not detrimentally alter cardiac function. This study established two rat models of AT infection in which an electrocardiogram (ECG) was used to study cardiac electrical function. The first model utilised a commonly used laboratory strain, Trypanosoma brucei brucei Lister 427, whilst the second model used a laboratory strain that demonstrates a similar chronic infection profile to clinical cases, T. b. brucei TREU 927. The TREU 927 infection model showed a significantly increased frequency of cardiac arrhythmias in vivo. The arrhythmias observed in vivo in the TREU 927 infection model were lost upon isolation of the heart ex vivo, but returned with the addition of the β-adrenergic agonist—isoproterenol (ISO). These data suggest that AT alters the myocardium in such a way as to increase the propensity for ventricular arrhythmias when in the presence of a circulating factor or β-adrenergic stimulation (stress). Importantly, this animal model has sufficient sensitivity to measure changes in the frequency of AT-induced ventricular arrhythmias. Therefore the model is of value for studies aiming to develop new treatment strategies to specifically minimise AT-related cardiac dysfunction or to assess the presence of adverse cardiac electrical abnormalities when testing new trypanocidal-based AT treatments.
| Human African trypanosomiasis (HAT) is caused by the Trypanosoma brucei sub-species T. b. gambiense and T. b. rhodesiense, which are transmitted by the tsetse fly vector (Glossina spp.). Infection with these parasites leads to both neurological and cardiac dysfunction and can be fatal if untreated. The rate of disease progression and severity of clinical signs is dependent upon a number of factors including parasite species and strain. Whilst the neurological-related pathogenesis of African Trypanosomiasis (AT) infection has been an understandable research focus leading to its extensive characterisation [1–3], the pathogenesis associated with cardiac dysfunction is poorly understood. A better understanding of the cardiac-related pathogenesis of the disease is required because it is unknown: 1) how AT infection leads to cardiac pathology; 2) whether infection level or stage alters the severity and progression of cardiac dysfunction; 3) if the toxicity of drugs used to treat HAT could be exacerbated by AT-induced cardiac pathology; and 4) if direct treatment of the cardiac pathology can improve overall patient outcome.
HAT manifests clinically in two stages: Stage I, the haemolymphatic stage; and Stage II, the meningo-encephalitic stage where the parasites extravasate into organs including the brain [4–6] and heart [7;8]. Post-mortem studies clearly demonstrate that cardiac pathology occurs in 70–100% of HAT patients [7;9] and in experimental animal models of the disease [10–14]. The observed cardiac pathologies include myocarditis [9] with mononuclear inflammatory cell infiltration and fibrosis [7;8], both of which can lead to ventricular dysfunction and heart failure. In a study of sixteen HAT deaths from T. b. gambiense, Adams et al. (1986) found that two of the deaths were the result of pulmonary oedema from cardiac failure [7]. Furthermore, radiographs from two separate studies demonstrated an enlargement of the cardiac silhouette to >50% of the thoracic diameter (a potential indicator of cardiac failure) in 34% (of 118 [15]) and 44% (of 25 [16]) of patients with HAT (reviewed in Blum et al. 2008 [17]). Echocardiography of HAT patients is not commonly assessed, but Tsala et al. (1988) in a study of 25 HAT patients, identified potential indicators of heart failure including right-ventricular dilatation in 64% of HAT patients and pericardial effusion in 12% [16]. Fouchet et al. (1968) [18] and Bertrand et al. (1974) [15] have both reported cardiac conduction abnormalities, including type I Atrio-Ventricular (AV) block in 3.7% and 14% of patients examined and type II AV block in 1% and 2.5% of patients, respectively [15;18;19]. Conduction abnormalities can lead to disturbances in heart rate and rhythm, leading to impaired delivery of blood to the tissues. Palpitations resulting from ventricular arrhythmias called ventricular premature complexes (VPCs) are indicators of heart dysfunction. HAT patients have significantly more palpitations than controls [19]. A number of studies demonstrated a clear involvement of the cardiac pathology in death [7;15;20;21], but because of the logistical difficulties of performing significant field studies in the regions involved, understanding the true extent to which this occurs in HAT is currently limited. A more recent large scale study that collected significant quantities of cardiac data from T. b. gambiense HAT patients observed a high prevalence of cardiac electrical abnormalities in ~55% of 406 stage I HAT patients and ~71% of 60 stage II HAT patients [19;22]. This study also demonstrated a significant proportion of patients with: (i) prolonged QT interval which can lead to fatal arrhythmias; and (ii) increased levels of the peptide NT- proBNP, an indicator of left ventricular dysfunction [19]. The extent of electrical abnormalities reported in patients with HAT in the study by Blum et el (71%) [19] is in striking agreement with the proportion of patients with cardiac pathology observed at post-mortem [7;9]. In addition to T. b. gambiense HAT patients, ECG abnormalities have also been identified in 55% (22 of 40) of patients with East African trypanosomiasis caused by T. b. rhodesiense (unclassified for stage I or II disease) [23].
Besides the inevitable inflammatory response to trypanosomes that occurs in heart tissues, we have previously demonstrated that T. brucei directly interact with heart muscle cells (cardiomyocytes) to induce ventricular arrhythmias (VPCs) in ex vivo hearts independent of a systemic immune/inflammatory response [24]. However, it is unknown whether the arrhythmias observed in rat hearts ex vivo are also present during infection in vivo.
The main aim of the current study was to develop an animal model of AT infection to assess whether ventricular arrhythmias—in particular VPCs—could be identified in vivo through electrocardiographic (ECG) assessment. The data suggest for the first time that AT infections create an arrhythmogenic substrate in the heart, which when triggered by circulating factor/s associated with β-adrenergic stimulation, increase the propensity for ventricular arrhythmias.
T. b. brucei Lister 427 and T. b. brucei TREU 927 were grown separately for one passage in ICR mice for 2–3 days to adapt them to in vivo conditions, from being culture adapted in the case of T. b. brucei Lister 427 [25] and from cryopreservation for T. b. brucei TREU 927. When parasitaemia reached 1 x 108 parasites.mL-1 [26] the mice were euthanased and blood collected in Carter’s balanced salt solution (CBSS; mM; 25 HEPES, 120 NaCl, 5.4 KCl, 0.55 CaCl2, 0.4 MgSO4, 5.6 Na2PO4, 11.1 glucose, pH 7.4) containing 100 U/ml heparin. The trypanosomes were diluted to 1 x 105 parasites in a 200 μL volume of CBSS (approx. 1:200 dilution from 1 x108 parasites.mL-1) under sterile conditions. The 200 μL parasite suspension was prepared in a 1 mL syringe for injection. Matching volumes of CBSS were prepared as control injections. All mice and rat animal procedures were approved by the University of Glasgow Ethical Review Panel and licensed by the Home Office, UK (Project Licence Number 600/4503).
The care and use of animals was in accordance with the UK government Animals (Scientific procedures) Act 1986 (ASPA). Animals used were adult male Wistar rats (250–300g) with a 7 day acclimatisation period upon delivery with a 12 hr light/dark cycle. Animals were kept at the Cardiovascular Research Unit, University of Glasgow in a dedicated room licensed under the Specified Animal Pathogens (Scotland) Order, 2009 (SAPO).
Anaesthesia was induced by placing the animal in a pre-filled induction chamber with 5% isoflurane (Isoflo, Abbot Laboratories, USA) in 100% O2 at 1 L.min-1 until loss of righting and flexor withdrawal reflexes. The rats were then maintained on isoflurane delivered via facemask at 1–1.5% isoflurane in 1 L.min-1 O2. The lead II ECG was recorded via the placement of sterile intradermal electrodes. The placement sites on the rat were caudal aspects of the left and right carpi and the medial aspects of left and right crura. To ensure reproducibility for the same rats and between rats, all animals were positioned identically. The ECG was recorded for 15 min with an IWX228 bioamplifier and LabScribe 2 software (iWorx) at a sampling rate of 2.0 KHz. Rats were then injected via the intraperitoneal route with T. b. brucei Lister 427 or vehicle and then recovered.
The ECG data from the last 1 min of the 15 min period collected on both day 0 and day 4 were averaged using the advanced ECG analysis module (iWorx) and exported to Origin6.1 (OriginLab) for RR interval, heart rate, PR interval and QT interval measurement, with correction for heart rate using the Framingham method [27]. The whole trace was manually assessed for ventricular premature complexes (VPCs) as defined by the Lambeth Conventions [28]; specifically complexes that were premature, without a well-defined P wave and had a wide and bizarre shape.
Parasitaemia levels were measured daily by microscopy of blood from superficial venepuncture of the lateral tail vein, as previously described [26]. A parasitaemia level exceeding 5.0 x 108 parasites.mL-1 for more than two consecutive days was set as a cut-off point for the welfare of the animals. No animal exceeded this level during the study. Day 4 of the Lister 427 model was selected as the end point based on prior experience with this trypanosome strain in the rat infection model in terms of the level of parasitaemia and its impact on the health of the animals and in order to avoid sudden deaths.
For the longer TREU 927 infection model, CA-F40 biopotential recording devices (Data Sciences International) were implanted into rats to measure the ECG in conscious animals. This allowed the recording and analysis of ECG data without the potential cardiovascular depressive effects of isoflurane [29;30]. Briefly, adult male Wistar rats (250–300 g) were anaesthetised with isoflurane delivered in 100% O2. Animals were positioned in ventral recumbency on a heated pad. Peri-operative analgesia of 5.0 mg.kg-1 carprofen (Rimadyl, Pfizer Animal Health) was administered in 5.0 mL of 0.9% sterile saline subcutaneously. The telemetry device was implanted subcutaneously in the dorsal thoracic region. Tracts were tunnelled under the skin from the implant site to the right pectoral and xyphoid regions ventrally. The insulation was removed from the distal 5.0 mm of each ECG lead, which were then affixed to the layer of muscle at each point under the skin with 1.5 metric nylon suture (Johnson & Johnson). Exposed ends of the leads were re-covered with the removed insulating material. The surgical incisions were closed with 1.5 metric polyglactan 910 (Vicryl, Johnson & Johnson).
The animals were allowed to recover from the surgical procedure for 1 week before they were infected with 1.0 x 105 T. b. brucei TREU 927 in 200 μL CBSS via intraperitoneal injection. Control rats were injected with the same volume of CBSS. Parasitaemia levels by superficial tail venepuncture were assessed as previously described.
The implanted probes were activated magnetically and data was collected from receiver pads underneath the animals’ usual cages. Telemetry signals were relayed via a data exchange matrix to a computer loaded with the acquisition software Dataquest™ OpenART v4.2 (Data Sciences International). Raw ECG data was collected at a 2.0 kHz sampling frequency. Files were exported to Ponemah v4.8 (Data Sciences International) for analysis. ECG sections of 30 min were averaged and assessed for RR interval, heart rate, PR intervals, and QT intervals corrected for heart rate using the Framingham method (QTc = QT + 0.154 x (1-RR)). A one hour section of trace from day 3, 6, 9 and the end of the final day of the protocol—to enable analysis of time points corresponding with peak and trough parasitaemias—was assessed for arrhythmias using the Lambeth Conventions as described above and compared to a one hour section of trace after recovery but prior to infection (day 0).
Both control and infected animals for each model were sacrificed by cervical dislocation while under a brief exposure of low dose 1% isofluorane. Heart, liver and spleen were removed, organ mass recorded and compared to the tibial length defined as the length from lateral femoral epicondyle proximally to the lateral malleolus distally to control for individual size and mass.
The hearts removed from both groups (Lister 427 infection model and TREU 927 infection model, plus controls) were immersed in ice-cold Tyrodes solution (mM); 116 NaCl, 20 NaHCO3, 0.4 Na2HPO4, 1.0 MgSO4-7H2O, 4.0 KCl and 11.0 D-glucose. The solution was bubbled with 95% O2 / 5% CO2 for 15–20 min to oxygenate and buffer before CaCl2 was added to a concentration of 1.8 mM. Extraneous tissue was carefully dissected away to reveal the aorta. The hearts were blotted dry and quickly weighed before cannulation to a Langendorff perfusion apparatus. The hearts were perfused with the above Tyrodes solution at 10 mL.min-1 and immersed in a water-jacketed chamber filled with Tyrodes solution at 37°C. Electrodes were placed in the chamber in close approximation with the right atrium for the negative electrode and the apex of the left ventricle for the positive electrode and the pseudo-ECG recorded. Hearts were perfused for a period of 15 min (steady state) followed by 15 min periods each with the β-adrenergic agonist isoproterenol at concentrations of 100 nM, 1 μM, 10 μM and 100 μM.
ECG data were collected using the ETH-256 bioamplifier (iWorx) and LabChart 7 (ADInstruments) software at a sampling rate of 2.0 kHz. The ECG from the last 5 min of each 15 min period was averaged using the advanced ECG analysis module of the programme and the RR interval, heart rate, PR interval and the QT interval with correction for heart rate using the Framingham method [27] were measured. Traces were exported to Origin6.1 (OriginLab). The entirety of the traces was manually assessed for arrhythmic events according to the Lambeth Conventions [28] and recorded as the frequency.min-1.
After Langendorff perfusion, whole hearts were fixed in 10% neutral buffered formalin and sectioned to include a transverse section of the apex and one longitudinal section of the base (to include the atria) and the apex (3 sections from each heart). These were then routinely processed to paraffin wax and serial sections (5 μm) were stained with haematoxylin and eosin (H&E) for histopathological evaluation of inflammation and Picrosirius red for fibrosis. Selected sections were stained with Giemsa to highlight parasitic organisms. The sections were then semi-quantitatively scored and assigned a score of 0–3 for degree of inflammation and fibrosis. The scores of 0–3 were based on the findings of no pathology, mild, moderate and severe changes respectively. In addition, the presence (1) or absence (0) of parasites and their location (e.g. interstitial) was noted.
For the in vivo Lister 427 infection model, the mean ± SEM data from the last 1 min of ECG trace for the same animal from day 0 and day 4 were compared with a paired Student’s T-test. Comparisons of in vivo ECGs between control and infected animals were conducted with a two-sample Student’s T-test. For the in vivo TREU 927 infection model, mean ± SEM ECG telemetry data were taken from the last 30 min of ECG trace from day 0 and day 11 and compared within each animal by paired Student’s T-test. Comparisons between control and infected animals were performed by two-sample Student’s T-test. Arrhythmia data for the TREU 927 infection model were taken from 1 hr long traces at days 0, 3, 6, 9 and the end and compared to day 0 by ANOVA. Mean ± SEM ECG data was acquired from the average of the last 5 min of each 15 min section of trace for each concentration of ISO for the ex vivo Langendorff perfused hearts of both the Lister 427 and TREU 927 infection models and analysed using ANOVA. P<0.05 was taken to be statistically significant.
The average ECG traces from 15 min periods from days 0 and 4 from anaesthetised rats in both infection and control groups were analysed for heart rate (HR), PR interval and QT interval (Fig 1A). The Lister 427 infection model was of short duration—animals had to be euthanased on welfare grounds at 4 days post-inoculation. The relatively brief infection period was due to parasitaemia of the infected rats demonstrating exponential growth (rising to 2.51 x 108 ± 1.02 x 108 parasites.mL-1 on day 4; Fig 1B), as was anticipated with this monomorphic trypanosome strain. There was no significant difference in the HR in the control animals between days 0 and 4 (99.8 ± 8.6% of day 0; 249 ± 21 vs. 248 ± 21 bpm; Day 0 vs. Day 4; P>0.05), nor was there a significant difference in the infected animals (94.8 ± 7.9% of day 0; 238 ± 19 vs. 226 ± 19; Day 0 vs. Day 4; P>0.05; Fig 1C(i)). There was no significant difference in PR interval for the control rats (46.6 ± 0.8 vs. 46.4 ± 1.2 ms; Day 0 vs. Day 4; P>0.05; Fig 1C(ii)), nor for the infected rats (46.4 ± 0.6 vs. 45.8 ± 0.8 ms; Day 0 vs. Day 4; P>0.05; Fig 1C(ii)). There was also no significant difference in QTc for control (176.6 ± 3.0 vs. 179.1 ± 2.8 ms; Day 0 vs. Day 4; P>0.05; Fig 1C(iii)) or infected animals (181.3 ± 3.9 vs. 176.2 ± 3.9 ms; Day 0 vs. Day 4; P>0.05; Fig 1C(iii)). During the 15 min periods of ECG for both the start and end of the study protocol no ventricular arrhythmic events were observed.
An ECG was performed on perfused hearts isolated from infected animals to determine: (1) whether cardiac electrical activity was altered in the absence of a systemic autonomic nervous response or circulating factors and (2) the effect of β-adrenergic stimulation on ECG parameters using isoproterenol (ISO) at increasing doses (100 nM, 1 μM, 10 μM and 100 μM) every 15 min to simulate a stress response (Table 1 and Fig 2A).
When hearts were isolated and perfused ex vivo, no ECG parameters (HR, PR, QTc intervals) were significantly different between control and infected animals (Timepoint 0; Table 1 and Fig 2). HR of control animal hearts (normalised to no ISO) demonstrated a non-significant increase to 117% of no ISO level (100 ± 8.0 vs. 117 ± 5.0%; no ISO vs. 100 μM ISO; P>0.05; Fig 2B(i and ii) and Table 1). However, hearts from infected animals demonstrated a significant increase in HR to 135% of HR with no ISO (100 ± 8.0 vs. 135 ± 2.0%; no ISO vs. 100 μM ISO; P<0.05; Fig 2B(i and ii) and Table 1). PR or QTc interval (raw data and data normalised to no ISO) showed no significant difference between control or infected animals (Table 1 and Fig 2C and 2D respectively).
A limitation of the use of T. b. brucei Lister 427 in vivo is that the parasite continues to divide exponentially within the host’s bloodstream until the death of the host, typically after approximately 4 days in the rat model. Lister 427 is a monomorphic strain of T. b. brucei, meaning it remains as the mitotically dividing long slender form [31]. While this continual growth makes Lister 427 a useful in vitro model organism, it does not mimic the fluctuating parasitaemia and more chronic infections [32] observed in human patients and animals infected with trypanosomes, which is due to the pleomorphism of non-laboratory adapted trypanosomes. Wild-type pleomorphic trypanosomes such as TREU 927 terminally differentiate to short stumpy forms under a density-dependent trigger [33] in preparation for transmission to the insect vector [34;35]. The animals were monitored for health and their parasitaemia was measured daily for the duration of the study (Fig 3A). The parasitaemia in the T. b. brucei TREU 927 infected animals followed the classical fluctuating pattern with a first peak parasitaemia of 3.16 x 107 ± 2.51 x 105 parasites.mL-1 of blood on day 6, before becoming undetectable by microscopy, followed by a second peak of 3.56 x 107 ± 1.62 x 107 parasites.mL-1 blood on day 10 (Fig 3B).
The ECG was recorded in both control rats and rats infected with T. b. brucei TREU 927 via biopotential recording telemeters. Average ECG traces were analysed from day 0 and day 11 (Fig 3A). HR tended to decrease, but not significantly, in the control animals (404 ± 18 vs. 348 ± 12 bpm; Day 0 vs. Day 11; P>0.05). However, there was a significant increase in heart rate for TREU 927 infected animals of 117.8 ± 5.0% of day 0 (351 ± 22 vs. 413 ± 18 bpm; Day 0 vs. Day 11; P<0.05; Fig 3C(i)). There was no significant difference in PR interval for control (43.2 ± 1.7 vs. 42.6 ± 1.5 ms; Day 0 vs. Day 11; P>0.05; Fig 3C(ii)) or infected animals (41.5 ± 1.7 vs. 44.1 ± 3.2 ms; Day 0 vs. Day 11; P>0.05; Fig 3C(ii)). When the QT interval was corrected for HR using the Framingham method there was no significant difference for control animals (181.9 ± 4.4 vs. 190.1 ± 11.8 ms; Day 0 vs. Day 14; P>0.05; Fig 3C(iii)) or infected animals (185.2 ± 7.2 vs. 192.6 ± 6.9 bpm; Day 0 vs. Day 14; P>0.05; Fig 3C(iii)). The frequency of arrhythmic events including VPCs, did not significantly change in control animals (0.027 ± 0.019 vs. 0.047 ± 0.017 VPC.min-1; Day 0 vs. Day 11; P>0.05; Fig 3D(i)). However, during progression of infection with T. b. brucei TREU 927, when the arrhythmia frequency was examined at 3 day intervals to take into account both peaks and troughs of parasitaemia, the frequency increased over the course of the model to a final significant increase of 442% of Day 0 levels at Day 11 (0.113 ± 0.039 vs. 0.500 ± 0.234 VPC.min-1; Day 0 vs. Day 11; P<0.05; Fig 3D(i and ii)). Arrhythmia frequency did not parallel parasitaemia level since the arrhythmia frequency did not concomitantly reduce with the parasitaemia when examined at Day 9 (0.307 ± 0.199 VPC.min-1; Day 9; Fig 3D(i)).
As with the Lister 427 infection model, hearts were isolated from the TREU 927 infection model animals and an ECG performed in the presence and absence of ISO (Table 2 and Fig 4A). When hearts were isolated and perfused ex vivo, no ECG parameters (HR, PR, QTc intervals) were significantly different between control and infected animals (Timepoint 0; Table 2). The control animal HR (normalised to no ISO), demonstrated a significant increase to 117% of no ISO level (100 ± 6.0 vs. 117 ± 7.0%; no ISO vs. 100 μM ISO; P>0.05; Fig 4B(i and ii) and Table 2) compared to an increase to 149% of no ISO level for infected hearts (100 ± 5.0 vs. 149 ± 8.0; no ISO vs. 100 μM ISO; P<0.05; Fig 4B(i and ii) and Table 2). The PR interval (normalised to no ISO), demonstrated a significant difference between control and infected hearts but only at 1 μM ISO (100 ± 9.0 vs. 80.2 ± 4.9%; no ISO vs. 1μM ISO; P<0.05; Fig 4C(i & ii) and Table 2). No significant changes in the QTc (when normalised to no ISO) were observed (Fig 4D(i & ii) and Table 2).
The pseudo-ECGs from the Langendorff perfused hearts from both models were also assessed for the frequency of VPCs with and without the presence of ISO (Fig 5A). There was no significant increase in VPC frequency in control hearts or the Lister 427 infection model hearts (Fig 5B and 5D(i)) both in the absence and presence of ISO. However, the TREU 927 infection model hearts demonstrated an increase in VPC frequency in the presence of ISO compared with control hearts (significant at 1 μM) (0.24 ± 0.16 vs. 5.24 ± 4.99 vs. 5.48 ± 1.83 (P<0.05) vs. 10.4 ± 8.9 VPC.min-1; no ISO vs. 100 nM vs. 1 μM vs. 10 μM ISO; Fig 5C and 5D(ii)) but not in the absence of ISO.
Histopathological evaluation of the hearts from both control (Fig 6A(i)) and infection models (Fig 6A(ii)) subsequent to Langendorff perfusion revealed the presence of parasites within the interstitium (parasite extravasation) in close proximity to cardiomyocytes in 3 out of 3 (100%) of the TREU 927 infection model hearts but in 0 out of 4 hearts (0%) from the Lister 427 infection model and 11 control hearts (Fig 6A(iii) and 6B(i)). The level of inflammation was higher in hearts from the TREU 927 infection model compared to control and Lister 427 infected hearts (Fig 6A(iv) and 6B(ii)). There was no significant change in the degree of fibrosis between any group of hearts (Fig 6(iii)).
When animals were sacrificed, the mass/tibial length ratio of the heart, liver and spleen were recorded. Heart and liver mass/tibial length ratios were not significantly different between infected animals and controls for either the Lister 427 infection model or the TREU 927 infection model (see S1A(i–v) Fig and S2(i–iii) Fig However, the spleen mass/tibial length ratios were significantly increased in both the Lister 427 infection model by 199% (13.3 ± 0.4 vs. 26.5 ± 3.4 mg.mm-1; control vs. infected; P<0.05; S1(v) and S1B Fig) and the TREU 927 infection model by 292% (17.2 ± 1.6 vs. 50.2 ± 7.5 mg.mm-1; P<0.05; S2A(iv) Fig).
We have previously demonstrated that even in the absence of a host systemic inflammatory/immune response to the parasite, ventricular arrhythmias can be induced by Lister 427 trypanosomes ex vivo due to the effect of a secreted/excreted protease, T. brucei cathepsin-L (TbCatL), on cardiomyocytes [24]. However, no significant ECG abnormalities were apparent between the control and infected animals in vivo. Four explanations may account for this. First is the use of anaesthesia in the Lister 427 model compared with conscious ECG monitoring in the TREU 927 model. It is possible that the anaesthesia (isofluorane) used in the Lister 427 model reduced the propensity for arrhythmias, but to the best of our knowledge there is little evidence that isofluorane can reduce the incidence of arrhythmias. Secondly, the brevity of the infection period may limit appreciable extravasation of the parasite into the cardiac extracellular matrix. This possibility is supported by histological examination of Lister 427 infected hearts, in which no parasites could be identified in the interstitium, suggesting that extravasation had not occurred with this strain—at least within the limits of detection. Connected to the brevity of the infection period, exposure time of cardiomyocytes to parasites and parasite products in the Lister 427 model is clearly much less than that of the TREU 927 model. However, it is important to note that the parasite burden within Lister 427 infected animals is much greater than the TREU 927 model (2.51 x 108 ± 1.02 x 108 parasites.mL-1 compared to 3.56 x 107 ± 1.62 x 107 parasites.mL-1, respectively, at peak parasitaemia). On balance therefore it is difficult to conclude that the brevity of infection period leading to reduced exposure time of cardiomyocytes to parasites and parasite products exclusively explains an absence of arrhythmias in the 427 infection model. Thirdly, Lister 427 does not readily extravasate without existing tissue damage from an inflammatory processes [36]. In the current study, histological examination confirmed that the level of inflammation in Lister 427 hearts was not significantly different from that in control hearts. Finally, it is probable that the amount of TbCatL produced by Lister 427 is significantly lower than that produced by pleomorphic trypanosomes such as TREU 927. Caffrey et al. (2001) demonstrated that short-stumpy form trypanosomes produce almost five times the amount of TbCatL compared to long-slender forms [37]. Since the Lister 427 strain does not differentiate and remains as a long-slender form, the quantity of TbCatL produced is very likely to be lower and contribute to the lack of response observed.
During pathophysiological stress, there are increased levels of circulating catecholamines (adrenaline and noradrenaline). In a previous study, ex vivo hearts demonstrated a significantly higher frequency of ventricular arrhythmias (ventricular premature complexes, VPCs) in response to trypanosome culture supernatant and isoproterenol (ISO; a β-adrenergic agonist) [24]. This effect was mediated by a CaMKII-dependent increase of spontaneous sarcoplasmic reticulum (SR)-mediated Ca2+ release events (Ca2+ waves) [24;38]. In the current study, ex vivo hearts from the Lister 427 infection model demonstrated no increase in arrhythmic events but a significantly increased HR in response to ISO. HR is dictated by the rate that sinoatrial nodal pacemaker cells (SANC) in the heart fire spontaneous action potentials. It is possible that the Lister 427 infection model induces a higher HR due to: 1) a greater sensitivity of the sinoatrial nodal pacemaker cells (SANC) to β-adrenergic stimulation as can happen in heart failure [39]; 2) altered expression/heterodimerisation/polymorphisms of β1 adrenergic receptors [40;41] in SANC; 3) altered expression of ion channels governing SANC pace making such as hyperpolarisation and cyclic nucleotide (HCN) channels [42], G-protein-activated inwardly rectifying potassium (GIRK) channels [43] or calcium channels [44]; 4) altered rate of SANC action potential firing due to changes in Ca2+ handling protein activity [44;45]. CaMKII-mediated phosphorylation of Ca2+ handling proteins may alter HR via modulation of the rate of action potential firing from SANCs [45]. The possibility that the observed HR change in the current study may be a CaMKII-mediated effect on Ca2+ handling in SANCs in response to trypanosomes warrants future investigation. The implications of a stress-induced increased HR during trypanosome infection are important as this may in turn lead to an increased metabolic demand on the heart at the expense of the other organs [46;47]. Given that patients with AT can die from multiple organ failure [32] this increased metabolic demand is potentially significant.
The PR interval was not significantly altered by infection with Lister 427, but tended to decrease at 1 μM ISO in control animals while remaining unchanged in infected animals. A failure for the PR interval to shorten under β-adrenergic stimulation may reflect a decreased ability of atrio-ventricular (AV) node conduction velocity to respond to β-adrenergic stimulation (1st degree AV block) [48]. A PR interval of >200 ms has been reported in 3.7–14% of HAT patients and was defined as 1st degree AV block [15;18]. However, a separate study Blum et al. (2007) did not observe a significant change in PR interval in patients with HAT [19]. It is therefore unclear whether trypanosome infection consistently results in altered AV node conduction velocity.
The QTc was not significantly affected by infection with Lister 427. A prolongation of QTc during clinical infection may be attributed to myocardial inflammatory infiltration leading to altered electrical conductivity [9;12]. The cause of the QTc prolongation identified by Blum et al. (2007) was not investigated but ascribed to the historical observation of myocardial inflammatory infiltration in patients with HAT [19].
The pleomorphic strain T. b. brucei TREU 927 was used to more closely approximate a natural infection. This TREU 927 infection model demonstrated the classical undulating parasitaemia phenotype (Fig 3B) and was sustainable over a longer time period (up to 11 days) compared to the four-day infection model of Lister 427. Implanted biopotential recording devices had the additional advantage of avoiding any cardio-depressive effect of anaesthesia while assessing ECG parameters. In parallel with the Lister 427 infection model, the PR and QTc interval was not significantly altered in the TREU 927 infection model. However, the normalised HR of the TREU 927 infected animals in vivo was significantly increased. It is unknown whether the anaesthetic regime used during ECG assessment in the Lister 427 infection model masked an increased HR. Interestingly, a significant increase in the frequency of ventricular arrhythmias (VPCs) with the TREU 927 infection model was also observed in vivo. The change in arrhythmia frequency increased over the course of the model independent of the parasitaemia level since the arrhythmia frequency remained increased when the parasitaemia reduced after the first peak. This provides further evidence that African trypanosomes increase the propensity for ventricular arrhythmias in vivo.
As with the Lister 427 infection model, increasing concentrations of ISO elevated HR to a greater degree in TREU 927 infected hearts. This supports an effect of trypanosome infection on the response of the SANCs to ISO. Interestingly, HR in vivo was significantly increased in the TREU 927 infection model (Fig 3C(i)), an effect lost upon removal of the heart ex vivo, but reintroduced with the addition of ISO. It is possible that a circulating factor (which ISO simulates) underlies the effect on HR observed in the TREU 927 infection model in vivo.
In parallel with the Lister 427 infection model, the PR interval tended to decrease in control animals with a significant decrease in control animals at 1 μM ISO compared to no ISO, and was significantly reduced compared to PR intervals in infected animals at 0.1, 1 and 10 μM ISO. Although this suggests an effect on AV node conduction velocity similar to that reported in a proportion of HAT cases [15;18], the lack of PR interval effect in rats in vivo suggests that alteration of AV node conduction velocity may not be a significant feature in the rat in vivo model of trypanosome infection. The QTc was not significantly altered despite an increase in the level of inflammation (Fig 6B(ii)). It may be possible that the levels of inflammation in the TREU 927 infection model were insufficient to induce a prolongation of the QTc interval as observed in patients with HAT, which is understandable given that HAT is a chronic disease of months to years. However, it should be noted that direct comparison of QT intervals between rats and humans should be interpreted with care since different channels are responsible for QT interval control in these species [49].
Increased VPC frequency was observed in the TREU 927 infection model in vivo in parallel with increased levels of inflammation and the presence of extravasated parasites. However, VPCs were absent when the same hearts were isolated ex vivo, only re-emerging in the presence of ISO. These novel data suggest that African trypanosomes create an arrhythmogenic substrate in the heart, which when triggered by circulating factor(s) increases the propensity for ventricular arrhythmias. The generation of an arrhythmogenic substrate in the current study is not caused by a significant increase of fibrosis (a known arrhythmogenic factor) in the TREU 927 infection model (Fig 6). However, the increased levels of inflammation observed and the ability of T. b. brucei TREU 927 to extravasate into the myocardium—both of which occur despite the parasitaemia levels being approximately 7 fold lower than with T. b. brucei Lister 427—are likely contributors to the arrhythmogenic substrate. Importantly, the ex vivo data reveal that these two factors by themselves are not able to lead to arrhythmias without an additional circulating trigger. It is possible that β-adrenergic stimulation such as circulating adrenaline/noradrenaline in vivo is the trigger for ventricular arrhythmias during trypanosome infection and is mimicked by ISO when applied to ex vivo hearts from infected animals. Whether the expected elevated levels of trypanosome-secreted TbCatL (compared with the Lister 427 infection model) in vivo contribute to the trigger via increased CaMKII activity (as would occur with β-adrenergic stimulation [50]) remains unknown and warrants further investigation in future studies.
There are key differences in the infection profiles of the monomorphic and pleomorphic infection models that make direct comparison between the models difficult. The exponential nature of parasite multiplication in the monomorphic Lister 427 model removes the ability to examine infections beyond the first peak of parasitaemia. This biological limitation also results in a lack of representation of parasite life cycle stages (i.e. short stumpy trypanosomes) relevant to infections in the field, resulting in restricted utility of the monomorphic model for exploring cardiac pathogenesis in a meaningful manner. The novel data therefore highlight the requirement for infection models that are capable of reflecting the chronic infection profiles of HAT and AAT infections (e.g. pleomorphic TREU 927) to study HAT-induced cardiac pathophysiology.
The current study has successfully characterised in vivo cardiac electrical dysfunction in two infection models of African trypanosomes. We have demonstrated for the first time that infection with a pleomorphic strain of African trypanosome produces an increased frequency of ventricular arrhythmias and provide evidence that this phenomenon requires a circulating factor in combination with trypanosome infection. This novel animal model of African trypanosomiasis provides a highly attractive platform to further study the cardiac/other organ-related pathophysiology of African trypanosomiasis and the efficacy of therapeutic strategies to treat this neglected disease.
|
10.1371/journal.pntd.0006914 | Uncharted territory of the epidemiological burden of cutaneous leishmaniasis in sub-Saharan Africa—A systematic review | Cutaneous leishmaniasis (CL) is the most frequent form of leishmaniasis, with 0.7 to 1.2 million cases per year globally. However, the burden of CL is poorly documented in some regions. We carried out this review to synthesize knowledge on the epidemiological burden of CL in sub-Saharan Africa.
We systematically searched PubMed, CABI Global health, Africa Index Medicus databases for publications on CL and its burden. There were no restrictions on language/publication date. Case series with less than ten patients, species identification studies, reviews, non-human, and non-CL focused studies were excluded. Findings were extracted and described. The review was conducted following PRISMA guidelines; the protocol was registered in PROSPERO (42016036272).
From 289 identified records, 54 met eligibility criteria and were included in the synthesis. CL was reported from 13 of the 48 sub-Saharan African countries (3 eastern, nine western and one from southern Africa). More than half of the records (30/54; 56%) were from western Africa, notably Senegal, Burkina Faso and Mali. All studies were observational: 29 were descriptive case series (total 13,257 cases), and 24 followed a cross-sectional design. The majority (78%) of the studies were carried out before the year 2000. Forty-two studies mentioned the parasite species, but was either assumed or attributed on the historical account. Regional differences in clinical manifestations were reported. We found high variability across methodologies, leading to difficulties to compare or combine data. The prevalence in hospital settings among suspected cases ranged between 0.1 and 14.2%. At the community level, CL prevalence varied widely between studies. Outbreaks of thousands of cases occurred in Ethiopia, Ghana, and Sudan. Polymorphism of CL in HIV-infected people is a concern. Key information gaps in CL burden here include population-based CL prevalence/incidence, risk factors, and its socio-economic burden.
The evidence on CL epidemiology in sub-Saharan Africa is scanty. The CL frequency and severity are poorly identified. There is a need for population-based studies to define the CL burden better. Endemic countries should consider research and action to improve burden estimation and essential control measures including diagnosis and treatment capacity.
| Cutaneous leishmaniasis (CL) is the most common form of this group of parasitic diseases, transmitted by sandflies. In sub-Saharan Africa, its extent of the problem is unknown, while elsewhere its disfigurement and stigma may cause a severe impact. This study systematically searched the literature to find evidence on the epidemiological data on human CL in this part of the world. Historically, CL has been present for decades in both western and eastern Africa, but unfortunately, in the last decades, the data are irregular and patchy. The estimated burden, relying on detected cases, may only capture part of the true number of cases. This article shows that there is insufficient evidence to have accurate figures; the diversity of the disease, along with poor surveillance have resulted in unprecedented CL outbreaks in the past. Many knowledge gaps remain, and we highlight the importance of improving the current fragmented knowledge by increasing commitments to tackle CL and conduct better population studies. CL in sub-Saharan Africa appears to be a blind spot and should not remain so.
| Cutaneous leishmaniasis (CL) is the most common clinical manifestation of leishmaniasis, a parasitic neglected tropical disease (NTD) [1]. Caused by an obligate intracellular protozoa from the Leishmania species and transmitted by the bite of Phlebotomine sand flies, the clinical presentations of CL include localized skin nodules (often called oriental sores), diffuse non-ulcerated papules, dry or wet ulcers, and, in the mucocutaneous form, extensive mucosal destruction of nose, mouth, and throat. Transmission of CL may involve animal reservoir hosts (e.g., rodents, hyraxes) in zoonotic foci, while anthroponotic CL (where humans are the main parasite reservoir) occurs in urban or periurban settings [2]. Environmental changes in rural contexts such as agricultural activities, irrigation, migration, and urbanization may increase the exposure risk for humans and result in epidemics. Likewise, outbreaks in densely populated cities or settlements have occurred, especially in conflict-affected zones such as Afghanistan or Syria [3,4], in refugee camps and contexts of large-scale forced migration of populations.
Globally, the World Health Organization (WHO) considers CL as endemic in 20 countries in the New World (South and Central America) and in 67 countries in the Old World (southern Europe, Africa, the Middle East, parts of southwest Asia) [5]. Between 700,000 to 1,200,000 CL cases are estimated to occur annually worldwide, with >70% of cases in 2014 reported from Afghanistan, Algeria, Brazil, Colombia, Costa Rica, Ethiopia, the Islamic Republic of Iran, Peru, Sudan and the Syrian Arab Republic [5,6]. Multiple parasite species cause CL: in the Old World, these are L. major, L. aethiopica, L. tropica, and, rarely, the viscerotropic L. donovani (in Sudan), resembling similar a phenomenon more known for L. infantum [7–10]. Though CL is often considered self-healing, the duration varies for different species and can take months, or years [11].
Due to the clinical and epidemiological diversity in CL, its geographic clustering and lack of reliable surveillance data, estimating the CL burden are challenging [12]. The most widely used measure of disease burden known as the Disability Adjusted Life Year (DALY) combines estimated prevalence, incidence, and mortality, with an assigned disability weight for each disease [13]. However, the disability weights are defined using different approaches with regards to the expert panel composition, health state description, and valuation methods [14,15]. The specific stigma and psychosocial distress generated by a non-fatal condition are often overlooked [16,17], although the social impact of CL is potentially severe and has been well-documented [18,19].
Moreover, in sub-Saharan Africa (SSA), not only the disability but also the number of CL cases is largely underestimated. A recent global burden analysis listed 19 countries in SSA in the top 50 high burden countries [20]. The passive epidemiological surveillance system that prevails in these countries leads to the patchy data from this region. According to WHO, only Sudan and Ethiopia reported cases of CL [21]. The objective measures of burden such as prevalence and incidence of CL are scarce in this region, making it hard to advocate for funding and resources to tackle the disease.
Whereas attention has been given to CL in Northern Africa (Algeria, Libya, Morocco, Tunisia, Egypt) and the Middle East [22–24], data for sub-Saharan Africa is critically lacking, particularly in countries where CL is not a notifiable disease. This study focuses on SSA because it is a blind spot on the CL epidemiological burden map and the overall picture of what has been documented on CL is not known. We undertook a systematic review of the literature to synthesize current knowledge on CL burden in SSA.
We searched the following electronic databases: National Library of Medicine through Pubmed, Cochrane Register, Web of Science, CABGlobal Health, African Index Medicus and Google Scholar. We did an initial keyword search and subsequent searches based on Medical Subject Headings (MeSH) with various combinations of search terms “cutaneous leishman*” AND “Africa, South of the Sahara” (which also included “Africa, Western”; “Africa, Eastern”; and “Africa, Southern”) OR “Leishmaniasis, cutaneous” OR “Leishmaniasis, diffuse cutaneous” OR “Leishmaniasis, mucocutaneous” AND each individual sub-Saharan countries. The World Bank classification was used to define sub-Saharan African countries and to group them according to the region (i.e., southern, eastern, western, and middle Africa- see Box 1). No language restrictions were set for searches, while we limited the publication date until 31 May 2018. We hand-searched the reference lists of all recovered studies for additional references. We also explored and summarized information from the Global Health Observatory for leishmaniasis maintained by WHO for CL.
We included studies if they are reporting primary data that help to determine the burden of CL in countries in SSA. The burden is defined as elements of 1) severity of the problem (clinical, disability, case fatality,…) in human patients; 2) frequency (prevalence, incidence,…) and 3) economic cost (from patient, societal or health system perspective). We excluded animals or vector studies, studies on pathogenesis, immunology, histopathology, or on Leishmania species only, studies on diagnostic tests or treatment for CL and cases of Post Kala Azar Dermal Leishmaniasis (PKDL)–skin sequelae of VL. Case reports and case series of fewer than ten patients were also excluded. Sub-Saharan Africa as the main geographical interest refers to the settings where the studies were performed/conducted. Reviews about CL in a specific country or region without original data were excluded.
The systematic review was conducted in line with PRISMA guidelines [25,26]. The review protocol was registered in PROSPERO, an international prospective register of systematic reviews, in July 2016, number 42016036272 [27].
We selected the articles in a two-step process. In a first stage, titles and abstracts of all retrieved records were independently reviewed by two investigators (TS and KV). In a second stage, the selected full-text articles were again reviewed (by TS, KV, and a third person) for eligibility. When full-text articles were excluded, the reason for exclusion was registered and reported. Any discordances were resolved through discussion or seeking consensus with a third investigator (MB).
The data were extracted in parallel by two independent readers, using a specific data form, including information on the published record (year, author), setting (country), aim, study design, and main outcomes. We sought data on prevalence or incidence of CL among patients in health facilities and the community; demographic and clinical characteristics of CL patients, and the association between CL and other morbidities, notably Human Immunodeficiency Virus (HIV). We attempted to use the STROBE checklist (for reporting epidemiological studies) to assess the ‘risk of bias,’ but could not continue due to a large number of historical studies that are not in line with current reporting standards. The data analysis thus resulted in a narrative, qualitative synthesis of the included studies.
The flow diagram in Fig 1 shows the selection process: we identified 340 published articles, and after removing duplicates, we screened the title and abstracts of 289 articles, and exclude 184. The full-text articles of the remaining 105 were assessed for eligibility, after which a further 51 were excluded. The remaining 54 articles were included. (See Supporting Information 1 for all the included studies and the key information).
The studies were published between 1955 and 2016; with only 12 (22%) after 2010. The studies were conducted in 13 out of the 48 countries in Sub-Saharan Africa: in eastern Africa (Ethiopia, Kenya, Sudan), western Africa (Burkina Faso, Cameroon, Chad, Ghana, Guinea, Niger, Nigeria, Mali, Senegal) and southern Africa (pre-independent Namibia). More than half of the studies were from western Africa (30/54), notably Senegal (6), Burkina Faso (5) and Mali (5). Twenty-three studies studied CL in the community (including three among school-children), and 28 used data collected in health facilities (including 18 dermatology specialized services). The remaining three studies were mixed. All 54 studies were observational: 29 (54%) were descriptive case series (numbering a total of 13,257 cases), and 25 (46%) followed a cross-sectional design, usually survey with various tools employed such as clinical screening or questionnaires.
In eastern Africa, CL has been known for more than a century, with the first indigenous CL case recorded in 1911 in Sudan [28]. In Ethiopia, CL has been known since 1913, and diffuse CL (DCL) clinical form was documented in 1960 in the highlands [29]. The first report of L. aethiopica as a distinct taxonomic entity was published in 1978 [30,31], and since then, the species has also been found in the mountainous region of Kenya [32]. L. tropica was later reported from certain areas in Kenya during the 1990s, and since then considered to have a more restricted distribution than L. major [33,34].
In western Africa, only L. major has been thought to circulate in this region. The oldest case reports of CL come from Niger in 1911 [35], then from Nigeria in 1924, and from Senegal in 1933 [36]. Later more cases were reported from Cameroon, Mali, Mauritania, Burkina Faso and Guinea [37,38]. During the first half of the 20th century, the colonial medical officers documented sporadic case reports from an area that later became recognized as the ‘CL belt’ [38]. Several comprehensive ecological and epidemiological studies took place in suspected hyperendemic foci in Senegal [39–42], Mali and Niger [43]. Current Namibia (previously South West Africa), reported dozens of CL cases in the 1970s [44], but the disease was not considered as a public health problem by the authorities [45].
Twelve studies (Table 1) reported prevalence estimated by the Leishmanin Skin Test (LST)—also known as Montenegro test—to detect exposure to the parasites in CL foci. Through intradermal injection of Leishmania antigens, the induration is being read 48–72 hours later as a demonstration of a delayed type hypersensitivity reaction, much like a tuberculin skin test [11]. LST does not differentiate between past and present infection and not species specific, yet it is often used as a marker for cellular immunity against CL [46].
These studies were conducted at the community level in CL foci, and have shown fluctuation over time (Table 1). Changes from 4% to 91% in LST positivity rate were observed in the same villages following an outbreak in Sudan [47,48]. High variability across foci within one country has also been reported, for example in Ethiopia: in Ocholo, 57% of school children without CL lesions were LST positive [49], while another study in the central-Ethiopian Rift Valley, LST positivity was maximum 5%. A study conducted in two neighboring villages in central Mali also demonstrated high variability: prevalence of Leishmania infection in Kemena was 45%, with the incidence of 19% and 17%; higher than Sougoula with 20%, 6% and 6% for the same years [50]. Reasons for these discrepancies are not known but possibly linked with hyper-clustering of reservoirs and vectors, leading to different intensity of peridomestic transmissions in Kemena [50].
A 2014 study from Mali complemented LST surveys with PCR and finger prick blood sample to measure antibody levels to sand fly saliva in endemic districts [54]. The results showed uneven prevalence of LST positivity across three different climatic areas (49.9%, 24.9% and 2.6% in Diema, Kolokani, and Kolondieba respectively), linked with north-south declining vector density. PCR was used to confirm L. major as the causative agent. LST positivity was also shown to be correlated to higher levels of antibodies to sand fly salivary proteins [54].
Across the studies, a consistent finding is that the proportion of positive LST increased with age and areas where CL transmission is active, at least a third of the population have had exposure to the Leishmania parasite [37,43,47–51,54–56].
Twenty-one studies reported estimates of CL prevalence or incidence; five were using medical records from hospitals, and the remaining were population estimates obtained through active screening for CL lesions and scars at the community level. All diagnosis was based on clinical examination. Though additional confirmatory methods (microscopy/smear, histology, culture in NNN or combination of these) were mentioned in all studies but two, it is unclear whether these were used in some or all or none of the patients. Among the five studies that were hospital-based, two used the number of dermatology consultations as the denominator, and the CL cases proportion found is 2% in Ouagadougou, Burkina Faso [57] and 14% in Addis, Ethiopia [58]. If suspected cases were to be denominator to calculate the CL cases proportion, they were found to be 78% (251/320) in Mali [59] and 93%(74/80) in Burkina Faso [60].
In most of the studies in the community, the prevalence of active CL was less than 5%. In endemic areas, the frequency of CL scars usually exceeds that of CL active lesions, except in a few special settings (Table 2). In Utut, Rift Valley in Kenya, a higher lesion versus scar rate (50% vs. 18%) in migrant charcoal workers suggested a non-immune population’s encounter with the disease in an area where transmission occurs [34]. Also during an outbreak in a new focus in Silti, Ethiopia, the frequency of CL lesions was considerably more than that of CL scars [63]. In Sudan, 36% of the community were found to harbor active lesions during an outbreak [68].
To complement the findings from published studies, we also examined the data from the country official reporting system to WHO. The system record data from 1996 onwards, but clearly there are missing data (Fig 2A and 2B). The absolute number of CL cases reported from eastern Africa is always higher than from western Africa, with Sudan bearing most of the burden. In western Africa, the number of cases reported from different countries is highly variable, and recurrent outbreaks were occurring in a 5–7 years cycle [74]. The increased cases in Ghana during 2002–2003 was prominent, yet there was a vacuum between 2007 and 2010, and cases were reported again starting in 2011. Other countries contribute little, with <100 cases per year (Nigeria, Senegal). No data was reported from this region during 2015–2017 [75].
The majority (n = 28) of the included records are clinical case series based on medical files from dermatology clinics or hospitals as the main data source. These studies describe a cohort of CL patients over a certain period, ranging from two to nine years. Chronologically, 10 studies reported CL cases in periods before 1980 [41,45,47,52,74,76–80], 11 described patient groups observed between 1980–2000 [35,57,59,67,69,81–87], and seven between 2000 and 2013 [58,60,88–92].
Hospitals reported that CL patients mainly came from surrounding areas or outside the cities or capital, such as Dakar, Senegal [74,88,93] or Niamey, Niger [84]. Eighteen studies report cases seen in specialized dermatology services. The proportion of CL cases among patients seen in those dermatology clinics is consistently less than 5% [59,69,94]. In the context of an outbreak, CL patients who seek care in specialized services represent only the tip of an iceberg, as shown in Burkina Faso (further described below). Between 1999 and 2005, a total of 7444 cases were recorded from various health centers in the capital Ouagadougou [95,96], but during the same period, the dermatology hospital had only seen 251 CL cases [57]. Diagnosis in all the case series is obtained through clinical examination and smears or histopathology. In Chad, a hospital close to the Sudanese border reported a very high proportion of CL confirmed cases (580 out of 680 cases between 2008–2012) [89].
Three countries have published studies on CL outbreaks: Sudan, Ethiopia, and Ghana. The first ever epidemics in Sudan were reported in 1976–1977 along the Nile, in Shendi-Atbara north of Khartoum [68], while the second and third outbreaks occurred in 1985 and 1986–1987, respectively [97]. The last epidemic in Sudan was in Tuti island, and it affected at least 10,000 people in 7 months. Underestimation is likely mandatory reporting only started after the epidemic reached its peak [86]. People of both sexes, all age groups and all socio-economic classes were affected, which is suggestive of a disease ravaging in a non-immune population. The causal parasite was L. major LON-1 [98] and the outbreak was attributed to various factors such as immigration from west Sudan, the heavy rainfall in the year of the outbreak after a long period of drought—which led to increase in sandfly density as well as the rodent reservoir population—and waning of herd immunity of migrants from CL endemic areas in western Sudan (Sayda el-Safi, personal communication). In Ethiopia, a CL outbreak occurred in 2005 in a district 150 km south of Addis. A survey then established an overall prevalence of 4.8% (92/1907), and 1 in 5 cases had mucocutaneous lesions [63].
In Ghana, an outbreak of localized skin lesion consistent with CL occurred in Ho municipality, Volta region in 2003 [90]. The usual triggers of CL epidemics such as intrusion of humans into vector habitat through deforestation, road construction, wars or migration were not at work here. Previously, only one CL case had been reported from the country in 1999, although the arid, Sahelian area of northern Ghana is considered to be part of the West African CL belt. Through passive case detection (with biopsy as a confirmatory diagnosis) with medical records review and active case finding, it was estimated that there were about 8876 CL cases between 2002 and 2003 in Ghana (Fig 2A). All age groups were affected, and since then CL is considered endemic in this area. A study in the same district later found 60% parasite-confirmed cases among active CL suspects (41/68). A phylogenetic analysis identified this Ghanaian parasite as new member of Leishmania enriettii complex, a possible new subgenus of pathogenic human Leishmania parasites [99].
Thirty-two studies described the clinical presentations of CL lesions. The most commonly used categories of the lesions are as followed: the localized CL or LCL, otherwise known as the classic oriental sore, refers to the lesion at the site of sand fly bites that may get ulcerated. LCL may appear as dry, papular forms with crust, or the wet, ulcerative forms with indurated edges. LCL can be singular or multifocal. When the nodules are multiple and nonulcerative, this is typically called a diffuse CL or DCL. In Sudan, mucosal leishmaniasis is described as lesion(s) that involves destructive mucosal inflammation which does not always start with a cutaneous lesion. This differs from New World mucocutaneous leishmaniasis (MCL), which refers to a metastatic dissemination to the mucosal tissues starting from a distal cutaneous lesion [52,100]. Bacterial superinfection is common along with pain, itchiness, fever and the secondary inflammation often complicates clinical diagnosis [11,101].
The diagnosis documented in the medical files are often missing. A dermatology hospital in Addis, Ethiopia reported that among 234 confirmed CL cases, only 22% were categorized—consisting of 9% DCL, 10% MCL and 3% LCL [58]. The higher proportion of complicated or atypical lesions are frequently reported from teaching hospitals or specialized services. This includes sporotrichoid CL with painless subcutaneous nodules along the lymphatic vessels in Sudan [80,87], or the diffuse CL in Ethiopia, which appear pseudo-lepromatous and can result in fungating or tumor-like lesions [52,80].
In the majority of the studies, the natural history of the lesions is only briefly described (n = 51). The duration between the first bite to lesion formation for LCL varied between 3–12 weeks [62,90]. Although CL can heal spontaneously, this seems to be dependent on the reported parasite species: L. major heals within approximately 2 to 12 months and L. tropica within 15 months, with a terminal scar appearing after about 24 months [11]. The description of diffuse CL caused by L. aethiopica suggests that it presents initially with nodules which do not heal or ulcerate but can metastasize widely [76] and are known to be very difficult to treat. In the case of DCL, spontaneous cure almost never happens. Mucocutaneous leishmaniasis is rare in Africa, but cases have been reported from Sudan and Ethiopia [52,80,100]. The lesions tend to be infiltrative and result in chronic edematous inflammation involving the lips, nose, buccal mucosa and larynx are.
With regard to the locations of CL lesions, there appears to be a regional difference. CL lesions from eastern Africa are mostly found on the head (i.e., face including cheek, nose, forehead, ears, lips) and less on the arms, legs or trunk, while from western Africa the highest proportion of lesions are on the upper and lower extremities.
Amongst the 42 studies reporting the sex ratio of the patients (Fig 3), only 12 recorded more females than males affected [49,50,56,63,70,72,82,95,102] while the remaining described male preponderance, either due to hypothesized occupational exposure or males’ easier access to seek care in a health facility. Thirty-six out of the 54 studies reported the age of the CL cases: people of all ages are affected. However, when stratification according to age was reported, there is a broad tendency towards younger age groups (between 10–30 years old.
CL and HIV co-morbidities has been described in Burkina Faso [57,60,103], Cameroon [70], Mali [59], and Ethiopia [91], while sporadic cases have also been reported from Guinea, Ghana, Senegal, Nigeria, Ivory Coast and Sudan. Burkina Faso has recorded 13.5% (10/74) HIV positivity in a cohort of CL patients in 2000, and another cohort of 32 CL/HIV patients was described in 2003–2004 [60,103]. Six out of 10 DCL cases in Ouagadougou were co-infected with HIV [57]. In Bamako, Mali, the prevalence of HIV among CL patients was 2.4% [59]. In Tigray, Ethiopia, a study reported an HIV prevalence of 5.6%, which increased to 8% two years later in 167 CL patients [92,104]. The only study reporting CL/HIV prevalence in the community was done in Cameroon in 2008. Here, a total of 32 466 subjects were clinically screened, and amongst 146 active CL patients, seven (4.8%) tested positive for HIV-1 and/or HIV-2 [70].
The consistent finding is that the clinical forms of CL are more diverse and complex in HIV co-infected patients, posing significant challenges in diagnosis and treatment. The lesions tend to be more severe: there are reports of infiltrative, leprosy-like, diffuse, psoriasis-like, verrucous, sporotrichoid, and angiomatous or Kaposi-like. Patients are more likely to have more than one lesion and more than one clinical forms [103]. Also, the time to lesion healing was longer in immunosuppressed individuals [70], and particularly in atypical and severe CL patients with poor response to treatment [91].
Our review shows that CL is reported in at least 13 countries in SSA but the true burden remains unknown. Several foci in Mali, Guinea, and Senegal have been studied intensively in the last half of the 20th century, but the published literature on CL can best be described as irregular and patchy. There is a lack of population-based or longitudinal studies to measure prevalence and incidence. The current CL burden is difficult to estimate accurately as primary data are scarce and CL cases often clusters in pocket areas. The prevalence in western Africa appears to be low, yet unprecedented outbreaks have occurred, such as in Burkina Faso and Ghana. Several CL outbreaks probably never get reported [105,106]. In eastern Africa, although the number of CL cases are high, there is insufficient evidence on CL prevalence and incidence outside the context of CL outbreak or its spread to new areas.
The findings from this review provide further insights vis-à-vis the official data reported to the WHO’s global surveillance system. Based on reported cases in 2002–2009, WHO estimated a global CL incidence of 214,036 in 2012 with 35,300–90,500 cases from eastern Africa and a mere 790–1500 cases from the rest of SSA, albeit with 5–10 fold underestimation [5]. Data reported to WHO in 2005–2015 put the figure of global CL incidence at 187,855, and the estimated contribution of SSA remains negligible [107]. From the 2013 Global Burden of Disease (GBD) study which primarily used modeling, Sudan and Burkina Faso are the only two countries from SSA with significantly greater DALYs from CL than the global mean [20]. Our findings are in line with these, thus emphasising the critical need to improve on-the-ground data as sources for future estimates.
The quality of evidence found in our review is inadequate to establish a more accurate CL burden in this region. Case series provide a snapshot of a specific situation in a certain time and place, yet are hard to extrapolate. A considerable part of the data we reviewed originated from specialized dermatology services which only represent a small proportion of all CL cases. The patchwork distribution of CL within a country further hampers surveillance. The CL belt in SSA from West Africa to the Horn of Africa [38], confirmed with a modeled distribution map of CL [108], appears to be mainly supported by historical accounts. The currently available evidence is clearly limited.
Various factors have been attributed to the poor CL data from SSA [2,12,109]: 1) CL is not a notifiable disease in many of the endemic countries; 2) Patients do not seek care due to perceived self-healing nature of CL; 3) Poor access to health facilities as most affected people live in remote, rural areas; 4) Lack of control tools, including unavailability of diagnosis and limited capacity to offer effective treatment. Compared to other regions, the neglect of CL is obvious. For New World CL in Latin America, the Pan American Health Organization (PAHO) has coordinated efforts to standardize and centralize surveillance data [110]. A Regional Information System called SisLeish was eventually developed to become an essential tool to prioritize areas and guide control actions [111]. Understandably, the region bears a much higher burden than SSA (from 2001–2015, 843931 cases were reported from 17 countries in the Americas). Currently there is no regional approach to improve CL surveillance for SSA. Sudan is part of the WHO Eastern Mediterranean Region (EMRO) [112] while the rest of the SSA countries belong to the WHO African Region (AFRO).
Our review identified the fragmented knowledge on burden as one of the key challenges for CL control in SSA. Being a largely zoonotic disease, the control efforts for CL remains limited to care provision, while vector control or environmental measures are not feasible. The risk of outbreaks, however, should not be undermined. Co-infection with HIV, already a concern for VL, might pose further challenges in CL management. What can be done in the face of all these adversities?
In lights of the scanty data, steps should be taken to improve existing surveillance systems or establish one where it is non-existent. Each country could undertake a thorough review of CL epidemiological situation, using standardized methods, enabling compilation and comparison. The future actions must be adjusted to the country context. An integrated paradigm should be adopted: either in setting up rapid epidemiological assessments for CL alone or in taking opportunities to include CL with other skin-NTDs [113,114]. Recognising the common challenges of a vertical approach to each NTD affecting the skin, a common tool to monitor disability has been piloted [115]. Furthermore, WHO has recently released guidelines for the training of skin NTD for frontline health workers [116,117]. Building capacity in case detection through training or inclusion of CL in clinical guidelines is starting in Sudan and Ethiopia, following an algorithm developed for Eastern Mediterranean region by WHO [118].
The strengths of this review are the systematic search of the literature and the stringent process and reporting following a published protocol in PROSPERO. Furthermore, standardized reporting according to PRISMA guidelines is adhered to. The exclusion criteria for case series of fewer than ten patients have been chosen as the aim is to provide an idea on disease burden though we might risk missing individual case report(s) and may exclude countries which only has case report publications. By systematically assessing all published articles we aimed to draw attention to the importance of the disease and identify research priorities.
The major limitations of our study are first, the publication bias. Sub-national studies that are not published nor listed in the international electronic databases might be missed. Secondly, the weakness of passive detection and clinical case reporting. We could not provide a meta-analysis nor compare the results between studies, due to the high variability across individual studies (denominator, sampling strategy, …). We could not systematically assess the risk of bias in the individual records and apply the current standard of as many studies pre-dated this era. The quality of the data in the studies is relatively poor. However, with the limited data we had to rely on, we understand better the state of the evidence in regards to CL in SSA: still an uncharted territory.
Based on the gaps identified in this review, there are some research priorities to be addressed (see Table 3). Improving epidemiological knowledge on CL will help to advocate for actions and resources in SSA, where the burden of NTDs surpass all other regions [119]. Future studies on CL burden should explore not only physical but also the socio-economic impact of this morbidity. CL in sub-Saharan Africa should not remain an enigma.
The epidemiological burden of cutaneous leishmaniasis in sub-Saharan Africa appears to be poorly documented. There is a paucity of robust evidence on prevalence and incidence on CL in this region. The diversity of CL epidemiological characteristics in endemic countries is not yet fully investigated. Nevertheless, the burden of CL morbidity remains important and most likely to be underestimated. Surveillance and mapping should be improved to mitigate outbreak risk and address dual co-infection with HIV. The current fragmented knowledge should be approached regionally, and awareness must be raised. In addition to population-based studies that better define the CL burden in sub-Saharan Africa, health systems should consider studies and action to improve CL essential diagnosis and care.
|
10.1371/journal.ppat.1001165 | Identification and Genome-Wide Prediction of DNA Binding Specificities for the ApiAP2 Family of Regulators from the Malaria Parasite | The molecular mechanisms underlying transcriptional regulation in apicomplexan parasites remain poorly understood. Recently, the Apicomplexan AP2 (ApiAP2) family of DNA binding proteins was identified as a major class of transcriptional regulators that are found across all Apicomplexa. To gain insight into the regulatory role of these proteins in the malaria parasite, we have comprehensively surveyed the DNA-binding specificities of all 27 members of the ApiAP2 protein family from Plasmodium falciparum revealing unique binding preferences for the majority of these DNA binding proteins. In addition to high affinity primary motif interactions, we also observe interactions with secondary motifs. The ability of a number of ApiAP2 proteins to bind multiple, distinct motifs significantly increases the potential complexity of the transcriptional regulatory networks governed by the ApiAP2 family. Using these newly identified sequence motifs, we infer the trans-factors associated with previously reported plasmodial cis-elements and provide evidence that ApiAP2 proteins modulate key regulatory decisions at all stages of parasite development. Our results offer a detailed view of ApiAP2 DNA binding specificity and take the first step toward inferring comprehensive gene regulatory networks for P. falciparum.
| Plasmodium falciparum is the main cause of the devastating human disease malaria. This parasitic organism has a complex lifecycle spanning a variety of different cell types in the mosquito vector and human host. To adapt and survive in these different environments, the parasite precisely regulates the transcription of genes throughout its lifecycle. However, the mechanisms governing transcriptional regulation in P. falciparum are poorly understood. To date, a single family of specific transcription factors, the Apicomplexan AP2 (ApiAP2) proteins, has been identified. These DNA binding proteins are likely to play a major role in coordinating the development of this parasite and are therefore of major interest. Here, we determine the DNA binding specificities for the entire P. falciparum ApiAP2 family of DNA binding proteins. Our results demonstrate that these proteins bind diverse DNA sequence motifs and co-occur in functionally related sets of genes. By mapping these sequences throughout the parasite genome, we can begin to establish a regulatory network underlying parasite development. This study represents the first characterization of a family of DNA binding proteins in P. falciparum and provides an important step towards understanding gene regulation in this parasite.
| Plasmodium falciparum is responsible for the majority of human malaria cases and causes approximately 1 million deaths every year [1]. The complete lifecycle of P. falciparum includes three developmental stages, which occur in its mosquito vector, the human liver, and human blood. Within each developmental stage the parasite undergoes major morphological changes that are accompanied by precisely timed transcription of genes that are necessary for parasite growth, differentiation, and replication. Detailed transcriptome and proteome studies have been conducted across the different stages of the life cycle [2]–[12]. Despite these advances in our understanding of messenger RNA transcript dynamics in P. falciparum, very little is known regarding the mechanism of transcriptional regulation, including transcription factor binding and sequence specificity.
Basic transcriptional control in P. falciparum appears to resemble that of other eukaryotic organisms, with general transcription factors coordinating the recruitment of RNA polymerase II to core promoter elements [13]–[15]. Experiments aimed at identifying cis-acting sequences required for gene expression have successfully identified specific enhancer and repressor sequences upstream of the core promoter elements [16]–[26]. In the asexual blood stage, regulatory sequence elements have been identified for the gene encoding the knob-associated histidine-rich protein (kahrp) [16], glycophorin binding protein 130 (gbp130) [18], cytidine diphosphate-diacylglycerol synthase (pfcds) [19], the DNA polymerase delta gene [20], a subset of the heat shock protein (hsp) family [22], the rif genes [23] and the falcipains [24]. Additionally, three sequence motifs have been identified upstream of the var genes: the SPE1, CPE, and SPE2 motifs, of which the SPE2 motif has been hypothesized to be involved in silencing of var gene expression [25], [26]. In sexual blood stage parasites three distinct short sequence elements have been found to regulate expression of the gametocyte genes pfs16, pfs25 [17], and pgs28 [21]. In addition to these experimentally derived motifs, bioinformatic analyses of the P. falciparum genome have identified a number of potential cis-elements that may play a role in gene regulation [27]–[36]. However, attempts to identify trans-factors have been largely unsuccessful [13], [15], [37], [38], with the exceptions of Myb1 [39], [40] and the high mobility group box (HMGB) proteins [41], [42].
Recently, a large protein family was identified in P. falciparum, containing Apetala2 (AP2) domains [43]. AP2 domains were originally described in plants as DNA binding domains approximately 60 amino acids in length [44]. In plants, the AP2 family of transcription factors is one of the largest, playing key roles in developmental regulation [44] and stress responses [45]. The Apicomplexan AP2 (ApiAP2) proteins represent a lineage-specific expansion, and are highly conserved across all Plasmodium spp. and in other Apicomplexans including Theileria, Cryptosporidium [43] and Toxoplasma [46]. P. falciparum was initially predicted to contain 26 ApiAP2 factors, each containing one to three AP2 domains [43], while in Toxoplasma the family is expanded to over 50 ApiAP2 proteins [46]. We have noted a 27th highly conserved ApiAP2 protein (PF13_0267), which agrees with recent Pfam predictions for this protein [47]. Although other DNA binding proteins have been reported in the literature, ApiAP2 proteins represent the largest family of transcriptional regulators identified in P. falciparum, where they are expressed throughout the entire developmental lifecycle [43].
Previously, we established that two ApiAP2 proteins, PF14_0633 and PFF0200c, bind DNA with high sequence selectivity [48]. Subsequent work demonstrated that the P. berghei orthologue of PF14_0633 (PBANKA_132980) is essential for the formation of sporozoites [49], and specifically regulates sporozoite target genes by binding to the same GCATGCA motif that we identified [48]. More recently, PFF0200c was shown to function as a DNA tethering protein involved in heterochromatin formation and integrity [50] via binding to the previously identified SPE2 motif [25]. Importantly, PFF0200c does not appear to act as a transcriptional regulator in the blood stage. A third study identified a P. berghei protein, AP2-O [PBANKA_090590 (PF11_0442)], as an activator of genes required for invasion of the mosquito midgut during the mosquito stage of the life cycle [51]. Together, these studies highlight the importance of the ApiAP2 DNA binding proteins in modulating stage-specific gene regulation and chromatin integrity. Despite these recent advances, the regulatory function of the majority of ApiAP2 proteins remains unknown. The DNA sequences recognized by the members of this protein family are largely uncharacterized, and the target genes that these ApiAP2 factors bind are undefined.
Here we biochemically and computationally characterize the global DNA binding specificities for the entire ApiAP2 protein family from P. falciparum. Our results reveal a complex array of DNA sequence elements, with the majority of proteins binding to unique sequences. We demonstrate several cases where multiple AP2 domains within the same ApiAP2 protein are capable of binding distinct DNA sequences. The identification of these unique sequence motifs sheds light on the molecular mechanisms of transcriptional regulation by assigning correlations between putative cis-acting sequences in Plasmodium and the trans-factors that will bind at those sequences genome-wide. Our data reveal the likely identity of trans-acting ApiAP2 factors that specifically bind to previously described cis-motifs, illuminating some of the previously unknown effectors of plasmodial gene expression. For the many new motifs we identified, we predict putative targets for each of the ApiAP2 proteins. This work represents the first comprehensive analysis of the ApiAP2 DNA binding proteins in P. falciparum and provides a crucial missing link toward understanding their role in the regulation of parasite development.
The 27 plasmodial ApiAP2 proteins vary drastically in size (Figure S1), however, the predicted 60 amino acid AP2 domains, are well-defined and highly conserved. To determine the DNA binding specificity for the P. falciparum AP2 domains, we used protein binding microarrays (PBMs), which enable simultaneous screening of all possible DNA sequences up to ten nucleotides in length without sequence bias [52], [53]. Seminal studies from the Bulyk lab have used PBMs to comprehensively characterize individual transcription factors from a diverse array of organisms including yeast, worm, mouse and human [52]–[54]; and we have previously demonstrated its utility for Apicomplexan AP2 DNA binding proteins [48].
We created 50 constructs for PBM screening (Supplemental Text S1, Figure S2), including individual domains, full length proteins, and tandem domain arrangements (two AP2 domains separated by a short conserved linker sequence of 12 to 79 amino acids; designated DLD). Our analysis by PBM of these P. falciparum AP2 domains revealed motifs for 20 out of the 27 ApiAP2 proteins (Figure 1), including a motif for the recently identified ApiAP2 protein, PF13_0267, helping to confirm the new annotation for this protein. Results from at least two PBM experiments for each AP2 domain were used to generate position weight matrices (PWMs), which represent the DNA binding affinity for a given domain (Dataset S1, Figure 1). Replicate experiments had excellent correlation coefficients illustrating the robustness of the PBM methodology (see Supplemental Text S1). Enrichment scores (E-scores) were assigned for each 8-mer (allowing up to two gaps) [53], with a significance cut-off of 0.45 (E-scores range from -0.5 to +0.5) for specific 8-mers enriched above background. The E-score is a rank-based, nonparametric score that is robust to differences in protein concentration and reflects the relative preference for each 8-mer [55]. In total we identified sequence motifs for 24 AP2 domains found in a variety of protein architectures (Figure 1). While Figure 1 illustrates which motifs are linked to the blood stage of Plasmodium development, several motifs are also associated with ApiAP2 proteins during non-blood stages as well (see Supplemental Text S1). It is noteworthy that different AP2 domains from the same ApiAP2 protein bind distinct DNA sequence elements. However, we do find several motifs that are recognized by multiple ApiAP2 factors (see Supplemental Text S1 and Table S1). This complexity may allow for multifaceted transcriptional regulation using a smaller number of individual factors.
Protein-DNA interaction specificities are determined by the chemical interactions of amino acids and DNA bases [56]. Side chain flexibility and DNA distortions allow one DNA binding domain to interact with multiple distinct DNA sequences. For several of the AP2 domains there were significant differences among the top scoring 8-mer sequences that were bound, suggesting multi-motif recognition. Using the Seed and Wobble algorithm [57] we identified alternative motifs associated with 8-mers of high signal intensity that could not be explained by the primary motif for 14 AP2 domains (representing 13 ApiAP2 proteins) (Figure S3, Dataset S2). Some AP2 domains only had a single secondary motif, whereas others had up to four. The secondary motifs can be described based on their relationship with the corresponding primary motifs and fall into the broad categories of end modifications, core changes, variable spacer distances or alternate recognition interfaces [57] (see Supplemental Text S1). The ability of an individual domain to bind anywhere from one to five different DNA sequences would significantly increase the number of target genes that could be regulated by one factor.
We selected two ApiAP2 proteins for confirmation of secondary motif binding by electrophoretic mobility shift assays (EMSAs). Domain 2 of PFD0985w has three predicted secondary motifs in addition to the primary motif. A plot of the E-scores for all ungapped 8-mers reveals that the top 100 matches to both the primary motif and one of the secondary motifs (Figure 2A) are relatively equal in E-score. Therefore, PFD0985w_D2 should bind equally well to these two motifs. To test this hypothesis we generated 60 bp oligonucleotides with the specific motif sequence in the center flanked by random sequences. EMSAs with purified PFD0985w_D2 demonstrate that both oligonucleotides are bound equally well, and that the primary motif is capable of out-competing the secondary motif and vice versa (Figure 2B). No binding is observed with an unrelated non-specific oligonucleotide, indicating specificity for the predicted motifs (data not shown). The second ApiAP2 factor that we selected for confirmation of secondary motifs was PFL1900w_DLD. The highest scoring 8-mers for this tandem domain were represented by completely distinct sequences and a plot of all 8-mers and their E-scores revealed preferential binding with a primary, secondary, and tertiary motif (Figure 2C). PFL1900w_DLD was able to shift all three motifs, but with varying affinities (data not shown for the primary and tertiary motifs), and competition between the secondary and tertiary motifs revealed a clear preference for the secondary motif over the tertiary motif (Figure 2D). These results suggest that the secondary motifs detected represent bona fide sequences bound by the AP2 domains and the E-score distributions accurately reflect binding affinities.
Both computational predictions and experimental data have identified a number of DNA sequence motifs upstream of genes in Plasmodium [17]–[22], [27]–[32], but the specific trans-factors that bind to these motifs have mostly remained elusive. For three cases, we now establish plausible links between the newly identified AP2 DNA sequence motifs and these previous reports. Militello et al. identified a specific motif, (A/G)NGGGG(C/A) (called the G-box), upstream of 8 out of 18 Plasmodium heat shock genes [22]. The occurrence of this GC-rich motif in the genome is low (Table S2), suggesting that its presence in upstream sequences may be significant for transcriptional regulation. The sequence motif that we have identified for PF13_0235_D1 is nearly identical to the G-box element (Figure 3A). Furthermore, the expression profiles of pf13_0235, hsp86 (pf07_0029), and hsp70 (pf08_0054), two heat shock genes containing one or more G-boxes exhibit a strong positive correlation (r = 0.93) during the asexual blood stage [2] (Figure 3A), suggesting that PF13_0235 may play a role in regulating hsp gene expression. We performed EMSAs using both G-box elements of the hsp86 upstream region and found that PF13_0235_D1 interacts specifically with the G-box and deletion of both G-boxes is required to completely eliminate binding (Figure 3B, C). In the presence of only one G-box, binding is severely reduced (Figure 3B, C) suggesting that PF13_0235_D1 preferentially interacts with both G-boxes, perhaps through dimerization (see below). No binding is observed with an unrelated non-specific oligonucleotide at similar protein concentration (data not shown). This result is in agreement with in vivo data from transient transfections, where elimination of G-box 1 substantially reduced luciferase expression, but did not completely abolish it [22]. We also tested the G-box from the 5′ flanking region of hsp70 for in vitro binding by EMSA, and confirmed that PF13_0235_D1 binds this sequence in vitro. It is interesting to note that the binding of this single G-box motif is similar to that seen for hsp86 after deletion of one G-box, suggesting that higher affinity interactions require two occurrences of this motif.
Likewise, the sequence element bound by PF10_0075_D3, GTGCA, is enriched in the upstream sequences of genes involved in merozoite development and invasion [31], [58]. Using EMSAs we find that PF10_0075_D3 binds to the GTGCA motif upstream of msp1 (pfi1475w), msp10 (pff0995c) and rhopH 3 (pfi0265c) (Figure S4A), and no binding is observed with an unrelated non-specific oligonucleotide, indicating specificity for the predicted motifs (data not shown). Previous expression studies using a rhoptry gene promoter to drive luciferase expression have demonstrated that the GTGCA motif is important for rhoptry gene-like stage-specific expression [31]. Combined with our EMSA results, this suggests that PF10_0075 may play a role in regulating the expression of invasion-related genes in P. falciparum.
Finally, a specific 5 bp motif in the 5′-upstream region of gbp130 (pf10_0159), GTATT, was previously found to be bound by unknown nuclear factors in a sequence-specific manner [18]. The reverse complement of this 5 bp element is nearly identical to the motif we have identified for PF11_0091. EMSAs using the promoter region of gbp130 and the purified AP2 domain from PF11_0091 confirm its ability to interact with this sequence (Figure S4B), while no binding was observed with an unrelated non-specific oligonucleotide (data not shown), suggesting it is a possible regulator of GBP130 function. The PBM-derived motifs are useful to suggest putative targets for the ApiAP2 trans-factors, especially where previous characterization is available. However, in vivo assays will be required in all cases to validate these interactions on a protein-by-protein basis.
To begin to characterize the functional role of ApiAP2 proteins, we searched the P. falciparum genome for sequences in promoters and untranslated regions that may serve as regulatory sites for ApiAP2 binding. As a first analysis, we used our AP2-specific position weight matrices generated from the PBM data to search the 5′ upstream sequence elements of Plasmodium genes using ScanACE [59], which lists all matches to our position weight matrices within the user defined threshold. Although putative transcription start sites have been predicted [60], actual transcription start sites are still poorly defined in P. falciparum [61]. Therefore, we searched 2 kb upstream of the ATG start codon or until an upstream open reading frame was encountered. While this search provides a list of all possible motif occurrences determined from matches to a specific position weight matrix (Datasets S3 and S4), it is undoubtedly an overestimation of putative target genes. In reality, the presence of a regulatory element upstream of a gene does not confirm a regulatory interaction exists, and many motif occurrences may be inactive [62]. Furthermore, for a regulatory element to be functional, it needs to be accessible for binding, which is in part determined by nucleosome occupancy. Nucleosome occupancy has been mapped during the intraerythrocytic developmental cycle (IDC) of P. falciparum [63] and using this data we were able to determine that between 65 and 97% of our ScanACE predicted binding sites are accessible (nucleosome-free) at some point during the IDC (Table S2). This suggests that the majority of our predictions have the potential to be active; however, in vivo binding affinities may differ from in vitro determined affinities, possibly altering the weighting of specific nucleotide positions within the motifs. Ultimately, the actual target sequences of each ApiAP2 protein will need to be individually determined through experimental validation in vivo during the specific lifecycle stage of interest.
As a test of the ability of our ApiAP2 proteins to bind to the ScanACE predicted targets we selected a putative target for the newly annotated ApiAP2 protein PF13_0267; pfc0975c has a match to the CTAGAA motif at 1469 bp upstream of the start codon. EMSAs showed that the putative target sequence was bound by the purified AP2 domain from PF13_0267, while a mutant oligonucleotide lacking the predicted target sequence did not exhibit significant binding (Figure S5). Although these results demonstrate that our ScanACE-predicted target genes provide a good starting point to search for candidate genes for in vivo testing, this does not indicate if pfc0975c is a true target of PF13_0267. Indeed the motif bound by PF13_0267 is found upstream of almost all genes and in vivo validation will be required to identify actual targets. Complete AP2 motif occurrence data for the P. falciparum genome are available to the malaria community at PlasmoDB (www.plasmodb.org) [64].
While the ScanACE analysis provides a list of all occurrences for each motif, it is unlikely that the ApiAP2 proteins are binding to all possible motif occurrences, and instead that they bind to a smaller subset of promoters. Proteins that are co-localized in the cell or form sub-cellular structures such as the ribosome have been found to be transcriptionally co-regulated in other organisms such as yeast, and often are regulated by the same cis-elements [65]. Genes that are functionally distinct, but are co-expressed can also be regulated by the same cis-elements in their upstream regions. To narrow the ScanACE list to a more informative subset of putative target genes we used relative mRNA abundance profiles to define relationships between co-expressed Plasmodium genes [2]. We used linear regression to determine at each time point the extent to which each AP2 motif contributes to (or recapitulates) the overall expression of the genes that contain a given motif in the upstream regions (see Methods and Supplemental Text S1 for details [66]). Thus each motif at each time point is associated with a score (i.e. the fitted regression coefficient), which is positive if genes that have the motif tend to go up at that time point, or negative if they tend to go down. These scores define the predicted motif activity at each time point, and an activity profile across the entire IDC. Activity profiles reflect the predictive effect of individual AP2 motifs on gene expression of a set of target genes at a given IDC timepoint (Figure 4A), and are therefore independent of the mRNA expression profiles of the AP2 genes themselves. The activity profile for each motif was then used to iteratively identify genes containing the target motif in their 5′ upstream regions that share an expression profile similar to the activity profile. This provided a refined list of putative target genes that are co-expressed with one another (Dataset S5).
To illustrate how this approach improved our target predictions we focus on the G-box motif. The ScanACE predicted target list for this motif includes 522 genes (Dataset S3), and using the activity profile-based approach we refine this list to a set of 124 putative co-expressed target genes (Table 1, Dataset S5). A comparison of the average expression profiles for these genes with the expression of pf13_0235, the gene for the ApiAP2 factor that binds the G-box, shows a strong positive correlation (0.97; Figure 4). Similar comparisons were made for all ApiAP2 factors and their putative targets (Figure S6), and we observe either significant positive (r-values from 0.97 to 0.43) or negative correlations (r-values from −0.94 to −0.56) for many ApiAP2 factors, implying that this protein family may act as both activators and repressors of target gene expression.
Functional annotation of the predicted targets using the DAVID bioinformatics resource [67] identified enrichment of genes involved in specific cellular processes (Table 1). Targets of PF13_0235_D1, the ApiAP2 factor that binds the G-box motif, include genes involved in ribosome function or translation and heat shock response genes. Genes involved in these functions have previously been suggested to be regulated via the G-box element [22], [33], supporting our target gene predictions. Other notable examples include the enrichment of targets involved in cell invasion and host cell entry for the PF10_0075_D3 motif (GTGCAC) and DNA binding for the MAL8P1.153 motif (ACACA). The involvement of the GTGCAC motif in invasion related processes has been independently predicted by three bioinformatic studies [31]–[33], while the ACACA motif was previously associated with DNA replication [31]. Since the majority of these motifs have not been previously described in P. falciparum, our prediction of target gene functions are novel and warrant further characterization.
Similarly, we used a recently published P. falciparum growth perturbation dataset [68] as an alternative data source to create activity profiles to refine our target gene predictions (Figure S7, Dataset S6, Supplemental Text S1). Genes that respond in a similar manner to a perturbation are more likely to be regulated by the same factor and we observed narrower target gene lists for each motif, many of which overlap with the predictions made using the IDC co-expression data, and others that are novel target gene predictions (Table 1, Figure 4, Figure S8). Further details on the perturbation refinement of target genes can be found in the Supplemental Text S1. Combined, refinement of putative targets using the high resolution temporal gene expression data [2] and the perturbation dataset [68] produce manageable gene sets for further analysis.
We also looked at motif enrichment in the upstream regions of var genes. There are approximately 60 var genes in P. falciparum that encode the antigenically variant erythrocyte membrane protein 1 (PfEMP1), which is involved in sequestration of infected red blood cells in the vasculature [69]. The var genes have been divided into groups based on their location along the chromosome (internal or at chromosome ends), the direction of transcription, and the sequences of their intron and 5′ and 3′ untranslated regions [70]. Using our ScanACE prediction of binding sites, we observed a striking pattern of ApiAP2 motifs that were clustered in discrete positions of all three types of var promoters (Figure 5). In the upsB promoters we observed repeated motifs for PFF0200c that correspond to the previously identified bipartite SPE2 element [25], located at −2000 to −3000 bp upstream of the ATG start codon. While PF10_0075_D3 binds to a similar sequence as the SPE2 element, recent work has demonstrated that PFF0200c is the primary ApiAP2 factor that binds to this element in vivo [50]. We also predict the SPE1 element of upsB var genes at −1200 bp [25], which matches to the motif we have identified for PF14_0633. In addition to these previously identified sequence elements, we predict binding sites for PF11_0442, which binds to the sequence GCTAGC (Figure 5). Matches to this motif are conserved in all three major types (A, B and C) of var genes. The presence of multiple ApiAP2 binding sites upstream of var genes suggests multiple ApiAP2 factors may be recruited for binding upstream of the var genes. var promoters are silenced by default [71], and it is possible that this silencing is maintained by the co-ordinated action of multiple ApiAP2 factors. Further investigation of these discrete motif-enriched sites will be required to determine the precise role of additional ApiAP2 proteins in var gene regulation.
The IDC transcriptome of P. vivax [72] suggests that the regulation of development follows a similar cascade of gene expression as that seen for P. falciparum [2], [6]. All P. falciparum ApiAP2 proteins have syntenic homologues in P. vivax and are expressed at a similar stage of development during the IDC with the exception of pf14_0471 (pv118015) which is shifted from trophozoite to late schizont stage [72] (Figure S9A). AP2 domains are highly conserved across all Plasmodium spp. and will likely bind the same motifs. It follows that the ApiAP2 proteins may regulate similar or related target genes in P. vivax compared to P. falciparum. We calculated activity profiles using the P. vivax asexual blood stage transcriptome data [72] (Figure S9B) and as seen for P. falciparum, most of the motifs are associated with activation or repression of target genes at one or more timepoints. However, a comparison of the target gene lists for each motif in P. falciparum and P. vivax shows that the conservation of putative targets is low, ranging from 0 to 53% (Table S3). This implies that although the AP2 DNA binding domains are highly conserved, some of the regulons across these species have diverged and regulation of orthologous genes has evolved independently. It should be noted that a comparison of target genes in non-blood stages, including the mosquito stage, demonstrates substantial statistically significant conservation of motifs among co-regulated genes in P. vivax and P. falciparum [73]. Therefore it will be important to identify the actual target genes bound in vivo for each motif to determine the actual extent of conservation between species.
While the above analyses are limited to transcriptional regulation during the blood stage of the lifecycle, ApiAP2 function is not limited to the IDC. Accordingly, we analyzed gene expression data from P. falciparum gametocytes, zygotes and sporozoites [6], [10]. Activity profiles for ApiAP2 motifs in gametocyte data revealed activity for a number of the PBM derived motifs during gametocytogenesis (Figure S10A). We found two motifs to be active in zygotes, including the previously identified zygote motif for the P. berghei orthologue of PF11_0442 (PBANKA_090590), as well as the CACACA motif, which is bound by PF13_0026 (Figure S10A, Dataset S7). Activity profiles for AP2 DNA motifs in sporozoites identify PF14_0633, PFD0985w_D2, and PFF0670w_D1 as potentially active AP2 domains during this stage (Figure S10A). Since there is currently no liver stage data available for Plasmodium species that infect humans, we used data from the rodent malaria species P. yoelii [9]. Activity profiles for our PBM derived motifs in the P. yoelii liver stage yielded a number of motifs that are potentially active during this stage (Figure S11). Further discussion of the non-blood stage active motifs can be found in the Supplemental Text S1.
Understanding the molecular mechanisms and the regulatory processes that underlie gene expression during the development of P. falciparum is a major challenge in malaria research. The ability to map the recognition sites of DNA binding proteins genome-wide enables an improved understanding of how trans-factors regulate gene expression and has been undertaken for only a few large eukaryotic families of transcription factors [54], [57], [74]–[76]. Here, we have comprehensively characterized the P. falciparum ApiAP2 protein family and established their preferred DNA recognition motifs. A number of findings strongly implicate these factors as major regulators of gene expression in P. falciparum. First, the expression of the ApiAP2 genes at distinct times throughout the IDC suggests they are available to regulate target genes throughout the entire 48-hour blood stage of the parasite. Second, the diversity of sequences bound by these domains is sufficient to account for the full cascade of gene expression observed in the IDC. Furthermore, ApiAP2 genes form an interaction network with themselves during the IDC (Figure S12), suggesting that in addition to regulating target genes they may also regulate their own expression. For all three instances where in vivo data on binding sites for ApiAP2 proteins are available [49]–[51], our PBM derived motifs are excellent matches, emphasizing the quality of these data and the robustness of the PBMs at identifying preferred sequence elements. Finally, half of the motifs that we identified are nearly identical to motifs that were independently predicted computationally as putative cis-elements in the P. falciparum genome [27]–[29], [31]–[33].
Our success rate for identifying DNA binding specificities (20 out of 27 ApiAP2 proteins, 74%) using the PBMs is comparable to results obtained using the same technology with other transcription factor families from yeast and mouse (40–85% successful) [54], [57], [76]. For those AP2 domains that did not yield a result, there are numerous technical reasons why we may fail to detect binding. Although we were able to successfully express all of our GST-fusion constructs, possible reasons for failure to detect binding events include insufficient protein concentration, low protein stability, binding conditions (e.g. ionic strength) used, or low-affinity binding (undetectable by PBM). Another possibility is simply that some predicted AP2 domains lack specific DNA binding activity altogether. As mentioned above, the ApiAP2 proteins vary dramatically in size, including four proteins that are less than 40 kDa (Figure S1). Only one of the four smallest ApiAP2 proteins (<300 amino acids) binds to DNA (PF13_0026). Despite our testing both full-length ApiAP2 proteins and isolated AP2 domains from these small proteins, we could detect no DNA binding by PBM. This is surprising given their lack of any other predicted functional domains.
Both computational prediction of motifs [22], [31]–[33] and experimental data [25], [49]–[51] have identified a number of regulatory elements involving repeated iterations of the same motif. One established way to recruit multiple copies of the same factor to a particular site in the genome is through the formation of dimers or multimers of the same protein. A crystal structure has recently been solved of the AP2 domain from PF14_0633, which reveals that the AP2 domain forms a dimer when bound to DNA [77]. This structure suggests that other ApiAP2 proteins may also form homo- or heterodimers [78], thereby recognizing multiple DNA sequence motifs in concert. Further support for this idea is seen in our EMSA analysis of PF13_0235_D1, which shows that higher affinity binding to the G-boxes upstream of hsp86 (Figure 3) occurs when multiple copies of the motif are present. Similar results were obtained for PF10_0075_D3 and the predicted rhopH 3 target (Figure S4A) suggesting that AP2 domain dimerization may enhance binding to these sequences. Our ScanACE predictions of target genes identify motif repeats in upstream regions (Dataset S3), which may allow for tighter control of target gene expression by the ApiAP2 factors. Similarly, heterodimer formation may facilitate combinatorial regulation of gene expression at co- motifs. Evidence for such interactions can be found in yeast two-hybrid assays that have detected associations between ApiAP2 proteins [79]. Taken together, our genome-wide motif predictions and the ability of plasmodial AP2 domains to form dimers implies that these factors likely work together to regulate target gene expression.
The extremely high level of conservation (>95% identity) for each AP2 domain across all Plasmodium spp. [43] suggests that orthologues from other species will bind to similar DNA sequence motifs. Indeed, motifs for the P. falciparum AP2 domains of PF14_0633 and PF11_0442 are matches to the experimentally determined motifs for their P. berghei orthologues (AP2-Sp and AP2-O, respectively) [49], [51]. However, while the cis-elements bound by individual AP2 domains may be conserved across species, our predicted IDC target gene sets for ApiAP2 proteins appear to differ extensively. This is common in other eukaryotic organisms, where DNA binding domains are highly conserved across species, but downstream target genes are divergent [80]–[85]. Our data suggests that this may be true for Plasmodium spp., as demonstrated by the divergence of IDC target gene sets in P. vivax and P. yoelii compared to P. falciparum, which contrasts with the almost perfect conservation of AP2 domains and the similar temporal expression of ApiAP2 genes. We previously showed that the orthologous AP2 domains from PF14_0633 in P. falciparum and its distant Apicomplexan relative C. parvum (cgd2_3490) bind virtually identical sequence elements [48], but their predicted regulons had virtually no overlap. However, there are some examples of transcription factor binding site conservation among specific groups of target genes. For example, the G-box element has been predicted to regulate heat shock genes in both C. parvum [86] and P. falciparum [22], and both the PF14_0633 (TGCATGCA) and PF10_0075_D3/PFF0200c_DLD (GTGCAC) motifs are conserved among sporozoite and merozoite invasion genes in P. falciparum, P. vivax, P. yoelii, and P. knowlesi [87]. A better assessment of target gene conservation between species will be possible with more accurate target gene lists from chromatin immunoprecipitation experiments for each individual ApiAP2 factor. Indeed, this has recently been demonstrated in the asexual blood stages for PFF0200c [50], and only a subset of predicted targets were actually bound in vivo by this factor. This is also likely to be true for other ApiAP2 factors and further work identifying functional binding sites will clarify the level of conservation of transcription factor binding sites among the different Plasmodium spp.
Combinatorial gene regulation is an important aspect of transcription in many organisms. It controls the level of gene expression, the precise timing of expression, and determines the ability of a regulatory circuit to respond differently to a wide variety of extracellular signals. The finding that there are a relatively small number of specific transcription factors in P. falciparum prompted the hypothesis that combinatorial gene regulation plays an important role in the parasite [29]. Our finding that some ApiAP2 proteins can bind more than one sequence element significantly increases the potential complexity of the regulatory network. We identified secondary DNA binding preferences for 14 AP2 domains. These secondary motifs allow one AP2 domain to regulate a much broader range of targets than initially predicted from our ScanACE analysis using only the primary motifs. Precedent for this has been seen in yeast where the transcriptional activator, HAP1, binds two completely different regulatory sequences, allowing for the regulation of target genes with different promoter elements [88]. A similar observation was made in a recent PBM analysis of 104 mouse transcription factors, where it was found that almost half of the TFs recognized multiple sequence motifs [57]. A re-analysis of previously generated chromatin immunoprecipitation – microarray (ChIP-chip) data illustrated that these newly identified alternate motifs were also bound in vivo [57]. Similar in vivo data will be required to fully elucidate the role of individual ApiAP2 motif occurrences on target gene functions; however it is evident from our data that the ApiAP2 factors have sufficient diversity in sequence recognition to potentially regulate all P. falciparum genes.
While it is clear that the ApiAP2 factors bind DNA and recent work has begun to explore their contribution to transcriptional regulation [49], [51], these factors are also capable of binding DNA as scaffolding and recruitment proteins [50]. This finding opens up new possible functions for this family of DNA binding proteins. The next step in understanding ApiAP2 function will be to address the role of other domains in these proteins. The majority of ApiAP2 proteins are predicted to encode extremely large proteins, yet nothing is known regarding other functional domains outside of the AP2 domain. Yeast two-hybrid assays have identified interactions between ApiAP2 proteins and chromatin modifying proteins [79], which could help recruit ApiAP2 proteins to target sites in the genome. How and when ApiAP2 proteins are targeted to the nucleus and if this is actively regulated also remains unanswered. Our global mapping of ApiAP2 motif preferences represents the first characterization of a Plasmodium family of DNA binding proteins and provides a starting point to investigate transcriptional control in P. falciparum. These results provide an important step toward understanding the role of these proteins as major regulators throughout all stages of parasite development in Plasmodium spp. and other related Apicomplexan species.
Domain boundaries were defined as in [43] and extensions were made based on sequence homology both 5′ and 3′ of the AP2 domains amongst Plasmodium spp. orthologues, as well as using structural predictions from the online secondary structure prediction server Jpred3 [89]. In total 77 different versions of ApiAP2 domains from P. falciparum as well as one from P. berghei were cloned into pGEX-4T1 (GE Life Sciences) to create N-terminal glutathione S-transferase (GST) fusions (Figure S2). Proteins were expressed in BL21-CodonPlus(DE3)-RIL cells (Stratagene) with 0.2 mM IPTG at room temperature and affinity purified using glutathione resin (Clontech). The purity of each protein was estimated by silver stained SDS-PAGE and yields were calculated based on absorbance at 260 nm and specific molar extinction coefficients. All protein concentrations include contaminating products and will be an overestimation of the actual AP2 domain amounts.
All constructs in Figure S2 were tested at least once on the PBMs, and positive motifs were confirmed at least twice on independent arrays. The PBM experiments were performed as previously described [53]. Briefly, custom designed oligonucleotide arrays are double-stranded using a universal primer, incubated with GST-AP2 fusion proteins, visualized with Alexa-488 conjugated anti-GST antibody, and scanned using an Axon 4200A scanner. Proteins were used at the maximum concentration obtained from purification and represent one-fifth of the total reaction volume used on the PBM. In this study three different universal platforms were used covering all contiguous 8-mers as well as gapped 8-mers spanning up to 10 positions. After data normalization and calculation of enrichment scores [53], [55] the “Seed-and-Wobble” algorithm was applied to combine the data from two separate experiments and create position weight matrices (PWMs) [55]. An enrichment score cut-off of 0.45 was used to distinguish high affinity binding data from low affinity and non-specific binding. The score for each 8-mer reflects the affinity of a DNA binding domain for that sequence, with higher scores representing tighter interactions [55]. Secondary motifs were identified by running the “rerank” program until E-scores below 0.45 were obtained [55]. The PBM analysis suite was downloaded from the Bulyk lab (http://the_brain.bwh.harvard.edu/PBMAnalysisSuite/index.html). For public access, all motifs have been deposited in the UniPROBE database [90].
N-terminal GST fusions of the ApiAP2 domains were purified as described above. Single-stranded HPLC purified 5′ biotinylated oligonucleotides were purchased from Integrated DNA Technologies (http://www.idtdna.com) and annealed with complementary oligonucleotides to create double-stranded probes (Table S4). EMSAs were performed using the LightShift Chemiluminescent EMSA kit (Pierce). Briefly, purified protein was incubated with 50 ng/µL of poly(dI-dC) and 10 fmol of biotinylated probes. Competitor DNA was added in 50 or 250 fold molar excess. All reactions were incubated in the kit EMSA buffer with 2.5% glycerol, 5 mM MgCl2, 10 mM EDTA, 50 mM KCl, and 0.05% NP-40 at room temperature for 20 minutes. Electrophoresis and transfer to Nylon membrane (Hybond) was performed according to the manufacturer's instructions. The Chemiluminescent Nucleic Acid Detection Module (Pierce) was used according to the manufacturer's instructions to visualize the probes.
2 kb-long upstream regions (or up to the nearest ORF) were first extracted from whole genome sequences and associated gene annotation data (PlasmoDB 6.0). PBM motifs were trimmed down to their 6 most informative consecutive motif positions (motif cores). Then, we determined all core PBM motif occurrences in the 2 kb regions using the ScanACE approach [91]. G+C content was set to 13.1% in P. falciparum, 42.8% in P. vivax and 20.1% in P. yoelii. Score threshold was set to the average score of randomly drawn sequences from the PBM PWMs, minus two standard deviations (this is the default ScanACE setting).
In order to identify candidate target genes for each AP2, we reasoned that these target genes should be co-expressed and should share the AP2 binding site we identified using PBMs. Thus, we first identified groups of genes that are co-expressed across multiple experimental conditions or multiple time points (e.g., in the IDC). Then for each AP2 motif, we determined in which of the groups the motif was over-represented. For each of these groups, we extracted the genes associated with one or more motif occurrences. Thus, target genes in this definition can come from multiple co-expression groups (and not all genes in these co-expression groups end up in the target list because not all genes in these groups will have a motif occurrence in their promoter). In order to define co-expressed gene groups, we used the k-means approach together with the Pearson correlation. Motif scanning was performed using the ScanACE approach [91] as described above. In order to determine functional motif score threshold, we used an information-theoretic procedure analogous to that used in FIRE [27]: briefly, we determined the motif score threshold such that the resulting motif occurrences best explain the co-expression clusters obtained by k-means. At a given motif score threshold, motif over-representation in each cluster was assessed using the hypergeometric distribution; to correct for multiple testing (i.e. multiple clusters being evaluated), we used the Benjamini-Hochberg procedure; corrected p-values corresponding to an estimated overall FDR of 0.25 were considered significant and the genes associated with motif occurrences in these clusters were extracted. Because the k-means is dependent on initialization, we repeated the entire procedure 10 times; genes extracted 3 times or more (out of 10) were considered as candidate target genes. Thus, only genes associated with the considered motif and that are consistently found as co-expressed together with other genes sharing that same motif end up as candidate target genes in our analysis. More detailed methods are available in the Supplemental Text S1.
PlasmoDB (www.plasmodb.org) accession numbers for genes and proteins discussed in this publication are: hsp86 (PF07_0029); hsp70 (PF08_0054); msp1 (PFI1475w); msp10 (PFF0995c); rhopH 3 (PFI0265c); gbp130 (PF10_0159); AP2-O (PBANKA_090590); AP2-Sp (PBANKA_132980).
|
10.1371/journal.pgen.0030158 | Telomeric Trans-Silencing: An Epigenetic Repression Combining RNA Silencing and Heterochromatin Formation | The study of P-element repression in Drosophila melanogaster led to the discovery of the telomeric Trans-Silencing Effect (TSE), a repression mechanism by which a transposon or a transgene inserted in subtelomeric heterochromatin (Telomeric Associated Sequence or TAS) has the capacity to repress in trans in the female germline, a homologous transposon, or transgene located in euchromatin. TSE shows variegation among egg chambers in ovaries when silencing is incomplete. Here, we report that TSE displays an epigenetic transmission through meiosis, which involves an extrachromosomal maternally transmitted factor. We show that this silencing is highly sensitive to mutations affecting both heterochromatin formation (Su(var)205 encoding Heterochromatin Protein 1 and Su(var)3–7) and the repeat-associated small interfering RNA (or rasiRNA) silencing pathway (aubergine, homeless, armitage, and piwi). In contrast, TSE is not sensitive to mutations affecting r2d2, which is involved in the small interfering RNA (or siRNA) silencing pathway, nor is it sensitive to a mutation in loquacious, which is involved in the micro RNA (or miRNA) silencing pathway. These results, taken together with the recent discovery of TAS homologous small RNAs associated to PIWI proteins, support the proposition that TSE involves a repeat-associated small interfering RNA pathway linked to heterochromatin formation, which was co-opted by the P element to establish repression of its own transposition after its recent invasion of the D. melanogaster genome. Therefore, the study of TSE provides insight into the genetic properties of a germline-specific small RNA silencing pathway.
| The genome of the fruitfly was invaded in the last century by a mobile DNA element called the P element. After a transient period of genetic disorders due to P mobility, the P element established a repressive state for its transposition. We have shown that a major component of this repression comes from P copies inserted close to telomeres, the ends of linear chromosomes. One or two P copies inserted in subtelomeric heterochromatin (the DNA region highly compacted by protein complexes) can stabilize around 80 P copies. This finding allowed the discovery of a more general phenomenon called the “Trans-silencing effect” in which a transgene inserted in this subtelomeric heterochromatin represses, in the female germline, a homologous transgene, irrespective of the genetic location of the latter. We show that Trans-silencing requires not only the chromosomal copy of the telomeric silencer, but also a maternally transmitted factor whose influence can persist over generations. We have found that this epigenetic silencing is sensitive to mutations in genes involved in heterochromatin formation and in a recently discovered silencing pathway based on small RNAs. Trans-silencing thus provides a tool for mechanistic analysis of gene repression on the basis of chromatin changes combined with small RNA pathways in the germline.
| Repression of transposable elements (TEs) involves complex mechanisms that can be linked to either small RNA silencing pathways or chromatin structure modifications depending on the species and/or the TE family [1–6]. Drosophila species are particularly relevant to the study of these repression mechanisms since some families of TEs are recent invaders, allowing genetic analysis to be carried out on strains with or without these TEs [7,8]. In some cases, crossing these two types of strains induces hybrid dysgenesis, a syndrome of genetic abnormalities resulting from TE mobility [9,10]. In D. virilis, repression of hybrid dysgenesis has been correlated to RNA silencing since small RNAs of the retroelement Penelope, responsible for dysgenesis, were detected in nondysgenic embryos but not in dysgenic embryos [11]. In D. melanogaster, repression of retrotransposons can be established by noncoding fragments of the corresponding element (I factor [12,13], ZAM, and Idefix [14]) and can be in some cases (gypsy [15], mdg1 [16], copia [17], Het-A, TART [18,19], and ZAM, Idefix [C. Vaury and S. Desset, personal communication]) sensitive to mutations in genes from the Argonaute family involved in small RNA silencing pathways. In the same species, strong repression of the DNA P TE, by a cellular state that has been called “P cytotype” [10], can be established by one or two telomeric P elements inserted in heterochromatic “Telomeric Associated Sequences” (TAS) at the 1A cytological site corresponding to the left end of the X chromosome [20–24]. This includes repression of dysgenic sterility resulting from P transposition. We have previously shown that this P cytotype is sensitive to mutations affecting both Heterochromatin Protein 1 (HP1) [21] and the Argonaute family member AUBERGINE [25]. P repression corresponds to a new picture of TE repression shown, using an assay directly linked to transposition, to be affected by heterochromatin and small RNA silencing mutants.
In the course of the study of P cytotype, a new silencing phenomenon has been discovered. Indeed, a P-lacZ transgene or a single defective P element inserted in TAS can repress expression of euchromatic P-lacZ insertions in the female germline in trans, if a certain length of homology exists between telomeric and euchromatic insertions [23,26]. This homology-dependent silencing phenomenon has been termed Trans-Silencing Effect (TSE) [26]. Telomeric transgenes, but not centromeric transgenes, can be silencers and all euchromatic P-lacZ insertions tested can be targets [23,26]. TSE is restricted to the female germline (unpublished data) and has a maternal effect since repression occurs only when the telomeric transgene is maternally inherited [27]. Further, when TSE is not complete, variegating germline lacZ repression is observed from one egg chamber to another, suggesting a chromatin-based mechanism of repression [28]. Recently, an extensive analysis of small RNAs complexed with PIWI family proteins (AUBERGINE, PIWI, and AGO3) was performed in the Drosophila female germline [4]. The latter study showed that most of the RNA sequences associated to these proteins derive from TEs. TSE corresponds likely to such a situation.
Here, we analyze the genetic properties of TSE and show that it has an epigenetic transmission through meiosis, which involves an extrachromosomal maternally transmitted stimulating component. Further, in order to investigate the mechanism behind TSE, we performed a candidate gene analysis to identify genes whose mutations impair TSE. We found that TSE is strongly affected both by mutations in genes involved in heterochromatin formation and in the recently discovered small RNA silencing pathway called “repeat-associated small interfering RNAs” (rasiRNA) pathway [3,4,6,29]. In contrast, we show that TSE is not sensitive to genes specific to the classical RNA interference pathway linked to small interfering RNAs (siRNA) or to the micro RNA (miRNA) pathway. This suggests thus that TSE involves a rasiRNA pathway linked to heterochromatin formation and that such a mechanism, working in the germline, may underlie epigenetic transmission of repression through meiosis.
TSE was generated by combining telomeric transgene insertions (from the P-1152 line) as a silencer locus, with various euchromatic P-lacZ transgenes expressed in the germline as targets. Depending on the target, TSE can be almost total (Figure 1C) or intermediate (Figure 1E and 1F). When TSE is incomplete, variegation is observed since “on” and “off” lacZ expression is seen among egg chambers: egg chambers can show strong expression (dark blue) or no expression, but intermediate repression levels are not (or very rarely) found. In addition, a given ovary can present ovarioles showing all possible combinations of on or off egg chambers (Figure 1E). Simple quantification of TSE is thus possible by determining the percentage of repressed egg chambers (Figure 1F). We scored the number of repressed chambers among the first five egg chambers of a given ovariole for ten ovarioles chosen at random per ovary. For a given genotype, more than 1,000 egg chambers were classically counted (Table 1). This measure generally produces very reproducible results among replicate experiments allowing accurate quantification of TSE.
TSE was previously shown to have a maternal effect since strong repression occurs only in the progeny of crosses involving females carrying the telomeric silencer, whereas no or weak repression occurs if the telomeric silencer comes from the father [27,28]. Conversely, the parental origin of the target does not significantly affect TSE. We have thus tested whether TSE exhibits only a maternal effect (during one generation) or maternal inheritance over several generations as well. Two reciprocal G0 crosses were performed in which the telomeric silencer (P-1152) was maternally introduced (maternal lineage) or paternally introduced (paternal lineage) (Figure S1). Thus, the G1 females of the two lineages have the same genotype. These G1 females were backcrossed with males having the telomeric silencer and the target transgene in order to recover G2 females having one copy of the telomeric silencer locus and one copy of the target transgene. Such backcrosses were repeated for several generations in order to follow lineages with constant genotypes over generations. At each generation, the percentage of TSE was measured. Figure 2 shows that in the maternal lineage, strong TSE was detected in G1 (around 70%) and maintained over generations. By contrast, in G1 of the paternal lineage, the level of TSE was weak (around 10%), a result consistent with the maternal effect previously reported. However, in this paternal lineage, a strong level of repression was not observed in G2 females either (around 24%), despite the fact that G1 females were carrying a telomeric silencer. TSE thus shows a maternal inheritance. However, the level of TSE in G2 is greater than in G1 in the paternal lineage, and in succeeding generations TSE gradually increases. A total of six generations are necessary, however, to reach a repression level close to that of the maternal lineage. Thus, a memory of the initial maternal effect is observed over generations. In conclusion, TSE is characterized by maternal transmission that gets progressively reinforced over successive generations. It is therefore partially epigenetically transmitted through meiosis.
In the previous experiment, two main hypotheses can be proposed concerning the molecular basis of the difference inherited in G1 between the maternal and paternal lineages of TSE. First, maternally inherited telomeric transgenes may be imprinted while paternally inherited transgenes are not, and imprinting may be necessary for TSE to take place in the zygote. Second, a mother carrying telomeric transgenes may deposit an extrachromosomal factor in the oocyte, which is necessary for TSE to occur. This latter hypothesis was tested by using females hemizygous for the P-1152 telomeric silencer locus and a dominant genetic marker on the homologous chromosome to identify transmission of the chromosome carrying the telomeric silencer or the chromosome devoid of silencer (M5 balancer chromosome). Figure 3 shows that crossing these hemizygous females (“A” females) with males carrying a target transgene produced control G1 “B” females, which have inherited from their mother both the cytoplasm and a chromosomal copy of the telomeric silencer: in these females TSE is about 65%. However, sisters having inherited the M5 chromosome do not show any repression (“C” females, TSE = 0%). Thus, the cytoplasm of a P-1152 female without a chromosomal P-1152 copy cannot induce the TSE. Crossing P-1152 ; P-Z-target males with females devoid of telomeric silencer produces a weak repression in the progeny (2.7%) as shown by “E” females, a result consistent with the maternal effect of TSE reported previously [27]. Finally, crossing “A” females with males carrying a P-1152 telomeric silencer allows recovery of females having maternally inherited only a “P-1152” cytoplasm and paternally inherited a P-1152 chromosomal silencer. In that case, strong repression is observed (“D” females, 75% TSE). Thus the cytoplasmic component (incapable by itself of inducing TSE, as shown with “C” females) combined to a paternally inherited telomeric silencer can establish a strong TSE. Moreover, this repression is as strong as if the telomeric silencer was maternally inherited (“B” females). Consequently, the maternal effect of TSE cannot be attributed to a difference in imprinting between the maternally and paternally transmitted P-1152 telomeric silencers, but rather to an extrachromosomally transmitted factor likely deposited in the cytoplasm of the oocyte, which renders the telomeric silencers (paternally or maternally inherited) capable of establishing TSE.
Given the variegating phenotype of TSE and the interaction observed above between a maternally transmitted factor and the chromosomal copy of the telomeric silencer, it seemed possible that heterochromatin formation and RNA silencing could be involved in this repression. A candidate gene analysis was thus performed to identify genes whose mutations affect TSE among genes encoding heterochromatin components and actors of the small RNA silencing pathways. For a given assay, a P-1152 telomeric silencer was combined with a P-lacZ target expressed in the female germline, in the absence (TSE positive control) or presence of mutant alleles of the candidate gene. The P-1152 silencer was inherited in each case from a homozygous P-1152 female. When tested in the heterozygous state, the mutant allele was maternally inherited. The first gene tested was Su(var)205 (Figure 4A–4D), which encodes HP1 [30]. HP1 is a multifunctional protein that binds at centromeres, telomeres, and some scattered sites on the chromosomal arms [31,32]. The TSE positive control produced almost 90% repression (Figure 4B), whereas one copy of the null allele Su(var)2055, corresponding to an almost completely amorphic allele, has a strong negative dominant effect on TSE (15.8% TSE remaining, Figure 4C). The allele Su(var)2054 encoding a truncated HP1 protein also behaves as a genetically null allele and also strongly impairs TSE (29.5% TSE remaining, Figure 4D). TSE is therefore sensitive to the dose of HP1. The same analysis was performed for SU(VAR)3–7, another nonhistone heterochromatin protein that binds at centromeres and telomeres and is a partner of HP1 [33,34]. SU(VAR)3–7 contains seven zinc finger motifs, which were shown to bind DNA in vitro [35,36]. Again a dose effect was observed since females heterozygous for the Su(var)3–7R2A8 null allele showed a reduced TSE (less than 40%, Table 1). Since Su(var)3–7 mutants can be homozygous viable, mutations in this gene were also tested at the homozygous state and an almost complete loss of TSE was observed (3.3% TSE remaining, Table 1). Such a result was also obtained with a second Su(var)3–7 mutant allele (Su(var)3–7R14, Table 1). Su(var)3–9 encodes a histone methyl transferase (HMT) responsible for the methylation of histone H3 on lysine 9 [37]. This protein is also a partner of HP1. The null allele of this gene tested here at the homozygous state had no significant effect on TSE (58.5% in the mutant versus 62.9% in the wild-type control, Table 1).
We further tested genes involved in RNA silencing. Three primary RNA silencing pathways have been discovered so far: miRNA, siRNA (classically termed “RNA interference”), and rasiRNA pathways. These pathways may partially overlap, since they have certain actors in common, but they differ in terms of the biogenesis of short RNAs. miRNAs derive from hairpin RNAs and target numerous essential genes; siRNAs derive from double-strand RNAs and can serve as a defence mechanism against parasites such as viruses; and rasiRNAs derive from interactions between two complementary types of PIWI-interacting RNAs and have numerous targets but are more specific of TEs and heterochromatic sequences [3,4,6,29]. We tested three genes (aubergine [aub], homeless [hls], and armitage [armi]) involved in both the siRNA pathway [38,39] and the rasiRNA pathway in Drosophila ovaries [3,4,29]. Mutations of these genes can induce defaults in siRNA-guided cleavage or in the production of rasiRNAs depending on the target of the silencing [3,39,40]. These mutations are also responsible for disruption of embryonic axis specification linked to disturbance of microtubule polarisation, but this effect is mediated by a mechanism different from rasiRNA silencing itself [41]. The first gene tested is aub, an RNA binding protein and member of the Argonaute family (Figure 4E–4H) [42]. No significant dose effect of aub was detected (Figure 4G and unpublished data), but the two heteroallelic aub mutant genotypes tested completely abolish TSE (Figure 4H; Table 1). The second one is hls, which encodes an RNA helicase [43]. Again no dose effect was observed (unpublished data), but a heteroallelic hls mutant genotype completely abolishes TSE (Table 1). The third one is armi, which encodes a putative RNA helicase [44,45]. No dose effect was observed (unpublished data), but a complete loss of TSE was observed in a heteroallelic mutant context (Table 1). Finally piwi, another member of the Argonaute family [46] shown to be involved in the rasiRNA pathway [3,4,6,29,47] was tested. This gene is also necessary for germline stem-cell renewal [48] and was shown to be required for post-transcriptional transgene silencing (a phenomenon termed cosuppression)[49]. No dose effect on TSE was observed (unpublished data). Because piwi has deleterious effects on ovary structure at the homozygous state, we tested the effects of piwi mutant alleles at the heterozygous state in combination with a Su(var)205 mutant. In these combinations, piwi mutant alleles aggravate the negative effect of reduction of the dose of HP1: the single Su(var)205 mutant shows 29.5% TSE, whereas the double heterozygotes show 10.0% and 5.6% for piwi1 and piwi2 respectively (Table 1). These results indicate that the siRNA and/or the rasiRNA pathways are involved in TSE.
To determine which of these two pathways is responsible for TSE, several other mutations were tested. r2d2 is involved in the siRNA, but not the rasiRNA pathway [3,50,51]. R2D2 is a double-stranded RNA (dsRNA) binding protein involved in siRNA loading onto RISC [50]. In ovaries of females homozygous for the r2d21 null allele, the siRNA pathway is severely affected, as shown by a dsRNA-initiated RISC (or RNA-induced silencing complex) assay [52]. These ovaries show somewhat abnormal ovarioles (Figure S2), but quantification of TSE was nonetheless possible. The loss of function for r2d2 does not affect TSE (Table 1). A second protein DICER-2 (DCR-2), which combined with R2D2 is responsible for dsRNA cleavage, is also involved specifically in the siRNA pathway [53]. The loss of function of Dcr-2 is fertile. Unfortunately, we were not able to test the effect of a Dcr-2 mutant allele on TSE. Indeed, the control staining (in absence of a telomeric silencer) of the two different P-lacZ target transgenes that we tested in homozygous Dcr-2L811fsX mutants intriguingly showed almost no staining in ovaries, thus making it impossible to assay any target repression by a telomeric silencer in this context. Finally, a mutant allele of loquacious (loqs), a dsRNA binding protein involved in the miRNA pathway was tested. The loqsf00791 allele corresponds to a hypomorphic allele that alters miRNA-induced silencing in all tissues where it has been tested, including ovaries [54]. This allele is viable, allowing us to test homozygous mutant females. No effect on TSE was detected (Table 1).
In conclusion, TSE strongly depends on the function of two genes involved in heterochromatin formation and four genes involved in the rasiRNA silencing pathway but is not affected by mutations of genes involved in the siRNA or miRNA pathways.
Our study of the regulatory mechanism of a TE that recently invaded the D. melanogaster genome allowed the discovery of a master site for establishing strong transposon repression in the germline; indeed, one or two P elements inserted in TASs are sufficient to repress the mobility of the whole P family [21]. TSE was further discovered [26], a mechanism by which a transgene inserted in subtelomeric heterochromatin has the capacity to repress a homologous transgene located in euchromatin. This phenomenon appears general since it can be induced by transgenes inserted at the telomere of the X chromosome as well as autosomal telomeres, and the targets can be located on all major chromosomes [26,28]. TSE corresponds thus to a useful tool to investigate the properties of the telomeric master site of P repression. Here, we show that TSE requires genes encoding proteins involved in heterochromatin formation (including HP1, a major component of heterochromatin) and proteins involved in the rasiRNA silencing pathway. Another TE master regulatory locus has been described in Drosophila close to the centromere of the X chromosome. It was first shown to regulate the gypsy retrotransposon and referred to as the flamenco locus [55]. This locus was further characterized as regulating together two other retrotransposons, ZAM and Idefix, and proposed as a repression center for multiple TEs, known as “COM” (Centre Organisateur de Mobilisation) [14]. Repression by this locus is based on RNA silencing, but differs from TSE in that repression of gypsy, ZAM, and Idefix occurs in the somatic follicle cells of the ovary. This point is important since in somatic follicle cells, AUBERGINE, a major actor of the rasiRNA pathway, is not expressed [4]. Conversely, in the germline, where TSE is active, all known components of the rasiRNA machinery (AUBERGINE, PIWI, and AGO3) are present. Consequently, TSE is particularly appropriate to investigate the genetic properties of a complete rasiRNA-based repression machinery.
Here, we show that TSE exhibits a maternal memory that can be detected for six generations and has thus a partial epigenetic transmission through meiosis (Figure 2). The genetic analysis indicates that potentialization of the telomeric silencer (“potentialization step”) appears to be a prerequisite for repression of the target transgene by the telomeric silencer (“target repression step”). For the potentialization of the telomeric silencer, both an extrachromosomal maternally transmitted factor, produced by females carrying a telomeric silencer, and a chromosomal copy of the telomeric silencer are necessary (Figure 3). The molecular nature of this extrachromosomal maternally transmitted factor, responsible for the maternal effect, is not known at the moment but it may correspond to rasiRNAs since TAS homologous rasiRNAs have been reported (see Table S1 of [4]). We have attempted to identify short RNAs made by telomeric transgenes by northern blot on ovaries but did not detect them; this may be due to a low concentration of such short RNAs. We propose two nonmutually exclusive mechanisms. In the first, which derives from the “rasiRNA positive loop model” proposed by Brennecke et al. (“ping-pong model”) [4] and Gunawardane et al. [29], females deposit PIWI-interacting RNAs (piRNAs) in the oocyte produced by telomeric transgenes inserted in tandemly repeated TAS heterochromatic sequences. As proposed in their model, these RNAs would be primarily small antisense RNAs associated with PIWI or AUBERGINE. These piRNAs would interact in the embryo with sense RNAs produced by the telomeric transgenes and result in the production of small sense RNAs associated with AGO3, which in turn would promote the production of a new antisense piRNA, resulting in a positive feedback loop. Such a loop can explain the two-component interaction shown in Figure 3 (see below).
A second mechanism can be proposed for the potentialization step. Indeed, RNA silencing and heterochromatin formation have been shown to be connected since “short RNA-dependent heterochromatin formation” pathways have been described in several species (fission yeast [56], ciliates [57], plants [58], and Drosophila [59], for a review see [60–62]). According to this model, maternally transmitted small RNA molecules may modify the chromatin structure of the chromosomal copy of the (paternally or maternally transmitted) telomeric transgenes in the embryo and thereby confer to the telomeric transgene the capacity to be a silencer. Recently, RNA-dependent heterochromatin formation was observed at Drosophila telomeres since mutations of homeless were shown to cause an opening of chromatin at the level of telomeric retroelements in Drosophila ovaries (M. Klenov and V. Gvozdev, personal communication). Such a model is consistent with the results of our candidate gene analysis since mutations in two genes involved in heterochromatin formation (Su(var)205 and Su(var)3–7) strongly impair TSE. HP1 could play a role in heterochromatin formation, not only as a component of heterochromatin but also indirectly: indeed, in Schizosaccharomyces pombe CHP1 (a chromodomain protein) has been shown to be part of a complex that drives small RNAs to chromatin in order to modify histone H3 methylation [63]. Such a role for the HP1 chromodomain protein in a similar complex could also exist in Drosophila.
How can maternal memory of TSE over six generations be explained? Because of the modality of TSE transmission over several generations, it can be proposed that establishment of the hypothetical extrachromosomal component discussed above requires a number of generations to reach a sufficient concentration to establish strong repression. Under the “rasiRNA positive loop model” proposed by Brennecke et al. [4] and Gunawardane et al. [29], when males carrying a telomeric silencer are crossed with females bearing no silencer, the cytoplasm of the oocytes would lack piRNAs homologous to the transgene. However, synthesis of these RNAs would begin in these G1 females stimulating this pathway and increasing the piRNA concentration. At each generation, females would transmit to their daughters a higher concentration of rasiRNA molecules than at the previous generation. At generation six, a sufficient level would be reached to allow strong TSE. Under the “RNA-dependent heterochromatin formation” model, when males carrying a telomeric silencer are crossed with females bearing no silencer, again the cytoplasm of the oocytes would lack small RNAs homologous to the transgene. However, synthesis of these RNAs would begin in these G1 females stimulating heterochromatinization of the telomeric silencer. This change in the chromatin state would stimulate the production of rasiRNAs at the locus also establishing a positive loop, i.e., reciprocal stimulation between heterochromatin formation and production of small RNAs. Such a reciprocal positive loop between heterochromatin formation and small RNA silencing has been proposed for gene silencing in S. pombe [64,65]. Again, at each generation, females would receive from their mother and transmit to their daughters a higher concentration of RNA molecules than at the previous generation until a sufficient level is reached at generation six.
Repression of the target transgene by the telomeric silencer was shown to be homology dependent [23,26,28]. TSE however does not require homology at the level of the promoters of the telomeric and euchromatic insertions since a defective telomeric P element lacking the P promoter can repress a P-lacZ transgene in which lacZ is driven by a heat-shock promoter [23]. The TSE decision appears to be established at the cystoblast stage and stably maintained since the 15 deriving nurse cells are roughly identical in regard to their repression state. Another important point in understanding the repression mechanism is that the telomeric silencer can lack the lacZ gene, i.e., the gene repressed by TSE in the target transgene. For example, a P-white-yellow transgene [66] inserted at 1A can repress the euchromatic P-otu-lacZ-white (pCo) transgene used in this study (unpublished data). In this case, the homology between the telomeric and the euchromatic transgenes comes from the white transformation marker. This rules out the hypothesis that silencing of the target occurs via an interaction between rasiRNAs deriving from the telomeric silencer and the target transgene transcript; indeed, the targeted transcript in TSE corresponds to the lacZ sequence and no lacZ rasiRNAs can be produced by the silencer. Therefore, TSE would not occur by post-transcriptional gene silencing.
Target repression could thus occur via transcriptional gene silencing. This would fit with the variegated phenotype of repression observed. Such variegation strongly suggests that repressed targets undergo heterochromatinization. Two hypotheses can be proposed to explain such heterochromatinization. First, rasiRNAs produced by the telomeric silencer would interact with nascent transcripts on the target, and this interaction could provoke a local heterochromatinization of the target [64,65]. Heterochromatinization may then spread along the target as suggested by the repression of a P-otu-lacZ-white target by a telomeric P-white-yellow transgene. In this case heterochromatinization would start on the white sequence of the target and subsequently reach the lacZ sequence. HP1 and SU(VAR)3–7 could be involved in this spreading since centromeric heterochromatin spreading has been shown to be sensitive to the dose of these proteins [35,67]. An alternative hypothesis for target heterochromatinization is a DNA–DNA interaction between the telomeric and the euchromatic transgenes, leading to their pairing and to trans-heterochromatinization of the target following this pairing. Indeed, telomeric silencers are themselves in a heterochromatic state due to the heterochromatic nature of the TAS [68–71]. This is illustrated by a phenomenon referred to as telomeric position effect, in which a transgene inserted in TAS shows variegation in the eye for the transformation marker [72–75]. Again HP1 and SU(VAR)3–7 could play a role in this trans-heterochromatinization.
Drosophila telomeres can be divided into three domains with respect to chromatin structure [71]: (1) the subtelomeric cluster of heterochromatic TAS repeats [68–70] in which the tested P elements or P transgenes are inserted; (2) distal to the TAS, a telomeric retrotransposon array tandemly repeated in the same orientation [76–80], the retrotransposon array is partially euchromatic [71], and transgenes located inside this domain do not show variegation for an eye marker [81]; and (3) at the extremity of the retrotransposon array, the telomere protein cap that prevents telomere fusion and regulates telomeric retrotransposon transcription and transposition [82–85]. Interestingly, HP1 (Su(var)205) and SU(VAR)3–7, both implicated in TSE, are part of the cap and therefore could be required for the telomeric locus to be a silencer. This hypothesis is supported by the analysis of the different Su(var)205 alleles. In addition to the two null alleles of Su(var)205 tested here, which showed a strong effect on TSE (Table 1), we also tested a deficiency and the Su(var)2052 allele corresponding to a mutation in the chromodomain (involved in the histone H3 K9-methylated binding), which leaves the capping activity intact [85,86]. The deletion of Su(var)205 as expected has a strong effect on TSE but Su(var)2052 has no effect (unpublished data). This suggests that HP1 effect can be mediated (at least partially) via its capping activity. Therefore, HP1 and SU(VAR)3–7 could affect the telomeric transgenes in TAS indirectly via such capping activity, for example, by affecting the localisation of the telomere inside the nucleus.
Finally, it could appear paradoxical that an amorphic mutant allele of the Su(var)3–9 HMT has no effect on TSE since RNA-dependent heterochromatin formation in most cases involves the histone H3 methylation transition from lysine 4 to lysine 9 to establish the link between short RNAs and formation of heterochromatin. However, this result can be explained by the fact that this HMT has been recently shown not to be responsible for histone H3 methylation on lysine 9 at Drosophila telomeres [85]; another unidentified HMT likely plays this role. We are therefore pursuing the candidate gene analysis in order to identify a HMT involved in TSE.
Telomeric TSE appears to be a complex repression mechanism that requires genes involved in heterochromatin formation and in RNA silencing. In Drosophila, interaction between small RNA silencing pathways and transcriptional repression was previously shown to exist in somatic tissues for cosuppression phenomena or for variegation phenotypes with the white marker in the eye [49,59]. TSE shows that interaction between RNA silencing and heterochromatin formation can also occur in the germline. This type of silencing appears to be the basic mechanism for P-element repression [28], although some P-encoded repressors encoded by euchromatic copies may contribute in some cases to P repression [10,87]. TSE has the same complex inheritance as the P cytotype [10,88,89]. The subtelomeric heterochromatin thus represents a piRNA producing “platform” [4] available for recent invaders of the genomes to establish their own repression. The study of TSE illustrates the genetic properties of such a platform.
All crosses were performed at 25 °C and involved three to five couples in most cases. All ovary lacZ expression assays were carried out using X-gal overnight staining as described in Lemaitre et al. 1993 [90], except that ovaries were fixed for 6 min.
P-lacZ fusion enhancer trap transgenes (P-1152, BQ16, and BC69) contain an in-frame translational fusion of the Escherichia coli lacZ gene to the second exon of the P transposase gene and contain rosy+ as a transformation marker [91]. The P-1152 insertion comes from stock number 11152 of the Bloomington Stock Center (http://flystocks.bio.indiana.edu) and was mapped at the telomere of the X chromosome (site 1A); this stock was previously described to carry a single P-lacZ insertion in TAS [26]. However, in our number 11152 stock, we have mapped two P-lacZ insertions in the same TAS unit and in the same orientation that might have resulted from an unequal recombination event duplicating the P-lacZ transgene. P-1152 is homozygous viable and fertile. BQ16 is located at 64C in euchromatin of the third chromosome and is homozygous viable and fertile. BC69 is inserted on Chromosome 2 in the first exon of the vasa gene and results in a vasa loss of function; it is consequently homozygous female sterile. P-1152 shows no lacZ expression in the ovary (Figure 1A), whereas BQ16 and BC69 are strongly expressed in the nurse cells and in the oocyte (Figure 1B and 1D, respectively). P-Co1 is an insertion of the pCo transgene (P-otu-lacZ) on the third chromosome (87A-B), which is homozygous viable and fertile. β-galactosidase expression of the pCo transgene is driven by the otu promoter and is therefore strongly detected in both nurse cells and the mature oocyte [27]. This transgene contains a white gene as a transformation marker.
Lines carrying transgenes have M genetic backgrounds (devoid of P TEs), as well as the multimarked balancer stocks used in genetic experiments (yw; Cy; TM3Sb / T(2;3)apXa, M5; Cy / T(2;3)apXa and M5; TM3Sb/ T(2;3)apXa) and the strains carrying mutations used for the candidate gene analysis (see below).
Su(var)205, aub, piwi, r2d2, and loqs genes are located on Chromosome 2, whereas Su(var)3–7, Su(var)3–9, hls (synonym of spindle-E), and armi are located on Chromosome 3. Loss of function is lethal in the case of Su(var)205 and loqs, female sterile in the case of aub, piwi, hls, r2d2, and armi, and is viable and fertile in the case of Su(var)3–9. Loss of function of Su(var)3–7 is viable when the mother is heterozygous mutant but lethal when the mother is homozygous mutant [92].
Su(var)2–505 was x-ray induced and corresponds to a null allele since it only encodes the first ten amino acids of the HP1 protein [67]. Su(var)2–504 was EMS induced and encodes a truncated HP1 protein, ∼85% of wild type size [67], missing a domain necessary for targeting of HP1 to the nucleus [93]. Lines carrying these mutations were kindly provided by Gunter Reuter.
Su(var)3–7R2A8 and Su(var)3–7R14 were generated by homologous recombination [92,94]. Su(var)3–7R14 produces a chimeric protein and Su(var)3–7R2A8 deletes most of the protein-coding sequence of the gene. These two alleles behave as genetic null mutations. Lines carrying these mutations were kindly provided by Marion Delattre.
Su(var)3–906 was generated by x-ray and corresponds to an amorphic allele [37,95]. Line carrying this mutation was kindly provided by Gunter Reuter.
Three strong mutant alleles of aub induced by EMS were used. All of them are homozygous female sterile. aubQC42 [96] comes from the Bloomington Stock Center (stock number 4968). aubHN2 [96] has an amino acid substitution. aubN11 [97] has a 154-bp deletion, resulting in a frameshift that is predicted to add 16 novel amino acids after residue 740 [98]. aubHN2 and aubN11 were kindly provided by Paul Macdonald.
piwi1 and piwi2 are P-element–induced mutations; piwi1 contains a P transgene inserted in the first exon [46]. The orientation of the transgene is opposite to that of the gene. piwi2 contains a P transgene inserted in exon 4 in the same orientation as the gene. In both cases, homozygous females are sterile. Lines carrying these mutations were kindly provided by Alain Bucheton.
hlsE616 (synonym of spn-E1) was generated by EMS [43,99]. hlsΔ125 (synonym of spn-EhlsΔ125) was generated by P-element excision resulting in deletion of coding sequences and adjacent sequences. Deletion may extend into other genes [43]. Lines carrying these mutations were kindly provided by Utpal Bhadra.
armi1 and armi72.1 are P-element–induced mutations; armi1 has an insertion of a P transgene in the 5′ UTR of the armitage gene [44]. armi72.1 derives from armi1 by an imprecise excision of the P element, resulting in a deletion of armi sequences in the 5′ untranslated region [39]. These two alleles are female sterile and come from the Bloomington Stock Center (stock numbers 8513 and 8544).
r2d21 results from a deletion induced by the remobilization of a P transgene located in 5′ of the r2d2 gene [50]. The 4.9-kb deletion removes the entire r2d2 open reading frame [52]. This allele is an amorphic allele and homozygous females are sterile due to abnormal ovaries. The siRNA pathway is severely affected as shown by a dsRNA-initiated RISC assay in ovaries of homozygous mutant females [52]. This line comes from the Bloomington Stock Center (stock number 8518).
loqsf00791 results from a piggyBac transgene insertion 57 bp upstream of the transcription start site of loqs [54], a gene necessary for miRNA-directed silencing. loqsf00791 is a hypomorphic allele that is homozygous viable. At the homozygous state, this allele has a strong effect on miRNA processing in ovaries. It also has a negative effect on siRNA silencing [54]. This allele comes from the Bloomington Stock Center (number 18371).
All the alleles described above are maintained over a balancer chromosome. Additional information about mutants and stocks are available at FlyBase (http://flybase.bio.indiana.edu).
The National Center for Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov) and FlyBase (http://flybase.bio.indiana.edu) accession numbers for the genes described in this article are (respectively): armitage (CG11513, FBgn0041164); Su(var)205 (CG8409, FBgn0003607); aubergine (CG6137, FBgn0000146); piwi (CG6122, FBgn0004872); r2d2 (CG7138, FBgn0031951); loquacious (CG6866, FBgn0032515); Su(var)3–7 (CG8599, FBgn0003598); Su(var)3–9 (CG6476, FBgn0003600); and homeless (CG3158, FBgn0003483). For the P element, see FBgn0003055. For general properties of TAS, see FBgn0041614. For TAS data sequences, see LO3284, U58967, U58968, and U58969. Information about X-TAS piRNAs are in Table S1 of [4]. |
10.1371/journal.pgen.1003378 | Association Mapping and the Genomic Consequences of Selection in Sunflower | The combination of large-scale population genomic analyses and trait-based mapping approaches has the potential to provide novel insights into the evolutionary history and genome organization of crop plants. Here, we describe the detailed genotypic and phenotypic analysis of a sunflower (Helianthus annuus L.) association mapping population that captures nearly 90% of the allelic diversity present within the cultivated sunflower germplasm collection. We used these data to characterize overall patterns of genomic diversity and to perform association analyses on plant architecture (i.e., branching) and flowering time, successfully identifying numerous associations underlying these agronomically and evolutionarily important traits. Overall, we found variable levels of linkage disequilibrium (LD) across the genome. In general, islands of elevated LD correspond to genomic regions underlying traits that are known to have been targeted by selection during the evolution of cultivated sunflower. In many cases, these regions also showed significantly elevated levels of differentiation between the two major sunflower breeding groups, consistent with the occurrence of divergence due to strong selection. One of these regions, which harbors a major branching locus, spans a surprisingly long genetic interval (ca. 25 cM), indicating the occurrence of an extended selective sweep in an otherwise recombinogenic interval.
| Selection during the evolution of crop plants has resulted in dramatic phenotypic differentiation, and these same selective pressures are expected to have had a significant impact on underlying genomic diversity. Population genomic analyses, especially when coupled with trait-based mapping approaches, thus have the potential to provide unique insights into the evolution of crop plants and their genomes. In this study, we performed a genome-wide analysis of genetic variation in cultivated sunflower and used the resulting data to genetically dissect variation in plant architecture (i.e., branching) and flowering time. We found substantial variation in levels of linkage disequilibrium (LD) across the genome, with islands of elevated LD generally corresponding to genomic regions underlying traits that have been targeted by selection during the evolution of cultivated sunflower. A number of these same regions also exhibited strong population genetic differentiation across the sunflower gene pool, suggesting that they may harbor genes underlying adaptation following domestication. Our analyses also identified numerous genomic regions underlying variation in both plant architecture and flowering time, many of which fall in genomic regions that have not previously been shown to influence these traits using more traditional quantitative genetic approaches.
| Strong selection during the evolution of crop plants has resulted in dramatic phenotypic differentiation. Undoubtedly, these same selective pressures will also have produced significant genomic consequences. Indeed, genomic regions that have been targeted by selection during crop evolution are expected to exhibit characteristic changes in their levels and/or patterns of nucleotide diversity. For example, under strong directional selection, one would expect a marked decrease in genetic variation in and near the targeted loci (e.g., tb1 [1]; waxy [2]; GIF1 [3]). However, under divergent selection, an increase in population genetic differentiation would be expected between the divergently selected lineages, coupled with localized decreases in genetic variation (e.g., [4], [5]). The extent of these effects will be jointly determined by the strength of selection and the local recombination rate [6], with stronger selection and/or reduced recombination affecting diversity across a larger chromosomal region. The use of large-scale population genomic analyses, especially when coupled with trait-based mapping approaches, thus has the potential to provide novel insights into the evolutionary history of crop plants and their genomes.
Traditional quantitative trait locus (QTL) mapping analyses have provided considerable insight into the genetic basis of the phenotypic changes that have occurred during crop evolution (e.g., [7]–[9]). This general approach is, however, somewhat limited in terms of both mapping resolution and the amount of diversity assayed. Association mapping, which involves the correlation of molecular polymorphisms with phenotypic variation in a diverse assemblage of individuals, solves both of these problems, and is thus a useful alternative to standard QTL mapping approaches [10]. Because association populations typically capture many generations of historical recombination, linkage disequilibrium (LD; the non-random association of alleles between loci) is expected to be substantially lower than in family-based mapping populations, resulting in much higher mapping resolution. Moreover, the high level of diversity in a typical association mapping population allows for the simultaneous investigation of the effects of a broad spectrum of alleles across multiple genetic backgrounds. The downside of such analyses is that structure in the focal population can produce spurious marker/trait correlations in the absence of physical linkage [11], [12]. Statistical advances have, however, made it possible to minimize the likelihood of false associations by accounting for relatedness amongst individuals (i.e., kinship) and population structure (e.g., [13]–[15]).
Ultimately, detailed insights into standing levels of nucleotide diversity, background patterns of LD, and relatedness amongst individuals within the focal population are critically important for the successful application of association mapping approaches. The use of high-density single nucleotide polymorphism (SNP) data derived from known genomic locations can facilitate the development of these insights and thus has the potential to enable the genetic dissection of important phenotypes in crop plants. Moreover, the outcomes of such analyses have potential downstream applications in marker-assisted breeding programs [16]. Here, we investigate SNP diversity, population differentiation, and the structure of LD across the 3.6 Gbp genome of sunflower (Helianthus annuus L.) and perform association analyses of plant architecture and flowering time in this valuable crop species.
Cultivated sunflower is a globally important oilseed crop that was domesticated from the wild, common sunflower (also H. annuus) approximately 4,000 years ago by Native Americans. Following its domestication, sunflower was originally used as a source of edible seeds and for a variety non-food applications (e.g., as a source of dye for textiles and for ceremonial purposes) [17]–[19]. The transformation of sunflower into an oilseed crop began in 18th century in Eastern Europe where breeding efforts increasingly focused on improving oil yield in a subset of the available germplasm. Commercial production commenced in North America in the mid-20th century, along with a focus on the development of sunflower as a hybrid crop. Modern sunflower is maintained in two primary breeding pools: the unbranched female (A) lines (differing only in cytoplasm from paired maintainer or B lines), and the typically recessively-branching, multi-headed male restorer (R) lines that are crossed to generate the unbranched, fertile hybrids grown by producers. In this study, we used an Illumina Infinium 10 k SNP array [20], [21] to genotype a diverse collection of publicly-available sunflower lines. We then used these data, along with phenotypic data collected from multiple locations, to analyze genome-wide patterns of genetic variation, characterize the extent of LD and population differentiation, and investigate the genetic basis of variation in plant architecture and flowering time.
The sunflower association mapping population utilized in this study was composed of 271 lines that have previously been shown to capture nearly 90% of the allelic diversity present within the cultivated sunflower gene pool [22]. This population is composed of accessions from the collections held by both the USDA North Central Regional Plant Introduction Station (NCRPIS) and the French National Institute for Agricultural Research (INRA) (Table S1). These accessions include numerous inbred lines and historically important open-pollinated varieties (OPVs; including high-oil Eastern European cultivars), as well as oilseed and confectionery (non-oil) accessions from elsewhere in the world. Where necessary, accessions were advanced via single-seed descent for one or two generations to minimize residual heterozygosity.
All accessions were assigned to one of ten categories based on their origin (USDA or INRA), breeding history (maintainer [B] lines = HA, typically unbranched; restorer [R] lines = RHA, typically branched), and agronomic use (oil vs. non-oil). Note that an oilseed vs. confectionery designation was not available for the INRA accessions; therefore, these were divided into INRA-derived B and R lines (denoted INRA-HA and INRA-RHA, respectively). For the USDA accessions, the following categories were defined: HA non-oil, HA oil, RHA non-oil, RHA oil, introgressed, OPVs, other non-oil, and other oil. Accessions designated ‘non-oil’ were either confectionery types, or could not be clearly defined as being oil types. The ‘introgressed’ category included accessions with a recent history of introgression from wild Helianthus species as indicated by the available pedigree information (e.g., [23], [24]). The OPV category included named sunflower accessions that represent open-pollinated varieties of the pre-hybrid era of sunflower breeding, including Jupiter, Manchurian, Jumbo, VIR 847, Mammoth, etc. (BS Hulke, USDA-ARS, pers. comm.) along with two Native American landraces, Hopi and Mandan. The ‘other oil’ and ‘other non-oil’ categories included accessions of each type for which a B vs. R designation could not be made.
In the spring of 2010, we planted our association mapping population in replicate at three locations: the Plant Sciences Farm in Watkinsville, GA, USA, the North Central Regional Plant Introduction Station in Ames, IA, USA, and the University of British Columbia Campus in Vancouver, BC, Canada (12 seeds/plot x 271 lines x 2 replicates x 3 locales) (Figure S1). Replicates were planted in an alpha lattice design constructed using the computer module ALPHA 6.0, available from Design Computing (http://www.designcomputing.net/).
Total DNA was extracted from bulked tissue collected from four individuals of each line using a CTAB extraction protocol [25]. Total DNA was quantified using Picogreen (ABI), and the quality of DNA was inspected using a Nanodrop 1000 spectrophotometer. All lines were then genotyped on an Illumina Infinium 10 k SNP array designed for cultivated sunflower. The array was designed from a large collection of sunflower ESTs and included no more than one SNP per gene [20]. Genotyping was performed according to the manufacturer's recommendations on the Illumina iScan System (Illumina Inc., San Diego, CA) at the Emory University Biomarker Service Center. Prior to hybridization of the Beadchips, DNA was diluted to 50 ng/µl and quality was assessed via UV spectrophotometry and agarose gel electrophoresis. All SNP data analyses were performed using the raw intensity data from the Illumina Beadchip and Genome Studio ver. 2011.1 (Illumina) following the methods outlined in Bowers et al. [21]. Note that only those SNPs that showed clearly interpretable clustering patterns were used in this study, thereby eliminating probes that hybridized to multiple gene copies from our dataset. Map positions were obtained from the sunflower consensus map [21].
Population-wide estimates of genetic diversity, including allele frequencies, observed heterozygosity, and unbiased gene diversity [26], were calculated using GenAlEx v. 6.4 [27].
Population structure was investigated using the Bayesian, model-based clustering algorithm implemented in the software package STRUCTURE [28]. For this analysis, we used only polymorphic SNPs with a minor allele frequency (MAF) ≥10%. Briefly, individuals were assigned to K population genetic clusters based on their multi-locus genotypes. Clusters were assembled so as to minimize intra-cluster Hardy-Weinberg and linkage disequilibrium and, for each individual, the proportion of membership in each cluster is estimated. We employed the admixture model without the use of prior population information (i.e., USEPOPINFO was turned off). For each analysis, we evaluated K = 1–12 population genetic clusters with 5 runs per K value and averaged the probability values across runs for each cluster. For each run, the initial burn-in period was set to 50,000 with 100,000 MCMC iterations. The most likely number of clusters was then determined using the DeltaK method of Evanno et al. [29].
Genetic relationships amongst the cultivated sunflower accessions were also investigated graphically via principal coordinates analysis (PCoA) using GenAlEx and the same set of polymorphic SNPs (MAF ≥10%) that were used for the STRUCTURE analyses. A standard genetic distance matrix [26] was constructed based on the multi-locus genotypes. This distance matrix was then used for the PCoA, and the first two principal coordinates were graphed in two-dimensional space.
A relative kinship matrix was then estimated from this set of SNPs using the program SPAGeDi [30]. Negative values between pairs of individuals, indicating that there was less relationship than that expected between two randomly chosen individuals, were set to 0 in the resulting matrix.
To investigate the extent of linkage disequilibrium (LD) across the genome, a correlation matrix of r2 values, the squared allele frequency correlations, was constructed between all possible pairs of polymorphic loci with MAF ≥10%. Following the methods of Macdonald et al. [31], we summarized the observed r2 values using the k-smooth function in the statistical programming language R (http://www.R-project.org/). We also visualized the extent of LD and genetic variation across the genome by averaging the r2 and UHe values, respectively, in a 5 cM sliding window across each linkage group (LG).
We calculated FST for all polymorphic SNPs between the two major heterotic groups (including 125 RHA lines and 100 HA lines) and plotted the results as a function of map position. We also performed outlier analyses on these data using the software BayeScan [32]. The program utilizes the Bayesian model from Beaumont and Balding (2004) and a reversible jump Markov chain Monte Carlo method to identify outlier loci that are putatively under selection based on FST estimates. Ten pilot runs of 5,000 iterations and an additional burn-in of 50,000 iterations were first performed. We then used 100,000 iterations to identify loci under selection based on locus-specific Bayes factors. Strong evidence for selection is indicated by a Bayes factor above 10, or log10 = 1.0 [32].
In order to visualize genome-wide haplotypic structure in the association population, graphical genotypes were constructed by defining haplotypic blocks of 25 or more consecutive SNPs (based on the map order from [21]) that were identical between two or more cultivars. To do this, each accession was compared to all other accessions in the dataset to determine the percentage of the SNPs contained in shared haplotypic blocks. The accession with the highest fraction of the genome shared with all other accessions in the data set was set as the “template” for genotype #1 (G1). The raw data for the template accession and all matching haplotypic blocks in other accessions were converted into G1 from the raw scores. This process was repeated for 24 additional cycles, masking the data that had previously been assigned to haplotypic blocks to produce G2, G3, …, G25. The most common genotypes were then color-coded and visually presented using spreadsheet software.
The number of days to flower (DTF; calculated from the planting date) was recorded at the R-5.1 reproductive stage. The R-5 stage commences at the onset of flowering, and is divided into substages according to the percentage of disc florets that have opened; R-5.1 corresponds to the stage at which 10% of the disc florets have opened. The total number of branches per plant (hereafter referred to as “branching”) was measured in the field at the R-9 reproductive stage. This stage is regarded as physiological maturity and is characterized by the presence of yellow/brown bracts on the back of the sunflower head. Four plants per accession were scored for each replicate at each of three locations (4×271×2×3 = 6,504 plants scored).
Data were analyzed using SAS software version 9.3 (SAS Institute, Cary, NC). We calculated Pearson pairwise correlation coefficients for the branching data and the average DTF across locations using PROC CORR and corrected the resulting significance levels for multiple tests using a sequential Bonferroni correction (Holm 1979). We also analyzed our data using the GLM procedure of SAS. Because our initial analyses revealed highly significant (P<0.001) genotype x environment (G×E) interactions (data not shown), all subsequent analyses were performed separately by location. At each location, the entry (i.e., genotype) was treated as a fixed effect and blocks and reps were treated as random effects. For branching, the block and rep effects were not significant in the model and thus raw means were used for association testing (below). Because there were significant block and rep effects for DTF, the least-squares means (LS means) were used for the association mapping of this trait. Finally, variance components using the VARCOMP function in SAS were calculated for branching and DTF and were used to estimate broad-sense heritabilities (H2) as the total genotypic variance divided by the total phenotypic variance.
Association mapping of branching and DTF were performed in the software package TASSEL v. 3.0 [33] using all SNPs with MAF ≥10%. Three different association mapping models were run for each trait including a mixed linear model (MLM) accounting for kinship (i.e., familial relatedness; K-matrix) and two MLMs using kinship and population structure as estimated via either principle component analysis (PCA) (P-matrix) or the program STRUCTURE (Q-matrix) [28], [29]. Model effects for individual SNPs were output from TASSEL for each MLM [33].
Linear model testing was performed by plotting the observed P-values from the association test against an expected (cumulative) probability distribution. These quantile-quantile (q-q) plots indicate the extent to which the analysis produced more significant results than expected by chance. Models that follow the expected line more closely are assumed to have produced fewer false positives. Given that non-independence of linked makers in the dataset could lead to overly conservative significance thresholds [34], we used the multiple testing correction method of Gao et al. [35] to evaluate the significance of our results. This approach accounts for correlations amongst markers while controlling the type I error rate (alpha = 0.05). Using both real and simulated data, this correction has been shown to be an efficient and accurate method of minimizing false positives in the presence of inter-marker LD.
Where possible, to enable the identification of novel genetic effects, we also compared the genetic map positions of significant associations to those of previously mapped branching and flowering time QTL. This was done by projecting QTL onto the sunflower consensus map (which, as noted above, was also used for ordering the SNPs employed in the present study) based on shared markers. Note that some previous QTL results could not be included in this comparison due to a lack of shared markers and/or differences in linkage group nomenclature. The linkage group names were standardized by Tang et al. [36], though the new naming scheme was not immediately adopted by all researchers.
The total number of readily scorable, bi-allelic SNPs in the focal population was 5,788. The number of SNPs with a MAF ≥10% was 5,359. Expected heterozygosity, or Nei's unbiased gene diversity, averaged 0.404±0.005 (mean ± standard error), and ranged from 0.007 to 0.5. Observed heterozygosity averaged 0.034±0.0044, and ranged from 0 to 0.38. Gene diversity and observed heterozygosity for each line classification grouping are found in Table 1.
Our STRUCTURE results using the full set of SNPs with MAF ≥10% indicated that K = 3 (hereafter referred to as Q = 3 corresponding to the Q matrix for the association testing results below), providing support for the existence of three genetically distinct clusters in our association panel. STRUCTURE results are grouped and graphed according to the line classifications (see Methods) in Figure 1. DeltaK and the mean likelihood values are plotted in Figure S2. Clusters one and three largely consist of the maintainer (HA) lines whereas the majority of the restorer-oil (RHA-oil) lines exhibit substantial membership in cluster two. The PCoA analysis was largely consistent with the STRUCTURE results (Figure S3). In order to simplify the PCoA plot, we combined categories by grouping the lines into either HA, RHA-nonoil, RHA-oil, or other (this category contained all remaining lines/accessions). The RHA-oil lines are generally separated from the balance of the cultivated germplasm along the first and second axes, while the HA lines are generally distinct along the first axis.
Relative kinship was also estimated using the full set of markers (MAF ≥10%). Approximately 60% of the pairwise kinship estimates were near zero (i.e., less than 0.005), indicating the lines were essentially unrelated (Figure S4). The remaining estimates ranged from 0.05 to just less than 1, with a rapidly decreasing number of sunflower pairs exhibiting higher levels of relatedness.
The results of our genome-wide analysis of linkage disequilibrium (LD) are summarized in Figure 2 and Figure 3 and Figure S5. Figure 2 displays an LD matrix of the squared allele frequency correlations (r2) plotted for the ordered markers (MAF ≥10%); the x- and y-axes correspond to the 17 LGs in sunflower. Regions of the genome where LD extends for considerable genetic map distance are visible as yellow to red squares on the figure (e.g., on LGs 5 and 10). Figure S5 shows plots of the squared allele frequency correlations (r2) for 10,000 random pairs of SNPs within 50 cM as a function of genetic map distance between SNPs for the 17 LGs. Looking across chromosomes, different overall patterns of LD are apparent. For example, LG 10 exhibits a relatively slow overall decay in LD, largely due to strong haplotypic structure across a portion of the chromosome (see also Figure 2 and Figure 3), whereas LG 11 shows a much more rapid decay of LD. Following the methods of Macdonald et al. [31], the red line on each graph in Figure S5 summarizes the observed r2 values as a function of map distance using the ksmooth function in the statistical software R (http://www.R-project.org/). On a per chromosome basis, the average genetic distance at which r2 dropped below 0.1 ranged from 6.95 cM to 12.6 cM with the medians ranging from 3.93 cM to 10.1 cM. The sliding window analysis of r2 further illustrates the variability in LD across the genome (Figure 3). In some cases, entire chromosomes show very low levels of LD. In other cases, elevated LD is visible in specific chromosomal regions, including portions of LGs 1, 5, 8, 10, and 13. The sliding window analysis of genetic diversity likewise revealed variation in UHe across the genome (Figure S6).
We also estimated FST between the two primary breeding pools (i.e., RHA vs. HA lines) for the full set of markers (MAF ≥10%), plotted the results against genetic map position (Figure 4), and tested for selection using BayeScan. Genomic regions exhibiting significantly elevated differentiation are visible as spikes in FST (with individually significant markers being colored in red) on several chromosomes, including portions of LGs 8, 10, and 13, and a single marker on LG 14.
Our association mapping population exhibited substantial phenotypic diversity for both plant architecture and flowering time, expressed here in terms of branching and DTF. Significant positive correlations were found for both branching and flowering time across locations (i.e., more or less branched lines tended to behave similarly across locations, and the same was seen for DTF in terms of earlier vs. later flowering lines), whereas there was an overall significant negative correlation between branching and DTF (i.e., more highly branched plants tended to flower earlier; Figure S7). As noted above, there was a significant G×E interaction (P<0.01) for both traits studied. An important consequence of such G×E interactions is that different associations may be detected across environments; thus, we performed the association analyses separately for each location. The estimates of broad-sense heritability for branching and DTF were 0.861 and 0.124, respectively. The number of plants that showed a complete lack of branching was 79, 110, and 70 in Georgia, Iowa, and British Columbia respectively. The number of branches per genotype averaged 7.3, 7.2, and 7.6 and ranged from 1–29, 1–19, and 1–30 in Georgia, Iowa, and British Columbia respectively. DTF averaged 57.1 (range 41–77), 68.7 (45–95), and 80.4 (63–104) in Georgia, Iowa, and British Columbia respectively.
In terms of branching, our analyses revealed significant associations on LGs 2, 4, 5, 6, 7, 8, 9, 10, 12, 13, 14, and 17 (Figure 5; Table 2; Table S2). In general terms, each of these LGs had a single peak of significant marker-trait associations. The exceptions were: LG 7, which had two peaks in IA (one near 10 cM and another near 53 cM); LG 10, which had a primary peak centered near 25 cM in all locations and a secondary peak near 74 cM in IA and BC; LG 13, which had a primary peak near 41 cM and a secondary peak near 64 cM in all locations, as well as a third peak near 5 cM in BC; and LG 17, which had a primary peak near 41 cM in GA and IA and a secondary peak near 21 cM in IA. For DTF, our analyses revealed significant associations on LGs 1, 3, 4, 9, 10, 12, 13, and 17 (Figure 6; Table 3; Table S2). Here again, each of these LGs typically had a single peak of significant associations. The exception was LG 13, which had a primary peak near 70 cM in GA, a secondary peak near 3 cM in GA, and a third peak near 21 cM in IA. Table 2 and Table 3 summarize these results across locations for the three mixed models. Figure 5 and Figure 6 (upper panels) show the Manhattan plots for branching and DTF, respectively, in each of the three locations. For each location, the three mixed models are plotted as kinship (K, red), population structure as measured by PCA plus kinship (P+K, blue), and population structure as measured by STRUCTURE plus kinship (Q+K, dark grey). Points above the dashed threshold line are significant after correcting for multiple tests, as detailed above. Figure 5 and Figure 6 (lower panels) also show the quantile-quantile (q-q) plots of the observed P-values versus the expected for each of the three models as well as a naive model that does not account for population structure or kinship. As can be seen from the q-q plots, the distribution of observed P-values in the naive model greatly deviated from the expected distribution whereas the other models followed the expected distribution much more closely. This result reflects the potential confounding effects of population structure and relatedness in the dataset. Note, however, that for DTF in BC, two of the models (P+K and Q+K) provided fewer significant results than expected by chance, suggesting that these models may be overly conservative (e.g., [37]). The full set of results, including functional annotations for the genes from which the SNPs were derived (see [20]), significance values, and SNP effects for all individual loci at each location and using all three models, are provided in Table S2.
The graphical genotypes for all 17 LGs across the full population of 271 accessions are presented in Figure S8. The genotype with the highest fraction of shared haplotypes across the genomes of the 271 lines (G1 in red) corresponds to the accession HA89, a line of great historical importance in sunflower breeding. HA89 accounted for an average of 16.2% of the genomes of the 271 lines. Overall, the 25 most common genotypes accounted for an average of 63.5% of the genomes of the 271 lines examined. For ease of interpretation, only the top nine genotypes are color-coded (beyond this point, each additional genotype individually accounted for ca. 1% or less of the genome); in total, these nine genotypes accounted for an overall average of 50.7% of the genomes of the 271 lines (see Table S3). White regions either correspond to non-major genotypes or reflect stretches with fewer than 25 consecutive, homozygous SNPs. Figure 7 depicts the results for LG 10 with the data sorted by the average number of branches produced by plants of each accession at all three locations (see heat map along the top). As noted above, the upper portion of this LG exhibits strong haplotypic structure and elevated LD across an extended region along with a major effect on branching at all three locations. See below for a detailed discussion of the historical and biological significance of these results.
The work presented herein represents the largest and most comprehensive analysis of population genomic diversity in cultivated sunflower to date. In terms of overall levels of SNP diversity, our data indicate that sunflower exhibits considerable molecular variation, on par with estimates derived from large-scale SNP surveys of other crops (e.g., maize [38]–[40], barley [41], and rice [42]). We also documented substantial phenotypic variation in terms of both plant architecture and flowering time, ranging from a complete lack of branching to whole plant branching and including accessions that reached reproductive maturity over a period spanning greater than 30 days. Thus, despite the population genetic bottlenecks that are known to have occurred during domestication and improvement, the cultivated sunflower gene pool harbors substantial variability.
In terms of overall population structure, our results are largely in agreement with our prior analyses based on a much smaller set of simple-sequence repeat markers (SSRs) [22]. This general agreement between our current findings and the earlier, SSR-based work suggests that any possible ascertainment bias during SNP discovery and selection had minimal effects on our population genetic results. In fact, the preferential usage of SNPs with high heterozygosity would be expected to result in an underestimate of the magnitude of structure [43] but the FST estimates between the B and R lines is virtually identical between the SNPs (FST = 0.049) and the SSRs (FST = 0.047). The much larger number of markers in the present study has, however, allowed us to refine our earlier findings. Notably, we found evidence for the presence of three genetically distinct groups within the germplasm collection. One of these groups was primarily composed of the RHA lines, while a second group consisted of a large number of HA lines, and the third included a more diverse assemblage of lines. The PCoA likewise demonstrated a split between the RHA and HA lines, with an even clearer division between the RHA-oil and HA lines. This genetic distinction between B and R lines is expected given the breeding history of sunflower, which involves the maintenance of distinct gene pools to maximize heterosis in hybrid crosses [44], [45]. Taken together, these results underscore the need to account for population structure when performing association analyses in sunflower.
Our analysis of LD revealed considerable variability across the genome. In most regions, LD declined quite rapidly as a function of genetic distance and the correlation between most pairs of SNPs fell to negligible levels (i.e., r2≤0.10) within 3 cM. In some instances, however, LD remained elevated (on average) over greater distances (e.g., LG 10; see Figure S5). Inspection of Figure 2 and Figure 3 reveals the existence of a number of localized islands of LD, including blocks on LGs 1, 5, 8, 10, 13, 16, and 17. In looking more closely at these localized regions exhibiting elevated LD, it is apparent that many of them occur in close proximity to genes or QTL underlying traits that have been targeted by selection during sunflower domestication and/or improvement. Given the history of breeding for resistance to diseases in sunflower [46] it is noteworthy that the four of these spikes in LD (on LGs 1, 5, 8, and 13) co-localize with QTL and/or candidate genes for resistance to several important diseases. Note that co-localization of QTL and LD spikes/associations (below) is based upon concordance of shared genetic markers (most often microsatellites) between the previous QTL map(s) and the map of Bowers et al. [21]. More specifically, QTL and/or genes for resistance to downy mildew (Plasmopara halstedii; [47]–[52]) co-localize with spikes in LD on LGs 1 and 8. Similarly, the block of elevated LD observed on LG 5 co-localizes with a QTL for resistance to black stem (Phoma macdonaldii, [53]). Finally, the spike in LD on LG 13 co-localizes with sunflower rust resistance genes [50], [52], [54]. It thus appears that selection for disease resistance may have played a role in shaping genome-wide patterns of LD in sunflower, though we cannot rule out the possibility of selection on other traits (see below).
The important role that selection on plant architecture played during the domestication and subsequent improvement of sunflower also appears to have shaped patterns of genetic diversity across the sunflower genome, especially with respect to LG 10. During the initial domestication of sunflower, unbranched, monocephalic landraces were preferentially propagated and the domesticated lineage moved away from the intensely branched, multi-headed phenotype that is characteristic of wild sunflower [17], [55]–[57]. Until the mid-20th century, modern cultivars were thus typically unbranched; however, beginning in the late 1960s, the transition to hybrid breeding and the associated desire for a prolonged flowering period in male lines resulted in the re-introduction of branching into the sunflower gene pool [45]. This resulted in selection favoring the fixation of a recessive branching allele at the so-called B-locus in R lines. The B-locus has since been mapped to approximately 27 cM from the top of LG 10 [58].
In viewing the graphical genotypes, the B-locus is visible as differentiated haplotypic blocks that span this region on LG 10 and which clearly correlate with the extent of branching (Figure 7). Interestingly, the re-introduced branching haplotype (in dark blue) spans ca. 25 cM, whereas the unbranched haplotype (primarily in red) appears to span ca. 10 cM. Thus, the effects of the very recent re-introduction of branching to the cultivated gene pool can be visualized as a large haplotypic block presumably resulting from a recent, strong selective sweep in the branched R (RHA) lines. This pattern can also be seen in the sliding window analyses of LD (r2) and the plots of population differentiation (FST) vs. genetic map position. In both cases, spikes are clearly visible in that same region along LG 10 (Figure 3 and Figure 4). Notably, our BayeScan analysis indicated that this region of the genome, along with portions of LG 8 and 13 and a single marker on LG 14, exhibits significantly elevated FST (Figure 4). This finding is consistent with the notion that this differentiation – which spans a remarkably long (at least in genetic terms) and otherwise recombinogenic genomic interval – was driven by strong selection.
Not surprisingly, this same region of LG 10 harbored highly significant associations for branching in all three locations, as well as for DTF in GA. Overall, we found significant associations for branching in 17 genomic regions on 12 of the 17 LGs in sunflower. In five cases, these associations overlapped with previously identified QTL for branching in sunflower on LGs 10 (a), 12, 13 (a and b) and 17 (Figure S9) [9], [59]–[61]; the remainder were novel effects for number of branches that have not previously been documented. Similarly, we found significant associations for DTF in 10 genomic regions located on 8 of the 17 sunflower LGs (Figure S10) [9], [62]; the remainder (on LGs 1, 3, 4, 10, 12, 13) were novel effects. It is noteworthy that the detected associations generally spanned much smaller intervals than are typical of traditional QTL studies. This result is consistent with our finding that LD decays relatively rapidly across much of the genome (see above) and suggests that even higher marker densities would be desirable for developing a complete picture of trait variation in sunflower.
In terms of the relationship between marker-trait associations and both LD and population differentiation, the branching and DTF associations co-localized with spikes in r2 and FST (on LGs 8, 10, and 13; Figure 3 and Figure 4). As noted above, all three of these regions, along with a single marker on LG 14, exhibited significantly elevated FST values, suggestive of selective divergence. Given the historical importance of the B-locus and the clear correspondence of the observed haplotypes to sunflower breeding groups (and thus branching architecture; see above), it seems likely that the driving force behind the pattern observed on LG 10 was selection on branching. The situation on LGs 8 and 13 is less clear; the observed patterns may have been driven be selection on branching, flowering time, disease resistance, or some combination thereof. It must be kept in mind, however, that the branching associations that we detected on LG 8 were most apparent in the kinship-only model, and disappeared almost entirely when we controlled for population structure. As such, this result in particular may have been a byproduct of population structure as opposed to a true functional association.
It is particularly noteworthy that the majority of significant associations identified herein are located in the aforementioned islands of LD. Given that our analyses relied on a single SNP in each of ca. 5,300 genes, we are almost certainly missing out on associations in regions of low LD. In fact, nearly 50% of the sliding windows analyzed had an average r2<0.1 and over 85% had an average r2<0.2. As such, future analyses aimed at assaying genetic variation at a much higher density (e.g., using genotyping-by-sequencing [63] or even whole genome re-sequencing) seem warranted and are likely to facilitate a much more detailed characterization of the molecular basis of phenotypic variation in sunflower.
|
10.1371/journal.pgen.0030235 | Deletion of the MBII-85 snoRNA Gene Cluster in Mice Results in Postnatal Growth Retardation | Prader-Willi syndrome (PWS [MIM 176270]) is a neurogenetic disorder characterized by decreased fetal activity, muscular hypotonia, failure to thrive, short stature, obesity, mental retardation, and hypogonadotropic hypogonadism. It is caused by the loss of function of one or more imprinted, paternally expressed genes on the proximal long arm of chromosome 15. Several potential PWS mouse models involving the orthologous region on chromosome 7C exist. Based on the analysis of deletions in the mouse and gene expression in PWS patients with chromosomal translocations, a critical region (PWScr) for neonatal lethality, failure to thrive, and growth retardation was narrowed to the locus containing a cluster of neuronally expressed MBII-85 small nucleolar RNA (snoRNA) genes. Here, we report the deletion of PWScr. Mice carrying the maternally inherited allele (PWScrm−/p+) are indistinguishable from wild-type littermates. All those with the paternally inherited allele (PWScrm+/p−) consistently display postnatal growth retardation, with about 15% postnatal lethality in C57BL/6, but not FVB/N crosses. This is the first example in a multicellular organism of genetic deletion of a C/D box snoRNA gene resulting in a pronounced phenotype.
| Prader-Willi syndrome, or PWS, is a complex neurogenetic disorder and the most common genetic cause of life-threatening childhood obesity. Newborns have poor muscle tone, making suckling difficult, which leads to poor weight gain. After infancy, they experience extreme hunger, leading to obesity. Other symptoms include short stature, mental retardation, and often infertility. In PWS patients, a complex set of genes on the paternal chromosome 15 (in the PWS region) is missing or unexpressed. In an attempt to understand this disorder, various protein-coding genes in this region have been deleted in mice, but none of the resulting phenotypes consistently correlated with the human disease. This region also contains a cluster of genes that encode functional non-protein-coding RNAs. We deleted specifically the MBII-85 small nucleolar RNA (snoRNA) gene cluster on the parental mouse chromosome, which did not affect expression of any of the other snoRNA or protein-coding genes in the PWS region. These mice consistently displayed postnatal growth retardation starting from day 5 to 6, low postnatal lethality only in certain genetic backgrounds (<15%), and no adolescent obesity. Thus, this mouse model, with the deletion of a small, brain-specific non-protein-coding RNA, should prove useful for teasing out the various molecular pathologies of PWS.
| The human genetic locus 15q11-q13 is subject to genomic imprinting that is controlled by a bipartite imprinting center (IC) [1]. Imprinting defects or chromosomal rearrangements/deletions within this locus (Figure 1) are responsible for the development of two clinical disorders—Angelman (AS) and Prader-Willi (PWS) syndromes. AS (MIM 105830) is a complex neurogenetic disorder characterized by mental retardation, severe limitations in speech and language and abnormal behavior, and results from loss of maternal expression of the Ube3A gene [2]. PWS (MIM 176270) is a complex neurogenetic disorder with a population prevalence of 1 in 10 000 to 50 000 [3–5] that is characterized by decreased fetal activity, muscular hypotonia, failure to thrive, short stature, obesity, mental retardation, and hypogonadotropic hypogonadism, for review, see [2,6]. PWS results from lack of paternal expression of one or several imprinted genes within the PWS/AS locus. Several paternally expressed protein-coding genes map to this locus, including NECDIN (NDN), MAGEL2, MKRN3, and the bi-cistronic SNURF-SNRPN (Figure 1B). There are also numerous paternally expressed C/D box snoRNA genes located downstream from the SNURF-SNRPN gene. Most of them are organized into two main clusters of HBII-85 and HBII-52 snoRNAs, containing 29 and 47 copies, respectively. Other snoRNAs are present as single (HBII-436, HBII-13 and HBII-437) or double copy (HBII-438a/438b) genes (Figure 1B). Most, if not all, snoRNAs are processed from a long, non-protein-coding RNA (npcRNA) transcript designated U-UBE3A-ATS in human and Lncat (large paternal non-protein-coding RNA, encompassing Snurf-Snrpn and Ipw exons together with the Ube3a antisense transcript) in mouse [7–9]. U-UBE3A-ATS extends ∼450 kb from the untranslated U exons upstream of the small nuclear ribonucleoprotein N (SNURF/SNRPN) gene to the UBE3A gene (Figure 1B).
In mouse the syntenic PWS/AS locus is located on chromosome 7C and contains all the aforementioned protein coding and non-protein-coding gene orthologs, except for the presence of protein-coding gene Frat 3 and absence of HBII-437 and HBII-438a/438b snoRNA genes (Figure 1C). Several mouse models for PWS have already been generated (Figure 1D). They can be divided into 3 groups: 1) transgenic mouse lines with disruptions of the PWS/AS locus; 2) mice with targeted elimination of the imprinting center (IC) controlling transcription of PWS genes, or targeted elimination of individual, single genes from the PWS locus; and 3) mice with uniparental paternal disomy (UPD) [10,11].
In all existing PWS mouse models (Figure 1D) that involve large deletions comprising several paternally expressed, imprinted genes, severe phenotypes (failure to thrive and early postnatal lethality) were observed [10,12–15]. Targeting of the single Mkrn3 and Snrpn genes, or some of the Ipw exons together with the MBII-52 snoRNA genes cluster [13,15–17] or deletion of an analogous region in human [18] all produced no PWS-like phenotypes. Elimination of the Magel 2 gene caused altered behavioral rhythmicity (Figure 1D8) [19]. However this gene is unlikely to be a main “PWS player” [20]. Currently, only elimination of the single Necdin gene leads to the development of early postnatal lethality and neurological abnormalities resembling the PWS, although the phenotypic effect depends on the targeted region and genetic background of the mice, some of which had no apparent phenotype [21–23]. The PWS critical region (PWScr) was narrowed to the locus containing the MBII-85 small nucleolar RNA (snoRNA) gene cluster based on existing mouse models [16] and gene expression analysis in PWS patients with chromosomal translocations [18,24,25]. These studies indicate that SNURF/SNRPN, MKRN3, NECDIN and MAGEL2 genes are unlikely to play a primary role in the pathogenesis of PWS. However, the question of their possible functional contribution to more severe phenotypic expression seen in typical PWS patients remains open [16,18,24]. We have applied the “chromosome engineering” technique [26] to delete the PWScr in mice. When the deleted allele is inherited maternally (PWScrm−/p+), no phenotypic abnormalities are visible. When it is inherited paternally (PWScrm+/p−), we consistently observe postnatal growth retardation in mice and less than 15 percent postnatal lethality in 129SV x C57BL/6 genetic crosses.
To test the hypothesis that the PWScr is the most probable candidate region for neonatal lethality, failure to thrive and postnatal growth retardation, we devised the following strategy for producing PWScrm+/p− mice (Figure 1E and 1F). Hypoxanthine-guanine phosphoribosyltransferase (HPRT)-deficient ES cells, AB2.2, were modified through homologous recombination (HR) using targeting constructs 5′HPRT/PWScr_targ and 3′HPRT/PWScr_targ to place loxP sites proximal to the 5′ flanking region of the MBII-85 snoRNA gene cluster and distal to the Ipw exon C, respectively (Figure 1E). Deletion of the PWScr harboring the entire cluster of MBII-85 snoRNA genes together with Ipw exons A-C was accomplished by expressing CRE recombinase in ES-targeted cells (Figures 1F and 2A) and injecting these into blastocysts. We identified two chimaeras derived from one of the PWScr-deleted ES clones with successful germ-line transmission. Deletion of the PWScr allele was confirmed several ways: 1) Resistance of ES cells to HAT media requires a functional HPRT gene. Restoring the intact HPRT gene through deletion of the PWScr leads to resistance of ES cells to HAT media. 2) Southern blot analysis of all HAT-resistant ES colonies (data not shown), as well as all PWScr-deleted mice, using 5′HR and 3′HR probes (Figure 2A and 2B), revealed identical bands corresponding to the correctly deleted PWScr allele (19998 bp). 3) We PCR amplified and sequenced flanking regions of the PWScr locus together with the inserted HPRT gene (Figures 1G, 2C, and S1) using PCR primers MB85seqD1 and MB85seqR1 (Table S1), and confirmed the deleted PWScr allele as well.
PWScrm+/p− pups born from chimaeras were significantly smaller than their wild-type siblings, such that on postnatal day 10, the PWScrm+/p− individuals (Figure 3) could be reliably predicted prior to DNA analysis. We monitored the weights of all mice over several weeks after continuous breeding and found that only the PWScrm+/p− and PWScrm−/p− mice displayed postnatal growth retardation compared to the PWScrm+/p+ siblings. Statistical analyses were performed on mice separated by genotype, genetic background and gender (Figures 4A–4D and S2), as well as separately for litters (data provided upon request). Postnatal growth retardation in PWScrm+/p− mice was observed, thus far, over six generations, independent of genetic background [e.g., 129SV x C57BL/6 (>85% C57BL/6 genetic background contribution, Figure 4A and 4B), 129SV x C57BL/6 x FVB/N (∼50% FVB/N contribution, Figure S2A and S2B), and 129SV x C57BL/6 x BALB/c (∼50% BALB/c contribution, Figure S2C and S2D)]. Differences in growth dynamics between PWScrm+/p+ and PWScrm−/p− or PWScrm+/p− mice continued to be statistically significant into adulthood (up to 1 year in BALB/c crosses) (Figure 4A, 4B, and S2). Moreover, not a single case of obesity was detected. Interestingly, when growth dynamics for mice were analyzed by gender, deficiencies in the PWScrm+/p− or PWScrm−/p− female mice tended to be less pronounced than those in the male deletion mice (Figure 4A, 4B, S2C, and S2D). This observation is well correlated with a statistical analysis of PWS patients indicating that the degree of short stature is more prominent in males than in females, irrespective of ethnic groups (genetic background) [27]. Furthermore, mice with the maternally transmitted PWScr-deleted allele were indistinguishable from their wild type littermates (Figure 4C). In addition, males (5′HPRT) derived from 5′HPRT/PWScr_targ-targeted ES cells were crossed with C57BL/6 females. As expected, we did not observe growth retardation in the 5′HPRTm−/p+ mice (p=0.275) (Figure 4D).
Notably, the observed postnatal growth retardation phenotype became apparent during the first week of life starting from postnatal day 5 in males and 6 in females (Figure 4A and 4B; Table S2), while there were no growth differences at early postnatal ages P1-P4 (Figure 4A and 4B; Table S2). In agreement with the absence of early postnatal growth retardation, no weight differences were observed during embryonal development at E12.5, E15.5 and E18.5, or in late gestation (E15.5 and E18.5) placentas (Figure 4E, and data not shown). Hence, one possible explanation for our observations might be poor breast-feeding in PWScrm+/p− pups as is the case with postnatal PWS patients. Insufficient milk intake resulting in growth retardation is consistent with the Holland hypothesis and observations in TgPWS mice (transgenic deletion PWS mouse model; Figure 1D1) suggesting, that the main basis of the PWS syndrome is not obesity and uncontrollable craving but early postnatal starvation [28,29].
The PWScrm+/p− mice exhibit postnatal growth retardation, but contrary to early predictions [16] postnatal lethality was observed in only 15 percent of the cases in 129SV x C57BL/6 genetic crosses (Figure 4F). On the other hand, we did not observe postnatal lethality in the FVB/N crosses. PWScrm+/p− mice were dying from postnatal days 1 to 22. The surviving PWScrm+/p− and PWScrm−/p− mice are alive and apparently well for at least 1 year. In future experiments, it would be interesting to test the possible role of MBII-13 and MBII-436 snoRNAs and/or the genomic region between the Snrpn and MBII-85 genes for their contributions to more severe phenotypes with higher rates of postnatal lethality. In addition, our PWScrm+/p− mouse model is in good agreement with two recently identified patients with a balanced chromosomal translocation involving SNRPN [24,30]. Both patients lack HBII-85 expression and exhibit a mild PWS phenotype, (e.g., they did not require gavage feeding, but were growth retarded.)
PWS is also characterized by hypogonadotropic hypogonadism and infertility in patients [6]. We studied the fertility of the PWScrm+/p− and PWScrm−/p+ mice and observed that PWScrm+/p− males and females transmitted the PWScr deleted allele to offspring. To further extend our observations, we established 10 breeding pairs from each of the PWScrm+/p− and PWScrm−/p+ males with 80 wild-type females. All matings resulted in pregnancies leading to successful live births with litter sizes around eight (7.93 ± 0.46 for PWScrm+/p− and 8.33 ± 0.48 for PWScrm−/p+ males), and always included wild-type mice and mice carrying the PWScr-deleted allele in a ratio of 1:1 (4.17 ± 0.34 : 3.66 ± 0.38 for PWScrm+/p− and 4.00 ± 0.54 : 4.17 ± 0.53 for PWScrm−/p+; Table S3). Thus, male mice containing the PWScr-deleted allele inherited maternally or paternally, are transmitting this allele in a Mendelian fashion.
The PWS locus includes several paternally expressed, protein coding genes, including Necdin, Magel2, Mkrn3, Frat3, and the bi-cistronic Snurf-Snrpn. To examine whether the deletion of the paternal PWScr from mouse chromosome 7C perturbed the expression of the aforementioned imprinted genes, we analyzed their expression levels in our mice by RT-PCR and real-time PCR. We failed to detect any significant differences in the expression levels of the investigated genes in PWScrm+/p− mice compared to control littermates PWScrm+/p+ (Figure 5A; Table 1). In addition, the controversial data concerning the involvement of Necdin in the PWS phenotype [21–23] prompted us to also examine its expression in more detail by Northern blot hybridization (Figure 5B). Consistent with the RT-PCR and real-time PCR results, there were no differences in the levels of Necdin mRNA in brains of the PWScrm+/p−, PWScrm−/p+ or control PWScrm+/p+, 5′HPRTm+/p+ and 3′HPRTm+/p+ mice (Figure 5B; Table 1 and data not shown).
We also examined the levels of the maternally expressed, imprinted Ube3a and Atp10a protein coding genes, located at the end of the PWS locus and involved in the development of AS [2] (Figure 1B and 1C). The RT-PCR (Figure 5A), real-time PCR (Table 1), and Northern blot hybridization (Figure 5C and 5D) analyses revealed similar expression levels of both genes in brains of PWScrm+/p− and PWScrm−/p+, as well as in control mice.
The mouse PWS locus on chromosome 7C also contains numerous neuronal, paternally expressed C/D box snoRNA genes, including two single copies of MBII-436 and MBII-13, and two multiple copy clusters, MBII-85 and MBII-52 [7,31]. Vertebrate snoRNAs are embedded in introns of protein coding genes, which are posttranscriptionally processed to yield mature mRNA and snoRNA. Occasionally the spliced exons are devoid of open reading frames, as if the sole function of the transcript is the expression of the snoRNA [32]. This is apparently the case for most, if not all, snoRNAs from the PWS locus that are co-transcribed with the large paternally expressed polycistronic Lncat npcRNA [7–9]. The Lncat transcript is complex and generates, by alternative splicing and other processing events, numerous RNA products (e.g., those represented by expressed sequenced tags (ESTs) and mature snoRNAs). The corresponding Ipw exons are a subset of Lncat-derived ESTs and map to the MBII-85 and MBII-52 snoRNA clusters that are interspersed with repeated exons A1/A2 and G1/G2, respectively [31,33]. The Ipw exons B, C, H, E, F map between both clusters. Northern blot analyses of RNA samples extracted from brains of PWScrm+/p− mice revealed the complete absence of MBII-85 snoRNA while expression of all other snoRNA genes in the PWS locus were unaffected (Figure 6A).
We then analyzed the expression of the Ipw exons using Northern blot hybridization with a cDNA probe containing the F and G exons, and found it to be slightly decreased in PWScrm+/p− mice (Figure 6B). However, expression of MBII-52 snoRNA was not altered, presumably because the primary Lncat transcript can undergo different processing pathways to yield mature MBII-52 snoRNA (Figure 7). Recently it was reported that lack of HBII-52 snoRNA genes together with most of the IPW exons did not result in a PWS phenotype in either human individuals [18] or in a mouse model [15,16]. Therefore, it is less likely that deletion of alternatively spliced Ipw exons A1/A2, B and C are responsible for the phenotype we obtained here, although we cannot completely exclude the possibility that lack of those exons in a long npcRNA can contribute to it. In future experiments, we will address the question, whether expression of MBII-85 snoRNA in a different host gene is sufficient to compensate the observed phenotype.
Lack of MBII-85 snoRNA expression in PWScrm+/p− compared to PWScrm−/p+ mice, along with the unaltered expression of paternally and maternally expressed genes from the PWS/AS locus in both heterozygous mice, indicates that deletion of PWScr does not affect imprinting of neighboring genes. Moreover, patients with translocations leading to a similar deletion encompassing the genes encoding HBII-85 and HBII-438A snoRNA exhibit normal biparental methylation patterns [24,30]. Thus, absence of MBII-85 snoRNA is the most likely cause for the phenotype observed in PWScrm+/p− and PWScrm−/p− mice (Figure 6A).
In Eukarya, most C/D box npcRNAs guide site-specific 2′-O-methylation in rRNAs and small nuclear RNAs (snRNAs) by complementarity to defined sites within these RNA targets [34]. However, most if not all C/D box small npcRNAs that map to the PWS locus, including MBII-85 and MBII-52 snoRNAs, lack significant complementarities to any rRNA or snRNA targets [7,31]. Although a major role for MBII-52 in the etiology of the PWS can be excluded, cell culture experiments have suggested that MBII-52 snoRNA might play a role in A to I editing and/or alternative splicing of the 5HT-2c serotonin receptor pre-mRNA [35,36]. In spite of these discrepancies, one might still be tempted to propose that MBII-85 snoRNA interacts (by complementarity) with a yet unidentified RNA target. However, other less orthodox functions for MBII-85 snoRNA and its role in postnatal growth retardation must also be seriously entertained. In any event, MBII-85 is the first example of a C/D box snoRNA gene, whose deletion results in an obvious phenotypic change in a multicellular organism. Future experiments might reveal additional, less obvious defects and deficiencies in PWScrm+/p− mice. Our mouse model will serve as an important tool for further investigations of the molecular pathogenesis of PWS in man.
For construction of the 5′HPRT/PWScr_targ and 3′HPRT/PWScr_targ targeting vectors, we isolated clones RPCIP711K19517Q6 and RPCIP711J18414Q6, respectively, from the RPCI21 mouse PAC library (RZPD German Resource Center for Genome Investigation) using the MBII-85 oligonucleotide (Table S1). The 5′HPRT/PWScr_targ construct was generated using the 5′ flanking region of the MBII-85 gene cluster as a template for PCR-amplification of 1075 bp and 7387 bp DNA fragments with primer pairs 5′FLAdir/5′FLArev and 5′FLBdir/5′FLBrev, respectively. The PCR fragments were used as homologous arms in the targeting vector. The cassette containing the 5′ portion of the HPRT gene, the loxP site, and the neomycin resistance gene was subcloned from a λ phage vector kindly provided by A. Bradley (Baylor College of Medicine, Houston, USA). The thymidine kinase (TK) gene was placed outside of the homologous arm (Figure 1E).
The 3′HPRT/PWScr_targ construct was cloned in a similar way. The 3′ flanking region of the MBII-85 gene cluster was used as a template to PCR-amplify 1098 bp and 6720 bp DNA fragments using primer pairs 3′FLBdirN/3′FLBrev and 3′FLAdir/3′FLArev, respectively. The insertion cassette, containing the 3′ portion of the HPRT gene, the loxP site, and the puromycin resistance gene was subcloned from the λ phage vector kindly provided by A. Bradley. The TK gene was placed outside of the homologous arm.
The 5′ HR probe for Southern blot hybridization was generated by using the PAC DNA containing the 5′ region of the MBII-85 gene cluster in two PCR reactions with oligonucleotide pairs 5′PRAdir/5′PRArev and 5′PRBdir/5′PRBrev, and ligating and cloning the resulting PCR products into the pSL300 vector [37]. The 3′ HR probe was cloned in a similar fashion. PCR products were obtained with oligonucleotide pairs 3′PRAdir/3′PRArev and 3′PRBdir/3′PRBrev (Table S1).
HPRT-deficient embryonic stem cells AB2.2 (from A. Bradley), passage 17, were expanded in HEPES-buffered, Dulbecco's modified Eagle's medium supplemented with 15% fetal bovine serum (HyClone), nonessential amino acids, L-glutamine, β-mercaptoethanol, 1000 U/ml recombinant LIF (Chemicon) and antibiotics (penicillin 100 U/ml and streptomycin 100 μg/ml) on a γ-irradiated monolayer of SNL6.7 cells (from A. Bradley) or mouse primary fibroblasts. For electroporation, 2 × 107 ES cells were resuspended in 20 mM HEPES pH 7.4, 173 mM NaCl, 5 mM KCl, 0.7 mM Na2HPO4, 6 mM dextrose, and 0.1 mM β-mercaptoethanol [38]. The NotI linearized replacement targeting vectors 5′HPRT/PWScr_targ, 3′HPRT/PWScr_targ, and intact CRE expressing cassette pOG231 (55 μg DNA of each) were electroporated at 25 μF and 400V (Gene Pulser; Bio-Rad). After electroporation, cells were plated onto 100 mm culture dishes containing a γ-irradiated monolayer of primary, G-418-resistant, or SNL6.7 (G-418, puromycin- and HAT-resistant) fibroblast feeder cells. Thirty-eight hours later, 350 μg/ml G418 (Invitrogen) and 0.2 μM 2′-deoxy-2′-fluoro-β-D-arabinofuranosyl-5-iodouracil (FIAU) (Moravek Biochemicals and Radiochemicals, USA) or 1.0 - 0.5 μg/ml puromycin (Sigma) and 0.2 μM FIAU, or HAT (Sigma) were added to the culture medium. The medium was replaced every day and colonies were picked and analyzed eight days after plating. 5′HPRT loxP-targeted ES cells were analyzed using oligonucleotide pairs 85–5′screen2d/85–5′screen2r and 85–5′screen3d/85–5′screen3r for a nested PCR approach. For analysis of 3′HPRT loxP-targeted cells we employed the 85–3′screen1d/85–3′screen1r and 85–3′screen2d/85–3′screen2r nested PCR primer pair combinations. DNA blot analysis was performed as described [39]. Membranes were hybridized with 32p−labeled 5′HR and 3′HR probes (Figure 2A). Several independent ES clones containing the CRE-mediated PWScr-deletion were injected into 3.5-day-old B6D2F1 (C57BL/6 x DBA) blastocysts, and the resulting embryos were transferred to CD-1 foster mice. Chimeras were identified by their agouti coat color.
The 5.7 kb DNA fragment containing flanking regions of PWScr together with the inserted HPRT gene was amplified and sequenced using PCR primers MB85seqD1 and MB85seqR1. Sequencing reactions were performed using the BigDye terminator cycle sequencing reaction kit (PE Applied Biosystems) and resolved on an ABI Prism 3100 (Perkin Elmer) capillary sequencing machine.
To genotype mice we performed PCR analysis of genomic DNA from tail biopsies using the primer pair MB85deld1/MB85delr1 (Figure 2A and 2C; Table S1). PCR cycling was done at 93 °C – 2 min; 7 cycles (93 °C – 40 sec, 70 oC – 20 sec, touch down −1 °C, extension 67 oC – 1 min 40 sec); 35 cycles (93 °C – 40 sec, 55 oC – 20 sec, 67 oC – 1 min 40 sec). The final extension was performed at 67 °C for 5 min.
Total RNA was isolated from mouse brains using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. RNA samples, 20 μg each, were treated with RNase-free DNase I (Roche). First strand cDNA synthesis was performed using Transcriptor reverse transcriptase (Roche) and hexamer oligonucleotides, followed by PCR amplification with gene specific oligonucleotides (Table S1). The cDNA probes for necdin, Ube3A, Atp10a mRNAs, and Ipw exons were PCR amplified, cloned in the pCRII vector (Invitrogen), and subsequently sequenced using gene specific oligonucleotides (Table S1). Approximately 20 μg of total RNA was denatured, fractionated on 1.2% agarose formaldehyde gels, and transferred to GeneScreen nylon membranes (NEN DuPont). Hybridization was performed with 32P-labeled cDNA probes. Northern blot analysis of snoRNAs was performed with specific oligonucleotides (Table S1) as described [31].
A PWScr-deficient mouse line was established by breeding male chimeras nos. 2 and 5 from one mutant ES cell line with female C57BL/6 mice to produce heterozygous mice. Subsequently, heterozygous mice were interbred or bred to C57BL/6 mice. All breeding occurred at the ZMBE animal facility of the University Clinics, Münster in a controlled (21 °C, 30–50% humidity) room with a 12:12 hour light-dark cycle, and mice were housed under non-enriched, standard conditions in individually ventilated (36 (l) x 20 (w) x 20 (h) cm) cages for up to five littermates. Pups were weaned 19 – 23 days after birth and females were kept separately from males.
Statistical analysis was performed using the StatView software package. Body weight was analyzed using Student's t-test, or ANOVA for each day of postnatal age. Weights of placentas and embryos were analyzed using Mann-Whitney nonparametric statistics. Postnatal lethality was analyzed with the chi-square test. |
10.1371/journal.ppat.1003899 | Electron Tomography of HIV-1 Infection in Gut-Associated Lymphoid Tissue | Critical aspects of HIV-1 infection occur in mucosal tissues, particularly in the gut, which contains large numbers of HIV-1 target cells that are depleted early in infection. We used electron tomography (ET) to image HIV-1 in gut-associated lymphoid tissue (GALT) of HIV-1–infected humanized mice, the first three-dimensional ultrastructural examination of HIV-1 infection in vivo. Human immune cells were successfully engrafted in the mice, and following infection with HIV-1, human T cells were reduced in GALT. Virions were found by ET at all stages of egress, including budding immature virions and free mature and immature viruses. Immuno-electron microscopy verified the virions were HIV-1 and showed CD4 sequestration in the endoplasmic reticulum of infected cells. Observation of HIV-1 in infected GALT tissue revealed that most HIV-1–infected cells, identified by immunolabeling and/or the presence of budding virions, were localized to intestinal crypts with pools of free virions concentrated in spaces between cells. Fewer infected cells were found in mucosal regions and the lamina propria. The preservation quality of reconstructed tissue volumes allowed details of budding virions, including structures interpreted as host-encoded scission machinery, to be resolved. Although HIV-1 virions released from infected cultured cells have been described as exclusively mature, we found pools of both immature and mature free virions within infected tissue. The pools could be classified as containing either mostly mature or mostly immature particles, and analyses of their proximities to the cell of origin supported a model of semi-synchronous waves of virion release. In addition to HIV-1 transmission by pools of free virus, we found evidence of transmission via virological synapses. Three-dimensional EM imaging of an active infection within tissue revealed important differences between cultured cell and tissue infection models and furthered the ultrastructural understanding of HIV-1 transmission within lymphoid tissue.
| HIV/AIDS remains a global public health problem with over 33 million people infected worldwide. High-resolution imaging of infected tissues by three-dimensional electron microscopy can reveal details of the structure of HIV-1, the virus that causes AIDS, how it infects cells, and how and where the virus accumulates within different tissue sub-structures. Three-dimensional electron microscopy had previously only been performed to image infected cultured cells or purified virus. Here we used three-dimensional electron microscopy to examine an active infection in the gastrointestinal tract of HIV-1–infected mice with humanized immune systems, allowing visualization of the interplay between the virus and host immune cells. Recapitulating the course of infection in humans, immune cells were depleted in infected humanized mouse gut-associated lymphoid tissue, and individual HIV-1 particles were detected as they budded from host cells and accumulated in pools between cells. HIV-1 was mapped to different substructures and cell types within the gut, and free virions were found to accumulate in pools between cells and also to infect adjacent cells via regions of cell-to-cell contact called virological synapses. Our three-dimensional imaging of an HIV-1 infection in tissue uncovered differences between cultured cell and tissue models of HIV-1 infection and therefore furthered our understanding of HIV-1/AIDS as a disease of mucosal tissues.
| HIV-1 remains a significant public health concern with over 33 million people infected world-wide [1]. Most HIV-1 transmissions occur across an epithelial barrier, resulting in generation of a founder population within the mucosa, viral dissemination to lymphatic tissue, and exponential viral replication throughout the lymphatic system [2]. These events result in depletion of most CD4-positive T cells in mucosal compartments, and establishment of a reservoir of resting cells with integrated provirus that is not susceptible to antiretroviral therapy. In the absence of therapy, progressive immune system collapse and progression towards AIDS ensue in most infected persons.
Accumulating evidence indicates that both acute and chronic HIV-1 infection profoundly affect the gastrointestinal (GI) tract [3], [4]. Studies of SIV infection in non-human primates demonstrated that intestinal CD4 T cell depletion occurs within days, even before T cell depletion can be detected in the peripheral blood or lymph nodes [5]; similar events occur in HIV-1–infected humans [2], [6]. Several features of the GI tract facilitate its susceptibility to HIV-1 infection: (i) the GI mucosa includes high levels of pro-inflammatory, HIV-1–stimulatory cytokines produced by exposure to antigens in the external environment, (ii) a dense clustering of cells that facilitates cell-to-cell transmission, and (iii) a majority of the activated memory T cells expressing CD4 and CCR5 that serve as the preferred target cells for HIV-1 infection [7], [8]. Indeed, the gut-associated lymphoid tissue (GALT) harbors the greatest concentration of potential HIV-1 target cells in the human body [9]; >50% of CD4 T cells from the lamina propria in the lower GI tract are destroyed during acute HIV-1 infection, and early infection of the GALT is believed to be central to chronic HIV-1 infection and disease progression [10], [11]. Furthermore, the presence of CD4 and CD8 T cells, dendritic cells, and macrophages in the GALT make this tissue an integral site for HIV-mediated immune depletion.
Mouse models with humanized immune systems are emerging as a tractable, cost-effective means by which to study HIV-1 infection in mucosal lymphoid tissue [12]. One such model, humanized bone marrow/liver/thymus (BLT) mice, are individually created by transferring human fetal thymic and liver organoid tissues, along with CD34-positive human stem cells, into immunocompromised mice. BLT mice reconstitute significant levels of human lymphoid immune cells; e.g., T and B cells, monocytes, dendritic cells and macrophages in peripheral blood and organs including the GI tract [13], [14]. Important aspects of human HIV-1 infection are recapitulated in this system, including T cell depletion in the gut and peripheral blood, and both systemic and mucosal virus transmission during the course of the disease [15], [16]. Furthermore, BLT mice exhibit high levels of human immune cell engraftment at mucosal sites and significant antigen specific immune responses by multiple cell types [17], [18].
Electron microscopy (EM) was instrumental in the original identification of HIV-1 [19], [20]. Subsequently, diagnostic EM analyses of biopsies from infected patients revealed important aspects of HIV-1 transmission in humans at varying stages of infection, from early acute disease to AIDS progression [21]. More recently, 3-D EM, specifically electron tomography (ET), cryoelectron tomography (cryoET) and ion-abrasion scanning electron microscopy, have been applied at increasingly higher resolutions, facilitating improved understanding of HIV-1 virion structure [22]–[24], virus budding [25], [26], and virus transmission between immune cells [27], [28]. 3-D EM of isolated virions and infected cells can provide a detailed understanding of HIV-1 ultrastructure and transmission between cultured cells, but does not address the complex cellular environment found in mucosal tissues within an organism experiencing an active infection.
Here we used ET to analyze GALT from humanized HIV-1–infected BLT mice in order to visualize HIV-1 infection in mucosal tissues in 3-D at ultrastructural resolution. These analyses allowed us to localize infected substructures within intestinal tissue, classify virions as mature or immature, identify infected cells, visualize structures we interpreted as components of the host cell machinery involved in viral budding, and assess the propensity for viral spread by cell-to-cell versus free virus routes of infection. In parallel studies, we used immunofluorescence (IF) and immuno-electron microscopy (immunoEM) to verify the identities of viral particles, locations of infected tissue, and to distinguish human from murine and infected from uninfected cells.
Human hematopoietic cells derived from transplanted human stem cells have been shown to repopulate the GALT of BLT mice, and HIV-1 infection of these mice results in CD4 T cell depletion, initially in GALT and then systemically [13], [16]. Following established protocols [13], BLT mice were infected with HIV-1 approximately 20 weeks after transfer of human immune tissues and cells, using only mice that met the following criteria for adequate human immune reconstitution: >25% of peripheral blood cells were within a lymphocyte gate on forward-versus-side scatter plots; >50% of cells in the lymphocyte gate were human (human CD45+/mouse CD45−); and >40% of human cells in the lymphocyte gate were T cells (human CD3+). Ten to twenty weeks post infection, mice were sacrificed and segments of small intestine and colon were excised. IF was used to survey locations of HIV-1–infected cells in GALT (Figure 1A,B). Following infection with HIV-1, human CD4 T cells were depleted from the lamina propria (Figure 1B), as previously reported [13], [16]. Staining for the p24 capsid protein of HIV-1 localized primarily in CD4+ cells in regions near the crypts (Figure 1B, inset), which harbor significant populations of immune cells and multipotent stem cells [29]. No evidence of human cells or HIV-1 infection was found in non-humanized infected controls (data not shown).
We next analyzed GALT samples in parallel by ET and immunoEM/ET. Tomography of frozen hydrated tissue samples by cryoET was not possible because the samples were too thick for imaging without sectioning and were infectious biohazards. We therefore imaged fixed and sectioned samples, either positively-stained plastic-embedded or negatively-stained methylcellulose-embedded sections. For ET alone, preservation quality was improved by lightly fixing HIV-1–infected tissue with aldehydes and then further processing them by high-pressure freezing and freeze substitution fixation [30]. This “hybrid” fixation method allowed for safe handling of infectious material and obviated the most structurally damaging steps of traditional chemical fixation [31], yielding well-preserved positively-stained samples. Tomograms were reconstructed from 200 nm or 300 nm sections, often in montaged serial sections of volumes up to 6.1 µm×6.1 µm×1.2 µm. Although these samples could not be used for immunoEM because antibody epitopes are rarely accessible in epoxy-embedded, positively-stained samples [32], analogous GALT samples generated from the same animal were prepared for immunoEM/ET as negatively-stained methylcellulose-embedded sections [33]. Measurements of virions and other structures reflected proportional thinning typical of plastic-embedded and negatively-stained samples [34]. Consequently most structures were ∼30% smaller than counterparts from cryoEM studies or virions in solution or in cultured cells [22]–[24], [35], [36].
ET surveys of HIV-1–infected BLT mouse GALT revealed budding virions (Figure 2A,B; Figure S1A) and free mature and immature particles (Figure 1C, Figure 2C–D, Figure S1B). Virions were detected in all HIV-1–infected mice, while none were found in mock-infected controls (data not shown). The virions were verified as HIV-1 using antibodies against HIV-1 p24 and the envelope spike (Figure 2E,F). Virions were imaged in tissue at all stages of egress, from early plasma membrane Gag assembly to nearly completed buds and fully mature, free HIV-1 (Figure S2). Budding profiles and immature free virions were distinguished by core structures that exhibited radial layers and often appeared as an incomplete internal sphere (a “C” shape in projection) [24], [26]. Mature HIV-1 particles were distinguished from immature particles by the collapse of their cores into a variety of conical shapes, typically “bullet-shaped” cones but often cylinders or ellipsoids [23], [37] (Figure S2). Although envelope spikes on HIV-1 and SIV can be distinguished in positively-stained samples [38], we observed few projections emanating from virion surfaces, consistent with biochemical and cryoET studies of purified HIV-1 virions that demonstrated a low number of envelope spikes: an average of ∼14 (ranging from 4–35) per virus particle [39], [40].
After establishing that HIV-1 could be identified in infected BLT GALT by ET and immunoEM, we surveyed GALT samples to determine locations of infection. Plastic-embedded sections of small intestine (jejunum and ileum) and large intestine (colon) were examined to find HIV-1 and infected cells, which were identified by budding profiles at their surfaces. Within a given animal, the extent of infection and the distribution of virions were similar between the small and large intestine. However, virions were found in differing amounts amongst sub-structures in the intestinal mucosa. The largest populations of HIV-1 virions and infected cells identified by EM were located in crypts (Figure 1A,C), consistent with IF (Figure 1B). Approximately one in ten crypts showed evidence of HIV-1 infection. The mucosal region surrounding the villus base and the crypts contained few free virions or infected cells (∼1 in >100); when present, infected cells were often near a capillary or venule (Figure S1A). The numbers of free virions and infected cells in the lamina propria were less than in the crypts (Figure S1B). Typically, infected lamina propria were in villi continuous with infected crypts. Few infected cells or virions were found in the smooth muscle layer surrounding the intestine. In addition, free virions were rarely found in blood vessels because even the high viral loads of the HIV-1–infected BLT mice from which the samples were derived (up to 126,000/mL in peripheral blood) translated to only ∼1×10−7 virions/µm3. Thus at the scale of individual EM images or even large-format tomograms, HIV-1 virions would be rarely seen, and our imaging of >50 blood vessels contained within tomograms yielded only two examples of free virions (data not shown).
To identify potential human target cells of HIV-1 infection, we conducted immunoEM (Figure S3A–D) using antibodies specific for human proteins. Human CD4 localized primarily to the plasma membrane in uninfected cells (Figure S3A), but we found extensive CD4 labeling in the endoplasmic reticulum (ER) of CD4-positive cells with budding virions or nearby free virions (Figure S3B), correlating with the finding that HIV-1 Vpu induces cell surface CD4 to redistribute to the ER to avoid surface retention of newly-forming virions [41]. Double labeling with antibodies against HIV-1 Nef and human CD4 (Figure S3C) or class I human leukocyte antigen (HLA) and human CD4 (Figure S3D) confirmed that cells exhibiting a predominantly ER localization of CD4 were human cells infected with HIV-1. No instances of Nef expression were found in uninfected or non-human cells (data not shown), which served as an internal control for the specific of the antibodies and further validated the BLT model of HIV-1 infection.
Tomograms of immature virions derived from negatively-stained infected tissue revealed detailed structural information. With the exception of the widening of lipid bilayer membranes, presumably caused by obligatory light fixation associated with this method, the overall architecture of the Gag shell in immature virions conformed to known properties of HIV-1 determined from studies of viruses isolated from cultured cells [22], [24], [35], [36], [42] (Figure 3; Figure S4). Indeed, the immature virions in our tissue samples (Figure 3, S4A,B) exhibited features observed in cryoET analyses of purified frozen hydrated HIV-1 [22], [24] (Figure S4C); e.g., individual layers of the Gag shell, including the hexagonal lattice of the capsid (CA) portion (Figure 3). The symmetry of the CA layer was confirmed by hexagonal features in the Fourier transforms of immature virions, but not in transforms of adjacent cytoplasm (Figure 3B; Figure S4A).
More than 50 crypts of Lieberkühn were imaged in the course of this study. In the ∼10% of crypts that were infected, HIV-1 virions were found primarily in pools within dilated regions of intercellular spaces (Figure 1C; Figure 4; Figure S5; Movie S1). Pools were defined as a population of virions within an intercellular space that was continuous within a given 3-D volume. Multiple intercellular spaces could be present within the volume, but unless the spaces were visually continuous, virions within them were regarded as separate pools (Figure 4B,C). The numbers of free virions in intercellular pools ranged from 5 to >200. In single-frame tomograms (3.2 µm×3.2 µm×200 nm), most pools contained 10–40 particles. Larger pools were observed in serial-section reconstructions encompassing greater tissue volumes. In longitudinal sections of crypts, most pools were found between the base and middle. Infected human immune cells, identified by the presence of budding virions, were often found near virion pools.
Virions within a given pool were distinguished as mature or immature based on the presence of a cone-shaped core in mature particles and radial Gag layers in immature particles (Figure S2). The numbers of mature and immature particles in intercellular pools were quantified within reconstructed volumes of infected crypts. Pools could be classified as either “mostly mature” or “mostly immature” (Figure S5A). Of >100 pools containing many hundreds of virions, approximately 90% of pools were classified as mostly mature and 10% were mostly immature.
Potential HIV-1 target cells and pools of virions were plentiful in GALT, particularly in crypts, thus it was not always possible to determine from which cell a particular virion population originated. In order to quantify virions from a particular cell and infer temporal data with respect to virion pools, we imaged regions of the intestinal smooth muscle layer (Figure 1A), which contains few HIV-1 target cells. Figure S5B shows an HIV-1-infected cell in the smooth muscle. The surface of this cell exhibited several HIV-1 budding profiles, and groups of free virions were located both in close proximity to and at varying distances from it. There were no other infected cells within several microns, thus we could be confident that nearby free virions had originated from that cell. We found that 62% of virions (n = 16) in immediate proximity (≤0.5 µm) to the cell were immature, while 73–75% of virions in groups located 0.8 µm (n = 15) and 1.3 µm (n = 32) away were mature.
Of >100 virion pools that were imaged, most were in obvious extracellular spaces. Some pools (∼5%) appeared to be intracellular, but were revealed by ET to be connected to the extracellular space by narrow channels that averaged ∼27 nm in width (range = 23–32 nm; n = 6) (Figure 4D–E) and contained 2–20 mature virions. A few of the budding regions were large enough that potential continuities with the plasma membrane were outside of the reconstructed volume. The presence of seemingly intracellular virion pools connected to microchannels could identify the cell as an infected macrophage, a cell type in which internal virus-containing compartments were proposed to represent specialized domains of the plasma membrane that were sequestered intracellularly [43], [44] and/or endosomal compartments [45], [46].
ET surveys of HIV-1 infected GALT showed evidence of virological synapses for direct cell-to-cell virus transmission, a route of HIV-1 transmission within tissues whereby a virus buds from an infected cell and directly contacts and infects an adjacent uninfected cell [47]. Formation of a virological synapse results from interaction of gp120 on an infected cell with its receptors on a target and also involves other host proteins such as LFA-1 and ICAM proteins on the surfaces of both the donor and target cells [48], [49]. A large format reconstruction (2×3-frame montage) of GALT revealed an HIV-1–infected cell, likely a dendritic cell or macrophage based on the convoluted processes intercalating between neighboring cells (Figure 5A; Movie S2). A presumptive virological synapse was visualized as a region of contact between a budding virion and an adjacent cell (Figure 5B; Movie S2). Although this positively-stained sample could not be examined by immunoEM, we found similar features in negatively-stained samples that labeled with antibodies against LFA-1 and ICAM-1 (Figure 5C,D), supporting the identification of these regions as virological synapses. In another example, an infected cell that showed numerous budding profiles included one that closely approached the surface of an adjacent cell although still attached to its host cell via a ∼50 nm neck (Figure S6A). The surface region of the cell proximal to the approaching bud was denser than surrounding surface regions and extended toward the bud. In a third example, a budding profile from an infected cell appeared to project into an invagination in the plasma membrane of an adjacent cell (Figure S6B,C). Tomographic views through the volume containing this region showed the boundaries of the invagination followed the contours of the budding profile (Figure S6C), suggesting a dynamic response to the approaching nascent virion.
By reconstructing a large 3-D volume of infected tissue, we could address whether direct cell-to-cell transmission was an obligatory means of virion transfer between two adjacent cells. Movie S1 shows a 1.4 µm×2.9 µm×1.2 µm tomogram in which the outlines of two adjacent cells were distinguished. Both cells were identified as infected by the presence of budding virions and were therefore HIV-1 targets. A region resembling a virological synapse was not observed in the reconstructed volume, however a large accumulation of free mature virions were present in the space between the cells, suggesting that direct cell transfer is not a required mechanism of HIV-1 transmission between closely apposed infected cells. The lack of an observed virological synapse in such cases could be the consequence of CD4 down-regulation in the infected cells. However the existence of natural recombinant HIV-1 strains, which could result from infection by one HIV-1 strain of a cell already infected with a different viral strain [50], suggests that residual CD4 remaining at an infected cell surface can allow for infection via free virus or direct cell-to-cell transfer.
The large number of budding virions within BLT GALT tomograms offered the opportunity to characterize structural aspects of HIV-1 budding in infected tissue (Figure S7). Actin filaments were often found near forming buds (Figure S7A) similar to those previously observed at HIV-1 budding sites in cultured cells [25]. Budding profiles exhibited varying lengths of necks, including some with no neck (Figure 3C,D; Figure S7B). In the colon, early budding virions without necks were often observed forming from surfaces that were not obviously plasma membrane. However, serial-section tomography revealed that these domains were usually continuous with the plasma membrane proper, indicating that they were convoluted regions of the cell surface and not distinct cytoplasmic compartments. Some budding virions exhibited necks with 50–80 nm lengths and varying widths (Figure S7C), with narrower necks likely representing those approaching scission. Virions were also observed budding at the ends of extremely long cellular projections (Figure 2A,B) that were likely filopodia extending from dendritic cells, as observed in culture [27], [51].
ET analyses of HIV-1 budding in cultured cells revealed a subset of RNA-free immature virions with a novel “thinner” Gag lattice lacking the nucleocapsid-RNA layer, which were suggested to represent aberrant, noninfectious virions resulting from premature activation of HIV-1 protease [25]. Using our measuring convention, the previously-described thin Gag lattice [25] measured 9–10 nm. Analysis of 100 free or budding immature virions from tissue samples yielded no examples with a thin (9–10 nm) Gag lattice that lacked discernable RNA densities. Instead, we found that the Gag lattice widths in all of the immature virions we surveyed (n = 100) within infected tissue was 14.6±0.8 nm (Figure S4B, Figure S7D,E); significantly different than the thin 9–10 nm Gag lattices previously described [25]. In addition, there were no systematic structural differences in Gag lattices correlating with the type of budding profile: the Gag shell thicknesses measured in 30 long-necked and 30 neck-free buds were similar and presumptive RNA densities were present in all cases (Figure S7D).
Release of HIV-1 virions from infected cells involves recruitment of the host endosomal sorting complexes required for transport (ESCRT) machinery to sites of virus assembly by the Gag polyprotein [52]. These interactions culminate with the polymerization of ESCRT-III proteins, recruitment of vacuolar protein sorting-associated protein 4 (VPS4) ATPase oligomers, fission of the cellular membrane attaching the virion to the host cell, and disassembly of the ESCRT machinery.
We used antibodies against ESCRT-III proteins, human charged multivesicular body proteins (hCHMPs) 1B and 2A, and ALG2-interacting protein X (ALIX), an ESCRT adaptor protein that facilitates the transport of Gag to the cell membrane [53] and can mediate interactions between ESCRT-I and ESCRT-III complexes [54], to detect components of the ESCRT pathway in infected tissue by immunoEM. We found that hCHMP1B, hCHMP2A and hALIX localized predominantly to the neck regions of budding HIV-1 virions (Figure 6A-C). The labeling was specific, but sparse due to the small number of epitopes and their availability only at section surfaces.
At scission regions of budding virions in which the neck of the bud was greater than half the diameter of the bud, clusters of 4–6 spoke-like projections nearly 20 nm in length radiating from a centralized origin at the base of the budding virion were sometimes observed (Figure 3C,D; Figure 6D; Figure S8A; Movie S3). As the larger neck diameter may define these buds as being at an initial stage of egress, these radial projections could represent components of the early portions of the ESCRT pathway such ESCRT-I or ALIX recruited by assembling HIV-1 Gag molecules. Indeed, the size and shape of the structures approximate models for the ESCRT-I-II supercomplex determined by a combination of spectral techniques [55]. By contrast, in tomograms of budding virions with narrower necks (less than half the diameter of the bud itself), we observed parallel electron dense striations circumscribing the neck of the bud in both positively- and negatively-stained sections (Figure 7A–E; Figure S8B,C; Movie S4) suggestive of ESCRT-III components polymerizing at membranes [56], [57]. Similar electron dense striations were detected at the necks of budding virions arrested at a late stage by expression of dominant-negative ESCRT-III or VPS4 proteins [58]. In addition, budding profiles in positively-stained samples often showed 1–5 electron-dense “spots” in the neck or base of a bud (Figure 7F,G; Movie S5). The spots were observed in over half of ∼50 budding profiles in which the diameter of the neck was half or less of the diameter of the budding virion; presumably a late stage of budding. Available antibodies against VPS4 did not stain efficiently by immunoEM, however their interpretation as VPS4 oligomers was consistent with fluorescence imaging showing recruitment of 2–5 VPS4 dodecamers to the sites of viral budding just prior to virion abscission [59], [60]. In addition, the size and relative shape of the putative VPS4 densities (Figure 7G) correlated with cryoEM reconstructions of VPS4 [61].
Many aspects of the pathologies related to HIV-1 infection, including immune cell death and tissue destruction, occur in GALT. However, 3-D ultrastructural details of a natural GALT infection were unknown because ET had not been applied to in vivo infection in GALT or other lymphatic tissues. BLT humanized mice are an emerging model for studying HIV-1 infection, and BLT GALT maintains cellular architecture, cell-cell interactions, immune cell populations and signaling more accurately than cell culture infection models [12]. As such, the BLT mouse system is a reliable model for structural studies of HIV-1 infection in a tissue environment. In addition, the inclusion of human thymic tissue in BLT mice allows for T cell maturation in the context of human, rather than murine, MHC proteins; an aspect that is not present in humanized mouse model systems produced with human hematopoietic stem cells but without thymic tissue.
Dense areas of HIV-1–infected cells, including CD4 T cells, macrophages and dendritic cells, and free HIV-1 virions were found in crypts within BLT GALT by IF, ET and immunoEM (Figure 1B,C). Blood vessels were imaged in mice with a wide range of viral loads; however, we were unable to correlate the relative abundance of virions detected in GALT with the viral load measured in the blood. In fact, only two examples of virions within blood vessels of BLT mice were detected as compared with hundreds of virions within mucosal tissue. This finding is consistent with reports highlighting a discrepancy between blood viral load and HIV-1 levels in tissues [62], [63]. Thus analysis of HIV-1–infected tissues by methods such as ET may provide valuable information in addition to blood viral load measurements when evaluating treatment regimens.
Potentially relevant to infection and immune cell recognition mechanisms, large pools of free HIV-1 were found within infected GALT (Figures 1C, Figure 4, Figure S5). Although most pools contained mainly mature virions, some pools contained a majority of immature virions (Figure S5A), a phenomenon not observed in EM studies of HIV-1 infection of cultured cells. Pools of virions were usually found between cells, but also in compartments that appeared to reside within cells. These compartments were often connected to the cell surface by microchannels 20–30 µm in width (Figure 4D). These narrow channels likely undergo dynamic changes in morphology, as their width would be too narrow to accommodate passage of HIV-1 to the extracellular space. We interpreted such channels as invaginations of the plasma membrane, consistent with reports that macrophages can assemble HIV-1 in intracellular virus-containing compartments created by internally sequestered plasma membrane [43], [44], [64]. In infected tissue, we found that pools of HIV-1 virions located between two cells could contain mature or immature virions (Figure S5A), whereas the intracellular pools connected by microchannels contained only mature virions (Figure 4D). One possibility for the difference in maturation states of inter- versus intracellular pools of HIV-1 is that intracellular virions connected to the extracellular space by microchannels are not subject to movement by interstitial fluid through intestinal tissue and could remain in a single location long enough to complete maturation, perhaps representing viral reservoirs that allow low levels of de novo infection to proceed in the presence of anti-retroviral therapy and/or antibodies [65].
Although the discovery of virion pools suggested that infection by free virus could occur within infected tissue, we also found evidence of direct cell-to-cell transmission of HIV-1 in infected GALT (Figure 5; Movie S2). The virological synapse is a mechanism of cell-to-cell transmission in which juxtaposition of an infected and uninfected cell promotes infection by directing viral assembly, budding, maturation, and fusion machinery to discrete locations of cellular contact between cells [47]. In a large 3-D reconstruction of two adjacent HIV-1–infectable target cells (Movie S1), we found a large pool of mature virions but no evidence for a virological synapse, suggesting that formation of virion pools and infection by free virus can occur even when adjacent cells are both infectable by HIV-1, or had been infectable prior to down-regulation of CD4. In addition, this result validated our frequent finding of large pools of free virions in HIV-1–infected tissue, demonstrating that this phenomenon was not necessarily the consequence of the juxtaposition of a human infected cell and a murine cell, as may occur in BLT GALT.
EM studies of HIV-1 virions produced in cultured cells suggested that maturation is a rapid process, because intermediate maturation states were not detected and because virions found near cells were predominantly mature [66]. However, our finding of pools containing immature virions in proximity to infected cells in tissue suggested maturation dynamics and/or virion diffusion properties differ between cells organized within tissue versus those cultured in vitro. In addition, we never found examples of RNA-negative budding virions with a thin Gag lattice in tissue samples, as had been observed in ∼18% of immature particles in cryoET analyses of HIV-1 produced in cultured cells [25]. Thus, higher numbers of aberrant particles and of exclusively mature virions in close proximity to producer cells could be artifacts of producing virions in cultured cells, suggesting that the BLT model of in vivo infection more accurately recapitulates the HIV-1 lifecycle than cell culture models.
Although ET relies on fixed tissue and cannot directly recapitulate virion dynamics in live cells, our studies provided a glimpse into temporal aspects of HIV-1 maturation. We determined that an isolated infected cell within a large tissue volume was the sole producer of several populations of imaged virions located at varying distances from the cell. This allowed us to determine that a single infected cell can produce at least 63 viruses (the number of virions in the three pools in Figure S5B). The total number of virions produced per cell is likely far larger, as regions above and below the cell were not represented in the reconstruction. Using a predicted rate of interstitial fluid movement in intestinal tissue of 0.1–2 µm/sec [67], a virion would travel 2 µm in 1–20 sec, indicating that maturation could occur just seconds after release from an infected cell. This argues that, in tissue, virions found ∼2 µm away from a producer cell budded only seconds earlier, supporting an assumption of rapid virus maturation. Furthermore, our finding of mostly immature virion pools in close proximity to the infected cell and mostly mature virion pools further away from the cell (Figure S5B) is consistent with synchronous release and subsequent maturation of HIV-1. The trigger(s) for and/or block(s) to maturation that could promote synchronized virus maturation in tissue could include proximity to an infected producer cell, lack of an adjacent target cell to form a virus synapse, and/or contact with a non-infectable cell.
Late events in HIV-1 budding had been visualized by fluorescence microscopy [59], [60] and ET of cultured cells [25], [26], [66], [68] but not yet in infected tissue. In our infected tissue samples, we detected distinct electron dense structures near virions at various stages of budding that may represent aspects of the host cell ESCRT machinery at sites of HIV-1 egress (Figures 6,7,S8). Although we could not identify the structures conclusively, our assignments of their possible identities are consistent with what is known temporally about the involvement of host cell machinery in HIV-1 budding and release from infected cells. Tomograms revealed that virions in the initial stages of budding contained 4–6 spoke-like projections emanating from the center of the forming neck of the budding virion (Figure 3C–D, Figure 6D, Figure S8A), potentially representing components of host ESCRT-I or ALIX recruited by assembled HIV-1 Gag. The shape, size, and temporal occurrence of these structures agree with a proposed model for vesicle budding and fission based on biophysical analyses of the ESCRT-I-II supercomplex in solution [55]. Virions at later stages of budding that were connected to the host cell membrane by thinner (<50 nm) elongated necks showed parallel, electron dense striations along the membrane surface of the neck (Figure 7A–E, Figure S8B) that we interpreted as features of polymerized ESCRT-III proteins [56], [57]. These late budding profiles often displayed dense spots along the center of the neck (Figure 7F,G) that we suggest were VPS4 oligomers recruited immediately prior to fission of the new virion from the cell membrane, consistent with fluorescence microscopy studies [59], [60]. ET of budding virions within tissue allowed a spatial and temporal interpretation of HIV-1 budding. First, the Gag lattice reached a sufficient point of closure, which allowed formation of a spoke-like structure at the base of the early budding virion. Next, the virion formed an elongated neck; concomitant with polymerization of host cell factors in a spiral around the inside of the membrane [56], [57]. In tomographic slices of budding profiles, these presumptive spirals appeared as two or more parallel lines bisecting the neck region (Figure 7A–E). Finally, the recruitment of large oligomers, possibly VPS4, coincided with the separation of the virion from the infected cell [59], [60], completing the budding process (Figure 7F,G).
In summary, our 3-D ultrastructural characterization of HIV-1–infected GALT identified dense regions of virus transmission, provided insights into the temporal nature of virus maturation, revealed HIV-1 transmission occurring by both free virus and direct cell-to-cell mechanisms, and demonstrated important differences between cultured cell and tissue HIV-1 infection models. Differences included the identification of free immature virions and the scarcity of aberrantly formed viral particles during an active infection. The high resolution of our positively- and negatively-stained tissue samples allowed 3-D visualization of HIV-1 transmission within lymphoid tissue, providing a new approach for understanding HIV-1 infection in vivo.
Humanized mice were prepared and cared for in an AAALAC-certified animal care facility at the Massachusetts General Hospital (OLAW Assurance #A3596-01), in accordance with a protocol approved by the MGH IACUC (Protocol #2009N000136/25). The protocol as submitted and reviewed conforms to the USDA Animal Welfare Act, PHS Policy on Humane Care and Use of Laboratory Animals, the “ILAR Guide for the Care and Use of Laboratory Animals” and other applicable laws and regulations. Every effort was made to minimize animal suffering throughout all experiments. Human tissue for preparing the humanized mice was procured and used in accordance with a protocol approved by the local Institutional Review Board (Partners Human Research Committee, Protocol #2012-P-000409/5).
NOD/SCID/IL2Rγ−/− mice (The Jackson Laboratory) were reconstituted with human tissue as described [13]. Approximately 20 weeks after transfer of human immune tissues and cells, mice were infected intraperitoneally with 1×105 TCID50 of JR-CSF HIV-1. Every 2 weeks after infection, ∼200 µl of blood was obtained through puncture of the retro-orbital sinus or submandibular vein for determination of HIV-1 plasma viral load. Viral RNA was isolated using the QIAamp Viral RNA Mini Kit (Qiagen) and viral loads were determined by quantitative RT-PCR using primers for HIV-1 Gag [69]. Immunofluorescence experiments were conducted using tissues from a mouse with a blood viral load of 940,000 copies/mL. ImmunoEM experiments were conducted using tissues from a mouse with a viral load of 100,000/mL. The remaining mice had blood viral loads as follows: 0 (control) 9,800, 8,400, 18,500 and 126,000 copies/mL. As previously shown, the range of blood viral loads did not correlate with virus populations found in tissue samples [62], [63]. Infected mice were sacrificed 10–20 weeks post infection, and then necropsied with segments of small intestine and colon excised and fixed. Immunofluorescence (IF) studies were conducted as described in the Supplementary Methods (Text S1).
For positively-stained samples, HIV-1–infected tissue was prepared by a hybrid method that employed primary chemical fixation followed by high-pressure freezing/freeze substitution fixation (see Text S1). Negatively-stained samples were prepared as described [33] and in the Supplementary Methods (Text S1).
200 nm positively-stained sections and 90 nm negatively-stained sections were imaged in a Tecnai-12 G2 transmission electron microscope at 120 KeV, and 300 nm sections were imaged in a Tecnai G2 TF30-FEG microscope at 300 KeV (FEI Company, Holland) in a dual-axis tomography holder (2040; Fischione Instruments, Export, PA). Dual axis tilt series (+/−60°; 1° intervals), including multi-frame montaged datasets, were acquired automatically using the SerialEM software package [70]. Tomographic data were aligned, backprojected, analyzed and segmented using IMOD [71].
The Gag lattice in the tomographic slice closest to the equator of each virion or budding profile slice was measured in five randomly selected areas as a line from the base of the innermost layer to the outside of the outermost layer (green lines in Figures S3B and S6D) using IMOD [71]. The values were combined to give an average Gag thickness for each virion. The symmetry of the Gag lattice was evaluated by Fourier transformation of Gag regions in negatively-stained tomograms. Budding profiles were viewed in tomographic slices taken near the surfaces of their Gag layers (Figure 3B) and images were displayed using the Slicer tool in IMOD, which allowed for 3-D rotation. When the Gag structure was optimally oriented, the image was transformed to Fourier space using an FFT algorithm within IMOD.
Selected tomographic datasets are available at http://www.br.caltech.edu/bjorker/ladinsky, on the Electron Microscopy Databank (http://www.emdatabank.org) under submission number 28207, or will be provided upon request.
|
10.1371/journal.pgen.1006379 | Effect of Insulin Resistance on Monounsaturated Fatty Acid Levels: A Multi-cohort Non-targeted Metabolomics and Mendelian Randomization Study | Insulin resistance (IR) and impaired insulin secretion contribute to type 2 diabetes and cardiovascular disease. Both are associated with changes in the circulating metabolome, but causal directions have been difficult to disentangle. We combined untargeted plasma metabolomics by liquid chromatography/mass spectrometry in three non-diabetic cohorts with Mendelian Randomization (MR) analysis to obtain new insights into early metabolic alterations in IR and impaired insulin secretion. In up to 910 elderly men we found associations of 52 metabolites with hyperinsulinemic-euglycemic clamp-measured IR and/or β-cell responsiveness (disposition index) during an oral glucose tolerance test. These implicated bile acid, glycerophospholipid and caffeine metabolism for IR and fatty acid biosynthesis for impaired insulin secretion. In MR analysis in two separate cohorts (n = 2,613) followed by replication in three independent studies profiled on different metabolomics platforms (n = 7,824 / 8,961 / 8,330), we discovered and replicated causal effects of IR on lower levels of palmitoleic acid and oleic acid. A trend for a causal effect of IR on higher levels of tyrosine reached significance only in meta-analysis. In one of the largest studies combining “gold standard” measures for insulin responsiveness with non-targeted metabolomics, we found distinct metabolic profiles related to IR or impaired insulin secretion. We speculate that the causal effects on monounsaturated fatty acid levels could explain parts of the raised cardiovascular disease risk in IR that is independent of diabetes development.
| Impaired glucose homeostasis leads to diabetes and cardiovascular disease and has two main components: failure to secrete enough insulin from pancreatic β-cells and reduced insulin-stimulated cellular uptake of glucose and other nutrients in target tissues (insulin resistance, IR). We used metabolomics analysis in non-diabetic persons to measure a non-selective range of small molecules including amino acids, lipids, and sugars. Pathway analysis highlighted distinct metabolic pathways linked to IR (e.g. bile acid production) and impaired insulin secretion (fatty acid biosynthesis), but causal directions remained unclear. Mendelian Randomization (MR) analysis can test for causal effects in observational studies in the absence of randomized controlled trials. Using MR analysis in up to four large independent studies, we found evidence that IR causes a decrease in levels of the main endogenous monounsaturated fatty acids palmitoleic acid and oleic acid, as well as suggestive evidence for higher levels of the amino acid tyrosine. We provide a possible explanation for parts of the diabetes-independent risk of cardiovascular disease in persons with IR.
| Insulin resistance (IR) is a major precursor of type 2 diabetes (T2D) [1], and constitutes an independent risk factor for cardiovascular disease (CVD) [2] and for certain cancer types [3, 4]. In IR, the demands on pancreatic β-cells to produce insulin increase and blood glucose levels rise if β-cell function is impaired. The metabolic effects of IR and declining β-cell function are not fully characterized and causal relationships are difficult to disentangle due to the lack of randomized controlled trials. Associations between the “gold standard” hyperinsulinemic-euglycemic clamp method [5] for measuring whole-body IR and non-targeted metabolomics profiling previously identified α-hydroxybutyrate as a biomarker for IR in 399 non-diabetic persons [6]. Additional insights for causal directions may come from profiling circulating metabolites combined with genotyping, as previously demonstrated in causal investigations of adiposity and the metabolome using a Mendelian Randomization (MR) approach [7] and for causal effects of uric acid on IR and T2D risk [8]. Mendelian randomization analysis can test the causal relationship between an exposure and an outcome variable in the absence of randomized controlled trials [9]. Exposure-associated single nucleotide polymorphisms (SNPs) can be used as instrumental variables (IVs) because allelic variants are randomly allocated at meiosis and therefore independent of bias from confounding and reverse causation. Genotype-based methods like MR can inform drug targeting: For example, the association between genetic variants in PCSK9, reduced low-density lipoprotein-cholesterol levels and lower coronary heart disease risk [10] predicted the clinical success of pro-protein convertase subtisilin kexin 9-inhibitors [11]. Conversely, MR studies confirmed the lack of a causal association between plasma high-density lipoprotein-cholesterol and cardiovascular events [12].
The causal effects of impaired insulin secretion and IR on blood metabolites have not been assessed before in a large-scale metabolomics framework and could pinpoint key mediators of the risk of adverse health events. We aimed to identify metabolic pathways related to IR and impaired early-phase insulin secretion during an oral glucose tolerance test (OGTT) in a large European sample and applied MR methods in additional cohorts to assess potential causal effects of impaired insulin secretion and IR. Using non-targeted metabolomics [13], we identified 52 circulating metabolites related to either IR or insulin secretion that implicated distinct metabolic pathways and we found evidence for a causal effect of IR on reduced palmitoleic acid (POA) and oleic acid (OA) levels, as well as on raised tyrosine levels.
The Uppsala Longitudinal Study of Adult Men (ULSAM) cohort of community-dwelling 71-year-old men provides the largest human sample combining plasma metabolomics with an OGTT and the hyperinsulinemic-euglycemic clamp method. We previously developed a bioinformatics pipeline for untargeted liquid chromatography/mass spectrometry (LC/MS) data [14] and were able to annotate among 10,162 spectral features 192 metabolites, on which this study is based. In up to 910 non-diabetic individuals, we used linear regression adjusted for age, sex, and sample quality indicators (S1 Text) to identify fasting metabolite levels associated with physiologic measures of insulin secretion and IR. Table 1 provides sample characteristics and Fig 1 illustrates the study flow. Three outcomes were assessed: IR (clamp M/I), the insulinogenic index (log-IGI30) as a measure of glucose-stimulated insulin secretion [15] and the disposition index (log-DI) for β-cell responsiveness [16], all scaled to SD-units. At the 5% false discovery rate (FDR), 47, 15, and zero metabolites were associated with clamp M/I, log-DI and log-IGI30, respectively (Fig 2, S1 Table). Reduced levels of lysophosphatidylethanolamine (LysoPE) 18:2 and hippuric acid were associated with IR and impaired insulin secretion. Shared positive associations were found for deoxycholic acid glycine conjugate, corticosterone, propranolol, piperine, and three unsaturated fatty acids (FAs; arachidonic, eicosatrienoic, and oleic acid). Higher levels of glycerolipids and several acylcarnitines, and lower levels of glycerophospholipids were exclusively associated with IR. Further, increased levels of two bilirubin species were exclusively associated with impaired insulin secretion. To identify metabolite associations independent of adiposity, we additionally adjusted each model for body mass index (BMI, S1 Table). This reduced the strength of associations to a lesser extent for impaired insulin secretion than for IR and preserved the direction of associations for all metabolites still significant at the 5% FDR (34 for IR and 8 for impaired insulin secretion). Six new associations were detected for IR after BMI adjustment (increased levels of three acylcarnitines, α-tocopherol, myristic acid and bilirubin, Fig 2). Pathway enrichment analysis carried out with MetaboAnalyst 3.0 [17] (http://www.metaboanalyst.ca/) indicated significant enrichment of IR-associated metabolites in primary bile acid synthesis (p = 0.009, 4 metabolites), glycerophospholipid metabolism (p = 0.006, 4 metabolites) and caffeine metabolism (p = 0.016, 2 metabolites) (S1 Fig). Impaired insulin secretion-associated metabolites were enriched in the FA biosynthesis pathway (p = 0.027, 3 metabolites).
We attempted replication for metabolites with suggestive evidence of causation (p < 0.1) in up to 7,824 European individuals in the KORA F4/TwinsUK cohorts that underwent untargeted metabolomics profiling on a different LC/MS platform [19]. The liberal p-value threshold was chosen as a compromise between limited sample size and the risk of missing true positive associations in the MR discovery set and we adopted the conventional threshold of p < 0.05 for significance in all replication analyses. Seven of the nine causally implicated metabolites for IR (excluding MAG(14:0) and 3α,6β,7β-trihydroxy-5b-cholanoic acid) and one of two (bilirubin) for impaired insulin secretion were available in KORA/TwinsUK. We replicated the causal effect of IR on POA (βIV = –0.48, 95% CI –0.93 to –0.03 p = 0.038, all in SD-units) and the tentative effect on OA (βIV = –0.38, 95% CI –0.81 to 0.04, p = 0.078) (Fig 3). We could not replicate the causal effects on tyrosine (βIV = 0.20, 95% CI –0.19 to 0.59, p = 0.316) and hippuric acid (βIV = –0.18, 95% CI –0.61 to 0.24, p = 0.398). A repeat analysis in a subset of 1,432 non-diabetic individuals in the KORA F4 cohort replicated the causal effect on POA (βIV = –1.14, 95% CI –2.13 to –0.15, p = 0.024) and the tentative effect on OA (βIV = –0.82, 95% CI –1.75 to 0.10, p = 0.082) (S1 Text).
In a second attempt to replicate the causal findings for POA and OA, we obtained the publicly available GWAS results for FA fractions in plasma phospholipids from the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium (http://www.chargeconsortium.com/main/results) [20]. The CHARGE study combines five independent cohorts of 8,961 Europeans in age, sex, and recruitment site-adjusted meta-analysis and we performed MR analysis based on the genetic IR score, as all 10 SNPs had been genotyped (S1 Text, FA fractions were converted to SD-units). The results replicated the causal effect of IR on POA (βIV = –0.29, 95% CI –0.54 to –0.05, p = 0.018) and OA (βIV = –0.48, 95% CI –0.73 to –0.24, p = 0.007).
We aimed to replicate the possible causal effect of IR on raised tyrosine using the summary meta-GWAS results from five cohorts of 8,330 Finnish individuals who underwent nuclear magnetic resonance metabolomics profiling [21]. In this sample, no significant effect on serum tyrosine levels was found (β = 0.21, 95% CI –0.16 to 0.57, p = 0.267).
Achieving adequate power in MR analysis requires large sample sizes [22]. However, the differences in analytical methods between studies for the same metabolite make the combination of within-study effects in post-hoc meta-analysis methodologically unsound. To nonetheless explore the effect of increased sample size on causal effects, we combined estimates for each metabolite in inverse variance-weighted fixed-effects meta-analysis and illustrate these exploratory estimates in Fig 3. In addition to the negative effects on POA and OA, the combined analysis indicated a causal effect of IR on higher tyrosine levels (p < 0.05) that was observed as a non-significant trend in each individual study. Given the risk of false positive findings due to multiple testing, these post-hoc findings should be interpreted with caution and require confirmation studies.
To assess for violations of MR assumptions due to genetic confounding, horizontal pleiotropy and IV heterogeneity, we examined a) the association of individual SNPs with the risk factors (S2B Fig)); b) the association of IVs with potential confounders (S3C Fig)); c) scatter plots of IV-outcome v. IV-risk factor associations and funnel plots of IV strength v. IV estimate (S4D Fig)); d) implemented sensitivity analysis for individual SNPs in inverse variance-weighted, log-likelihood and MR Egger regression, including heterogeneity tests (S4 Table, S1 Text [23–25]); and e) discuss in S1 Text the likelihood of bias from canalization and other effects.
We tested for associations of genetic scores with potential mediator variables in age- and sex-adjusted linear regression in ULSAM, PIVUS and TwinGene (S3 Fig) and confirmed the association between worse genetic IR and lower HDL-C as well as smaller waist-hip ratio as reported by Scott et al. [18] (who included ULSAM), and found an association with higher albumin levels. As in [18], the impaired insulin secretion score was positively associated with plasma glucose and unrelated to other traits apart from lower albumin levels and a trend for increased C-reactive protein. For all metabolites with indication for a causal effect (p < 0.1) in the MR discovery sample, we examined individual SNP effects on metabolite levels. The IV-exposure associations were derived from the GWAS results for homeostasis model assessment-IR (HOMA-IR, n = 46,186) and corrected insulin response (CIR, n = 5,318) in non-diabetic persons from the publicly available results of the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC; http://www.magicinvestigators.org/downloads/, converted to SD-units). The association between SNPs and risk factors in MAGIC were consistent and did not indicate heterogeneity apart from one variant (rs11605924) that was associated with CIR in the unexpected direction (S2 Fig). As reported in S4 Table, Cochran’s Q tests failed to detect significant heterogeneity between individual SNPs’ causal estimates. The significant effects of IR on OA and POA levels detected in the main analysis were replicated in all sensitivity tests except in the case of MR Egger regression for POA (slope estimate –0.96, 95% CI –2.15 to 0.22, p = 0.095). Although the intercept estimate in MR Egger regression for OA differed significantly from zero, it was small in magnitude (intercept estimate 0.02, 95% CI 0.00 to 0.04, p = 0.018) and the causal effect was reproduced (slope estimate –0.49, 95% CI –0.82 to –0.16, p = 0.010). One important assumption of MR Egger regression–that any pleiotropic effects of IVs on the outcome be independent of IV strength [23]–cannot be assessed with currently available methods. Because the IV score had been carefully constructed by Scott et al. [18] to limit the likelihood of pleiotropic interference (particularly from BMI) and based on the totality of all MR sensitivity analyses that reproduced the main results in direction and magnitude and failed to indicate significant heterogeneity, we are confident to have excluded pleiotropic effects that could invalidate our main findings as far as possible. However, as discussed in S1 Text, some sources of bias (e.g., canalization) cannot be excluded and the possibility of remaining pleiotropic effects pose limitations that mandate careful interpretation of any MR study.
To further assess our findings of a negative causal effect of IR on monounsaturated FA levels, we looked up gene expression data for SCD1 and its rodent equivalent scd1 in the EMBL-EBI Expression Atlas v3.0 (http://www.ebi.ac.uk/gxa/home). This gene encodes stearoyl-CoA desaturase 1 (SCD-1), the rate-limiting enzyme in the biosynthesis of OA and POA [26]. Among 252 uploaded experiments that reported significantly different expression (5% FDR) between experimental and control conditions, we extracted all experiments related to IR. There were eight studies in mice and rats–none in human beings (S1 Text). In all instances, the direction of differential scd1 expression was consistent with our findings—the IR-increasing condition down-regulated scd1 expression.
In the observational part of this multi-cohort study of blood metabolomics profiles, we identified bile acid, glycerophospholipid and caffeine metabolism as associated with IR, and FA biosynthesis as related to impaired insulin secretion. We discovered and replicated causal effects of IR on lower levels of the monounsaturated FAs POA and OA, as well as suggestive evidence for higher levels of the aromatic amino acid tyrosine. Sensitivity analyses did not indicate pleiotropic effects of the genetic instruments. Causal effects were largely unaffected by the exclusion of prevalent diabetes cases in the KORA/TwinsUK replication set. A small collection of publicly available experimental results in rodents supported our causal findings: All IR-increasing conditions were associated with a down-regulation of SCD-1, implying reduced endogenous production of OA and POA.
The liver and adipose tissue are the main sites of de novo lipogenesis and SCD-1 is the rate-limiting enzyme for monounsaturated FA biosynthesis [27]. It introduces double bonds into palmitic and stearic acid to produce POA (16:1n-7) and OA (18:1n-9), respectively–the major precursors for cholesteryl esters and triglycerides (TGs) that are packaged into very low-density lipoprotein (VLDL) particles and secreted by the liver [28]. In scd1 knockout mice and hypertriglyceridemic persons, plasma FA composition reflects hepatic SCD-1 activity [29, 30]. Hence, whilst dietary FAs contribute to overall plasma FA levels, the relative lipid composition, as assessed in the present study, is likely to reflect SCD-1 activity. Correspondingly, in ULSAM and PIVUS, we found good agreement between FA quantification by untargeted plasma metabolomics and targeted serum cholesteryl ester analysis (S1 Text).
Experimental evidence for the inhibition of SCD-1 by IR stems from liver-specific insulin receptor knockout (LIRKO) mice that had ~80% reduced hepatic scd1 expression and ~90% reduced microsomal scd1 transcript levels compared to control mice [31, 32]. In muscle-specific insulin receptor knockout (MIRKO) mice, scd1 expression was downregulated by ~23% compared to controls [33].
Little is known about the causal effects of a genetic predisposition for IR on SCD-1 [34]. Reduced SCD-1 activity in knockout mice has beneficial metabolic consequences, including reduced obesity [35] and improved IR [36] (reviewed in [28] and [34]). Yet, reduced SCD-1 activity has also been associated with adverse vascular outcomes: Inhibition of SCD-1 in hyperlipidemic mice markedly increased aortic atherosclerosis despite protective effects on obesity and IR [37]. Reduced SCD-1 activity caused enrichment of saturated FA in VLDL and LDL, which promoted atherogenesis through macrophage-induced vascular inflammation. Susceptibility to exacerbated inflammation was also demonstrated in scd1 knockout mice with induced colitis [38] or on a very low-fat diet that increased endoplasmic reticulum stress response [39]. Based on these competing effects on vascular and metabolic health, we speculate that IR reduces SCD-1 activity, which counteracts the metabolic consequences of IR by improving insulin signaling but concomitantly increases the risk for CVD through saturated FA-induced proinflammatory changes. Supported by our results and the above evidence, this hypothesis further derives from two facts: IR increases CVD risk independent of other risk factors [2, 40] and IR predicts CVD risk independent of T2D [41]. A summary of the presumed relationships is displayed in Fig 4. In our study, we could not evaluate the longitudinal effect on CVD, hence our hypothesis needs to be evaluated in cohorts with the available outcome data. In observational analysis, IR was positively associated with OA (FDR-adjusted p = 0.035, after adjustment for BMI pBMI > 0.05) and POA levels (p = 0.091, pBMI > 0.1) in contrast to the negative genetic associations in MR studies. This discrepancy is likely due to confounding factors such as dietary FA intake and highlights that MR studies have the power to disentangle causal mechanisms that may be obscured in cross-sectional studies. The hypothesized but untested relationship with cardiovascular disease requires investigation in future studies.
The effect of IR on elevated tyrosine levels was observed as a non-significant tendency in all cohorts but reached nominal significance only after combination of estimates in meta-analysis. Although there is a considerable risk that this result is a false positive due to multiple testing, we still consider it an interesting finding worth investigating in larger MR studies, as it has not been reported before. Associations between worse IR/T2D risk and circulating tyrosine levels have been established in observational and longitudinal studies [42, 43]. Possible molecular mechanisms could involve reduced tyrosine catabolism, either through IR-induced oxidative stress leading to elevations in methionine, cysteine and the antioxidant glutathione that result in tyrosine hydroxylase inhibition [44]; or through inactivation of tyrosine aminotransferase [45]. Hence, we speculate that the previously identified association between tyrosine and increased T2D risk could be caused by concomitant IR but caution that the observed trends in our study need to be verified in larger samples.
Limitations of our study include moderate power in MR studies and inherent limitations of the non-targeted metabolomics discovery platform to capture certain metabolites (e.g., polar amino acids) and separate sugars and other non-polar molecules. We used an untargeted metabolomics approach that detected several thousand metabolic features, yet the limited availability of standard compound spectra in in-house and public libraries precluded the quantification of the entire spectrum of the plasma metabolome and may have biased the pathway enrichment analysis. Observational associations were established in an exclusively male, elderly European cohort (ULSAM) with unknown generalizability to other age and ethnic groups. The different metabolomics platforms between studies and the expression of FA levels in % of plasma phospholipids rather than absolute levels in CHARGE somewhat limit methodological consistency. However, this diversity also supports the robustness of findings that were replicated on different platforms. Causal relationships were consistent across studies but require validation in physiologic models of IR, particularly as some sources of bias (e.g., canalization, unmeasured horizontal pleiotropy) cannot be excluded. Whilst the unique contribution of our study is to assess causal effects on the plasma metabolome, we could not examine the reverse causality as no genetic instruments were available for the majority of metabolites. As illustrated above, relative levels of circulating POA and OA likely reflect endogenous biosynthesis, however, the exact contributions of SCD-1, exogenous sources, catabolism and excretion could not be addressed by our study and require experimental investigation. We used proxy outcomes (HOMA-IR and CIR) with available large GWAS results in sensitivity analyses rather than the exact same measures as in the main analysis.
Strengths include the extensive testing for violations of MR assumptions, the replication of main findings in large independent cohorts that used different methods of quantification, and the use gold standard measures for IR. We detected no evidence of pleiotropy of genetic instruments whose associations with cardiometabolic traits were in the expected directions and causal estimates agreed between different analysis methods.
In summary, our study in multiple independent cohorts of community residents indicates a causal effect of IR on circulating OA, POA, and tyrosine levels and provides new insights into the metabolomic signature of IR and impaired insulin secretion. It is to our knowledge the first large-scale attempt to explore the causal effects of genetic IR on a non-selected set of plasma metabolites. The potential implications of the presumed IR-induced inhibition of monounsaturated FA biosynthesis on health outcomes require validation in experimental models but may form part of the explanation for the elevated CVD risk in IR that is independent of T2D development.
Detailed descriptions are available in S1 Text. In brief, ULSAM (http://www2.pubcare.uu.se/ULSAM/) was started in 1970 and enrolled 81.7% (n = 2,322) of all male residents of Uppsala county, Sweden, born between 1920 and 1924. On-going assessments every five to ten years include questionnaire, biochemical, and anthropometric examinations. The current study is based on the assessment at age 70 years, which included an OGTT and hyperinsulinemic-euglycemic clamp measurement.
The PIVUS study (http://www.medsci.uu.se/pivus/) enrolled 50% (n = 1,016) of a random sample of Uppsala community residents aged 70 years in 2001 and features assessments including health questionnaires, blood sampling and clinical measurements every five years. The current study is based on the assessment at age 70 years.
TwinGene (http://ki.se/en/meb/twingene-and-genomeeutwin) is a longitudinal study of 12,591 twins born before 1958 and registered in the Swedish Twin Registry with questionnaire assessments and blood sampling done between 1998 and 2002, and again between 2004 and 2008. The current study used a subcohort from a nested case-cohort design established for an earlier project [13] that was randomly selected within four age and sex strata to match a case group that included incident cases of T2D, coronary heart disease, ischemic stroke, and dementia (up to 31 Dec 2010) (S1 Text). In all three Swedish cohorts, prevalent cases of diabetes were excluded (criteria in S1 Text).
The Cooperative Health Research in the Region of Augsburg (Kooperative Gesundheitsforschung in der Region Augsburg, KORA [46]) study is a series of epidemiological surveys of the general population in Southern Germany that includes longitudinal health assessment and blood sample collection. The current study is based on the KORA F4 (2006–2008) survey of 32-to-77-year-old men and women (n = 1,768, 48.5% male, mean age 60.8 ± 8.8 years, mean BMI 28.2 ± 4.8 kg/m2).
TwinsUK is a predominantly female cohort of adult twins recruited from the general UK population. The current study is based on 17-to-85-year-old twins (n = 6,056, 7.1% male, mean age 53.4 ± 14.0 years, mean BMI 26.1 ± 4.9 kg/m2) who underwent blood metabolite profiling and health center assessment.
The CHARGE consortium GWAS for plasma FA fractions [20] combined 8,961 mostly middle-aged to older persons of European ancestry (45.0% male, mean age 59.7 ± 7.6 years, mean BMI 27.0 ± 4.8 kg/m2) from five cohorts—the Atherosclerosis Risk in Communities (ARIC) study, the Cardiovascular Health Study (CHS), the Coronary Artery Risk Development in Young Adults (CARDIA) study, the Invecchiare in Chianti (InCHIANTI) study, and the Multi-Ethnic Study of Atherosclerosis (MESA). Details on cohorts and recruitment are reported elsewhere [20] and documented online (http://chargeconsortium.com/).
The Finnish consortium [21] combined five cohorts of 8,330 individuals with serum nuclear magnetic resonance metabolomics results from the FINRISK 2007 Dietary, Lifestyle and Genetic determinants of Obesity and Metabolic syndrome (FINRISK-07/DILGOM) study, the Helsinki Birth Cohort Study (HBCS), the Health2000 GenMets study, the Northern Finland Birth Cohort 1966 (NFBC1966) study and the Cardiovascular Risk in Young Finns Study (YF). Across all cohorts with 46.9% males, mean age was 37.8 ± 3.2 years and mean was BMI 25.4 ± 4.3 kg/m2.
All participants provided written informed consent prior to inclusion in the study and the research was approved by the Ethics Committees of Uppsala University (ULSAM, PIVUS) and Karolinska Institutet (TwinGene), or the respective Institutional Review Boards for the other cohorts. The study was conducted according to the principles of the Declaration of Helsinki.
Untargeted metabolomics profiling of venous blood samples in the three Swedish cohorts was carried out by ultra-performance liquid chromatography (UPLC) on a Waters Acquity UPLC system coupled to a Waters Xevo G2-Time-Of-Flight-Mass Spectrometry (TOFMS) platform at Colorado State University (Fort Collins, CO, USA). Data acquisition in the positive electrospray ion mode with a mass-to-charge ratio (m/z) range of 50–1,200 at 5 Hz was alternately performed at collision energies of 6V and 15–30V. Details on sample handling and data processing by XCMS in R [47] are available in S1 Text and in [14]. Parameter selection for feature detection, alignment, grouping, and imputation was optimized in simulations of random sets of 20–40 samples. In total, 10,162 (ULSAM), 9,755 (TwinGene) and 7,522 (PIVUS) features were detected. Adjustment for factors of unwanted variability (plate effect, analysis date, retention time drift and sample collection) by analysis of variance-type standardization was followed by log-transformation and removal of spectra with abnormal intensities and/or low inter-duplicate correlations and/or retention times <35 sec. For each feature, retention time, m/z, and fragmentation pattern were compared to in-house and public database reference libraries and matched according to Metabolomics Standard Initiative guidelines [48]. The current study is based on all 192 metabolites identified in ULSAM. Common features between ULSAM, PIVUS, and TwinGene were identified by matching m/z and retention time, followed by manual inspection of fragmentation spectra. Full metabolomics data are available in the MetaboLights archive (study identifiers MTBLS90 for PIVUS, MTBLS124 for ULSAM, MTBLS93 for TwinGene; http://www.ebi.ac.uk/metabolights/).
Metabolomics analyses in KORA and TwinsUK were carried out by the commercial company Metabolon, Inc. (Durham, NC, USA), which combined positive and negative ion-mode UPLC/tandem-MS with gas chromatography/MS. Following protein precipitation in methanol, samples for analyzed in duplicates and spectral annotation was performed against a standard compound library (for details, see S1 Text and [19, 46]). Following the removal of outlying (>3 SD) features and those with <300 non-missing values, 276 and 258 metabolites in KORA and TwinsUK, respectively, were quantified and identified based on matching to in-house library standards.
In CHARGE, a targeted gas chromatography approach was used to quantify plasma phospholipid composition (except in the InCHIANTI cohort, where total plasma FA were measured). As detailed in [20], fasting plasma phospholipid isolation by thin-layer chromatography was followed by quantification of FA by targeted gas chromatography.
The Finnish study processed serum samples from all five sub-cohorts in one central laboratory by three complementary 1H-nuclear magnetic resonance analysis windows optimized for lipoproteins, low molecular weight metabolites and lipid species, respectively [21, 49]. For quantitative analysis, raw spectral data were pre-processed with baseline zeroing, peak alignment and correction for albumin background and validated against high-performance LC data.
We used IV analysis to estimate the causal effect of IR/impaired insulin secretion on metabolite levels in PIVUS and TwinGene. The Wald ratio estimator [50] was obtained as the ratio between the regression coefficients for the effect of the genetic IV on metabolite levels divided by the effect of the genetic IV on IR/insulin secretion (Eq 1). Standard errors were estimated by the delta method, which we previously validated (Eq 2) [51].
As IV, we used the IR and impaired insulin secretion genetic risk scores validated by Scott et al. [18] (S2 Table). Because the study included ULSAM (alongside the MRC-Ely, RISC, Fenland, and EPIC-Interact cohorts) to estimate the association between instrument and risk factor, we excluded the ULSAM cohort from MR analysis. We used proxy SNPs (linkage disequilibium r2 >0.8) for variants not directly genotyped. When no proxies were available, SNP scores were imputed as 2*risk allele frequency based on the 1000 Genomes Project phase 1 [52] (for the insulin secretion score this applied to rs1800574 in all three cohorts, as well as rs7957197 and rs10811661 in TwinGene) (S2 Table). Quality control included mean-imputation of SNP values for individuals with one missing value and exclusion of individuals with >1 missing values. All SNPs had call rates >95%. Additive non-weighted genetic risk scores were calculated in PLINK1.07 (http://pngu.mgh.harvard.edu/~purcell/plink/). The association between genetic IVs and metabolites was estimated in linear models adjusted for age, sex, the first three genetic principal components and cohort with metabolite levels (SD-unit) as outcome. Age- and sex-adjusted associations between genetic IVs and exposure (SD-unit) were obtained from Scott et al. [18]. For sensitivity analysis, SNP associations with IR and insulin secretion were obtained from summary GWAS results in MAGIC (see above). MR Egger regression and other sensitivity analyses were carried out in R according to the scripts provided in the data supplement by [23] and in appendix A.3 by [24], respectively.
We obtained the publicly available GWAS data for serum/plasma metabolite levels in 7,824 European adults in the KORA F4 and TwinsUK cohorts [19], as well as for OA and POA fractions of total plasma phospholipids in 8,961 Europeans in CHARGE http://chargeconsortium.com/). We also obtained the summary meta-GWAS statistics for serum tyrosine levels from a Finnish consortium study [21]. Details on genotyping and statistical analyses are available in S1 Text and elsewhere [19, 46]. We extracted β coefficients and standard errors for SNPs in the non-weighted IR/insulin secretion genetic scores and computed summary effect sizes with the grs.summary() function in the gtx package in R [53]. Regression coefficients expressed SD-unit change in metabolite levels. Causal effects were estimated by MR analysis as described above for all metabolites available in KORA/TwinsUK that passed p < 0.1 at the discovery stage, for OA and POA in CHARGE, and for tyrosine in the Finnish consortium. Causal estimate from all cohorts were combined in inverse variance-weighted, fixed effects meta-analysis via metafor in R.
R scripts for the full metabolomics pipeline in PIVUS, TwinGene and ULSAM are available online (https://github.com/andgan/metabolomics_pipeline), R scripts used for observational and MR analysis are available as well (https://github.com/chrnowak/metabolomics).
|
10.1371/journal.ppat.1003197 | Protein Complexes and Proteolytic Activation of the Cell Wall Hydrolase RipA Regulate Septal Resolution in Mycobacteria | Peptidoglycan hydrolases are a double-edged sword. They are required for normal cell division, but when dysregulated can become autolysins lethal to bacteria. How bacteria ensure that peptidoglycan hydrolases function only in the correct spatial and temporal context remains largely unknown. Here, we demonstrate that dysregulation converts the essential mycobacterial peptidoglycan hydrolase RipA to an autolysin that compromises cellular structural integrity. We find that mycobacteria control RipA activity through two interconnected levels of regulation in vivo—protein interactions coordinate PG hydrolysis, while proteolysis is necessary for RipA enzymatic activity. Dysregulation of RipA protein complexes by treatment with a peptidoglycan synthase inhibitor leads to excessive RipA activity and impairment of correct morphology. Furthermore, expression of a RipA dominant negative mutant or of differentially processed RipA homologues reveals that RipA is produced as a zymogen, requiring proteolytic processing for activity. The amount of RipA processing differs between fast-growing and slow-growing mycobacteria and correlates with the requirement for peptidoglycan hydrolase activity in these species. Together, the complex picture of RipA regulation is a part of a growing paradigm for careful control of cell wall hydrolysis by bacteria during growth, and may represent a novel target for chemotherapy development.
| Peptidoglycan (PG) is a major component of the bacterial cell wall, which forms a flexible, but strong mesh around the cell to oppose osmotic pressure and prevent lysis. PG is also dynamically modified, continually being disassembled and polymerized as the cell elongates and divides. It remains poorly understood how cells can titrate enough hydrolysis of the PG to allow bacterial growth without leading to excessive digestion and disruption of cellular integrity. In our work, we have identified two methods by which a critical PG hydrolase, RipA, is carefully controlled in Mycobacterium tuberculosis—protein interactions help prevent lethal RipA dysregulation, while proteolytic cleavage is used as a second step to activate the enzyme in order to separate daughter cells. Our work elaborates multiple post-transcriptional mechanisms for preventing PG hydrolases from becoming lethal autolysins. These different levels of regulation may serve as a more general paradigm for PG remodeling in other bacterial species.
| Mycobacterium tuberculosis is the causative agent of tuberculosis and accounts for up to 10 million symptomatic infections a year [1]. The spread of multi-, extensively- and now totally- drug resistant strains [2] has created a pressing need to understand essential mycobacterial processes in an effort to define novel targets for chemotherapy. One highly essential bacterial process is peptidoglycan (PG) synthesis and remodeling, which is critical for providing structural integrity in nearly all bacteria. PG forms a continuous macromolecular mesh that is part of the bacterial cell wall and is required for correct cellular morphology and opposition to osmotic forces. Despite extensive biochemical and genetic characterization of the enzymes responsible for the synthesis and degradation of PG (reviewed in [3], [4]), the mechanism by which these enzymes coordinate their activities remains poorly defined. It is clear, however, that dysregulation of this homeostatic balance frequently has lethal effects on the bacterium—inactivation of peptidoglycan synthases, either through the use of penicillin derivatives or overexpression of dominant negative forms of PG synthetic enzymes, induces lysis of cells [5], [6]. In many cases, this lethality can be suppressed by inactivation of several peptidoglycan hydrolases [5], [7], suggesting that PG hydrolase autolysin activity is restrained by functional interactions with PG synthases. This idea is consistent with a ‘make-then-break’ approach to cell wall synthesis where new PG subunits are first incorporated before the existing sacculus is cleaved to allow expansion [8]. One example of this is the formation of the septal PG—cells ensure that the septal PG is formed before PG hydrolases cleave apart the daughter cells.
Recent work suggests that the activity of PG synthetic and hydrolytic enzymes is regulated by the formation of protein complexes. In E. coli, the PG amidases AmiA, AmiB and AmiC can interact with non-enzymatic partners that upregulate septal peptidoglycan hydrolysis [9]. Conversely, the major bifunctional PG synthases, PBP1A and PBP1B in E. coli interact with and rely on essential lipoprotein partners for function [10]. In addition to interactions with non-enzymatic partners, several affinity chromatography and genetic studies have identified interactions between PG modulating enzymes themselves [11]. While the exact interactions may be species-specific, in general, PG synthases can associate with both other PG synthases and with PG hydrolases. Likewise, PG hydrolases can form predicted hydrolytic complexes with other autolysins [11]–[13]. These results suggest a general paradigm where PG modulating enzymes of both similar and opposing functions assemble as multi-protein complexes that spatially and temporally coordinate PG synthesis and hydrolysis during bacterial growth and division. An immediate challenge is to translate the many identified interactions into functional in vivo effects on the growth and division of bacteria.
Previously, we have studied regulation of the essential M. tuberculosis PG hydrolase, RipA (Rv1477). RipA belongs to the NLPC/p60 family, and has been characterized as a D,L D-glutamate-diaminopimelic acid (DAP) endopeptidase that cleaves within the pentapeptide bridges of the PG sacculus, thereby removing cell wall crosslinks [14]. The RipA homologue in Listeria (P60) and in Mycobacterium marinum (IipA) can be deleted, but this causes septal resolution defects [15], [16]. In contrast, RipA is essential in M. tuberculosis [17], and depletion of RipA produces a chaining phenotype in M. smegmatis, which causes severe growth inhibition [18]. This is unlike the case in E. coli, where extensive chaining and growth inhibition requires inactivation of several PG hydrolases [19].
In this work, we interrogate the mechanism by which RipA activity is regulated in vivo during vegetative growth. We report that RipA requires careful control to support growth and division without compromising the cell's structural integrity—RipA becomes a lethal autolysin when its activity is dysregulated. Under physiological conditions, RipA relies on protein interactions to correctly control its degradative capacity. These interactions are also necessary for proteolytic cleavage of RipA to produce active enzyme. RipA cleavage and activation is more robust in M. smegmatis than in the pathogenic M. tuberculosis or M. bovis BCG, which may be a reflection of the different PG hydrolysis requirements between fast and slow growing mycobacteria. However, bypassing RipA cleavage by overexpressing fully active truncated enzyme compromises the structural integrity of both M. smegmatis and M. tuberculosis, suggesting that RipA cleavage may be rate-limited in order to synchronize PG hydrolysis with the growth rate of the bacterium. These results suggest a model in which RipA is regulated by several interconnected post-transcriptional mechanisms—proteolytic processing produces active enzyme, while protein-protein interactions upstream and downstream of cleavage ensure RipA functions correctly at the septum.
When RipA is depleted, daughter cells are unable to separate and instead, grow as chains (Figure 1A). While cells require peptidoglycan hydrolysis to accomplish cell separation, excessive cell wall degradation can compromise structural integrity and lead to lysis. We hypothesized that RipA sits in this precarious situation, where the cell cannot tolerate either too little or too much RipA activity.
We investigated whether excessive RipA activity is toxic to mycobacteria by inducing M. smegmatis RipA (RipASm) from a tetracycline-inducible episomal plasmid in M. smegmatis. Unlike the chaining phenotype we have previously observed with RipA depletion, RipA overexpression caused the rod-shaped cells to become spherical and lyse, (Figure 1A, 1D (time-lapse movie in Video S1). This is dependent on catalytic activity, as overexpression of a catalytic mutant, RipASm C408A, does not display this phenotype (Figure 1A, 1D, (time-lapse movie in Video S2). The spherical phenotype of RipASm overexpression led to a severe growth defect by optical density (Figure 1B) and over one hundred fold killing, as determined by CFU enumeration (Figure 1C). Thus, excessive RipA activity in the cell is highly lethal.
To determine whether a more physiological level of RipA could be converted to a lethal autolysin, we dysregulated RipA activity through the use of the beta-lactam antibiotic meropenem. Beta-lactam antibiotics block PG precursor incorporation, which causes excessive PG hydrolase activity and cell lysis [7]. While M. tuberculosis is relatively resistant to most beta-lactams, recent work has shown that meropenem is more resistant to the endogenous mycobaterial beta-lactamase, and is highly effective at killing M. tuberculosis, especially in combination with the beta-lactamase inhibitor clavulanate [20]. Meropenem targets PBP2 and PBP3 in E. coli, as well as L,D transpeptidases in M. tuberculosis [21], [22]. Since RipA is known to interact with the PG synthase PBP1, which is required for normal vegetative growth [23] and morphology in mycobacteria [24] (depletion of the protein leads to rounded cells), we asked whether meropenem treatment can dysregulate RipA and convert the enzyme into a lethal autolysin.
We first treated M. smegmatis with 10 µg/mL meropenem and assessed morphological changes over time by microscopy. Treated cells filament and swell at the poles and septa, which are the sites of mycobacterial PG incorporation (Figure 2A, arrows). This morphological toxicity correlated with a decrease in optical density over time, which suggested lysis (Figure 2B). This was borne out by CFU analysis, which showed that 80% of treated cells were killed within 6 hours of meropenem treatment (Figure 2C, bar 2).
The bulging at sites of PG incorporation after meropenem treatment suggested an excess of PG hydrolase activity. Since RipA localizes both to poles and septa, we hypothesized that RipA may play a role in killing meropenem treated cells. To test this idea, we depleted M. smegmatis of RipA before meropenem treatment (Figure S1A) and then assessed survival with meropenem treatment by CFU enumeration. We found that unlike RipA replete cells (Figure 2C, bar 2), meropenem did not kill RipA depleted cells (Figure 2C, bar 4). Furthermore, while RipA replete cells bulged under meropenem treatment as expected (Figure 2D, arrows), the RipA depleted cells appeared refractory to swelling (Figure 2D). As a control, cells were treated with SDS (which causes non-specific cell wall and membrane stress), as well as streptomycin, which targets protein synthesis. These control cells showed no survival (Figure S1B) or morphological (Figure S2) differences between RipA replete and depleted cells, demonstrating RipA specifically interacts with meropenem-affected pathways. Given that RipA enzymatic activity is modulated through protein-protein interactions with different PG synthetic and hydrolytic partners [18], [24], the meropenem data suggest that RipA is held in check in complexes by an interacting protein, such as PBP1. Furthermore, although there are many hydrolases that could contribute to cell death when PG synthesis is blocked, we found that at least for the synthases blocked by the clinically relevant beta-lactam antibiotic meropenem, RipA is quantitatively the single most important hydrolase.
Our results demonstrate that RipA dysregulation is highly detrimental to the cell. Thus, mycobacteria must control the activity of RipA during growth—there must be enough PG hydrolase activity around to support growth and division, but not an excessive amount so as to compromise structural integrity. One way this control may be regulated is at the transcriptional level.
We assessed whether the cell downregulates RipA expression using quantitative PCR. Since RipA is required for septation, which does not occur in non-replicative conditions, we compared RipA expression between exponential and stationary phases. We found that RipA remained expressed from exponential phase through the transition into stationary phase (Figure S3), suggesting there may be post-transcriptional mechanisms responsible for restraining RipA activity when it is not needed. Thus, we investigated whether achieving tight control of RipA activity may rely on post-transcriptional processes.
We hypothesized that removal of wildtype RipA from its endogenous niche in vivo would inhibit correct septal resolution and therefore phenocopy RipA depletion. If RipA requires downstream interactions for activity, e.g. with members of septal complexes or with post-translational enzymatic regulatory proteins, then we should be able to create a dominant negative RipA mutant, where the critical catalytic cysteine [25] is mutated to a nonfunctional alanine. Overexpression of the RipASm C408A catalytic mutant should result in competition between nonfunctional RipA and endogenous RipA for required post-translational activation processes. If RipA requires these interactions to function, then we would observe chaining. Indeed, when we induced RipASm C408A, cells grew as short chains (Figure 1A, white arrows), suggesting that RipA interactions are necessary for correct septal PG hydrolysis.
While overexpression of the RipASm C408A mutant produced a severe growth defect (Figure 1B, 1C), it was not accompanied by the widespread lysis that was observed upon wildtype RipASm overexpression (Figure 1C). The apparent drop in optical density upon longer induction of the RipASm C408A strain (Figure 1B) was due to clumping of the culture, which though affecting optical density, did not lead to a drop in CFU, indicating growth inhibition rather than lysis (Figure 1C). However, we did observe occasional lysis of the RipASm C408A strain in addition to the dominant negative chaining phenotype. Some cells within a chain produced a slight bulging phenotype, which is indicative of an increase in PG hydrolytic activity (Figure 1A, red arrows). These bulging cells, like in RipA dysregulated cells, can go on to lyse (Figure 1D, Video S2), though this does not lead to detectable cell death by CFU enumeration (Figure 1C). It is possible that displacement of endogenous wildtype RipA from complexes at the septum leads to activity at ectopic sites in a subset of cells. Alternatively, RipA overexpression could stimulate other endogenous PG hydrolases.
When we used a RipA polyclonal antibody that recognizes a C-terminal epitope, we observed truncated RipA species from mycobacteria by Western blotting. When we overexpressed RipASm, we found several bands smaller than the predicted full length protein (Figure 3A, lane 3). Likewise, we saw these truncated bands when we overexpressed RipASm C408A in M. smegmatis (Figure 3A, brackets). These products were not due to non-specific cytoplasmic degradation of overexpressed RipA, as we fractionated RipASm C408A overexpressing cells and found that RipA processed species were enriched in the cell wall fraction (Figure S4A). The efficiency of fractionating mycobacteria was confirmed by Western blotting against RpoB (cytosolic) and mycobacterial antigen 85 (cell wall) markers (Figures S4B, S4C). Thus, these results suggest that RipA undergoes physiological post-translational processing in the periplasmic or cell wall compartment.
To further demonstrate that RipA processing is physiological and not an artifact of overexpression, we used Western blotting to estimate the size of RipA in wildtype M. smegmatis whole cell lysates. In mid-exponential phase cells, RipA formed a smear of ∼30 kDa (Figure 3A, lane 2, red arrow) with no detectable full length protein present. This signal was specific for RipA, as cells depleted for RipA (Figure 3A, lane 1) had decreased signal compared to wildtype cells (Figure 3A, lane 2) when equal amounts of total protein were analyzed (Figure 3B). Furthermore, processed endogenous RipA partitioned to the cell wall compartment (Figure S4A). The smear of processed RipA suggests there are multiple processing sites. This may represent multiple cleavage products or further modifications to the protein. A recent crystal structure of RipA suggests a protease labile loop exists between the N inhibitory and C terminal PG hydrolase domains; this loop is hypothesized to be the site of cleavage in vitro, which is required for RipA enzymatic activation [25]. We mutated candidate residues in the loop in an attempt to identify cleavage sites but were unsuccessful in blocking RipA processing (Figure S5).
Truncated RipA species were found associated with the cell wall compartment and in culture filtrates. In the culture filtrate, a RipA fragment appeared as a single band at approximately 25 kDa (Figure 3C, asterisk). We demonstrated this signal was specific by C-terminally tagging endogenous RipA on the chromosome of M. smegmatis with a FLAG epitope. This 25 kDa species exhibited altered mobility due to the epitope and could be detected by both anti-RipA (Figure 3C, left panel) and anti-FLAG antibodies (Figure 3C, right panels). These results indicate that RipA exists physiologically in a smaller form than the predicted full length protein.
The observation of RipA cleavage suggested this process could be required for the protein's function in vivo. To test whether RipA processing is correlated with division, we titrated overexpression of RipASm C408A and quantified the amount of induction needed to mediate chaining. We found that low level overexpression with 30 ng/mL inducer was sufficient to cause chaining (Figure 4A) without saturating the processing machinery, since these cells did not accumulate full length RipA (Figure 4B). Instead, mildly overexpressed RipA was processed down to two sets of smaller species at around 23 kDa and 12 kDa (Figure 4B). We also saw a dose-dependent saturation of the endogenous processing capacity, with high induction leading to accumulation of full length RipA (Figure 4B), as well as loss of processed endogenous RipA (Figure S6A). When we quantified recombinant protein levels by comparing densitometry with endogenous RipA protein (Figure 4C), we found that even mild RipASm C408A overexpression at 30 ng/mL of inducer (processed recombinant protein is approximately 10% of endogenous wild type RipA levels) was sufficient to cause chaining and cellular toxicity (Figure 4A). Together with qPCR data showing that RipASm C408A induction does not affect endogenous RipA transcription (Figure S6B), these data suggest direct competition between the RipAsm C408A mutant and endogenous RipA for processing machinery, and that the processed, not full length, form of RipA is required for division.
However, though correlated with function, the processed species we observed could be the product of an inactivating event. To test this, we took advantage of the observation that the M. tuberculosis homologue of RipA (RipATB) functions differently in M. smegmatis than its native counterpart, despite having the same general domain architecture (Figure S7). In contrast to RipASm, which is toxic when overexpressed, overexpression of RipATB in M. smegmatis, surprisingly caused no toxicity or cell morphological differences (Figure 5A,B), despite similar protein levels (Figure 3A). We examined whether RipASm toxicity was correlated with its processing by performing Western blot analysis on RipATB overexpressing M. smegmatis. In contrast to overexpression of RipASm, when wildtype RipATB is overexpressed we observed only a single full length band at 55 kDa (Figure 3A, right arrow). The absence of processing correlates with the lack of detectable RipATB toxicity in M. smegmatis. Thus, we hypothesize that proteolytic cleavage is required for activating RipA in vivo.
However, it could be formally possible that RipATB cannot recognize M. smegmatis peptidoglycan or is not intrinsically active enough in M. smegmatis to cause morphological defects. To test if RipATB can be enzymatically functional in M. smegmatis, we deleted the predicted N-terminal inhibitory segment by fusing the truncated active domain of RipATB to the RipA secretion signal peptide (RipATB-AD). As a control, we also produced a construct in which the M. smegmatis RipA active domain (RipASm-AD) can be secreted. None of the strains produced growth defects when uninduced (Figure S8). As expected, RipASm-AD like full length RipASm, was fully functional when induced and disrupted cell wall integrity, leading to bulging of the cells and a concomitant growth defect (Figure 6A,B). When RipATB-AD was secreted, we found that it was also functional and behaved in the same way as RipASm-AD (Figure 6A,B). Thus, the catalytic domain of RipATB can be active in M. smegmatis, but full length RipATB is not toxic because it does not undergo efficient processing in M. smegmatis.
Given the potentially toxic nature of hyperactive RipA we hypothesized that RipA processing and activation may be less robust in slow-growing mycobacteria in order to match their much slower rate of growth and consequent lower requirement for peptidoglycan hydrolysis. To investigate this model, we first determined whether RipA is processed in M. tuberculosis by overexpressing RipATB. By Western blot analysis, we found multiple immunoreactive smaller species of RipATB, suggesting processing in M. tuberculosis (Figure 7D, brackets). However, the induction of RipATB in M. tuberculosis did not produce morphological changes or growth defects, even after five days of induction (Figure 7A,B). This overexpression produced about 3 fold more protein (most of which is in the processed form) than endogenous full length RipA (Figure 7E), which is similar to the amount of overexpression needed to observe cell chaining in M. smegmatis with the RipASm C408A allele (Figure 4C). The lack of morphological changes in M. tuberculosis is also in contrast with the marked lethality of RipASm overexpression in M. smegmatis (Figure 1).
One explanation for this dichotomy may be that the slow growth of M. tuberculosis might mask morphological or growth defects caused by RipA overexpression. To test this possibility, we bypassed RipA processing by secreting truncated RipATB-AD in M. tuberculosis. Induction of RipATB-AD produced a severe growth defect in M. tuberculosis (Figure 7C), and concomitant cell rounding (Figure 7A) similar to that seen when RipASm was overexpressed in M. smegmatis. These results show that unchecked RipA activity is toxic even to slow-growing mycobacteria. Since full length RipATB induction does not produce this toxicity, M. tuberculosis may have intrinsically less robust RipA processing than in M. smegmatis. Indeed, in M. tuberculosis, we observed a band of 55 kDa on Western blots probed with anti-RipA antisera. This band is the same size as full length RipATB and appears in both uninduced and induced samples (Figure 7D, arrow). As this form was not detected in wildtype M. smegmatis lysates (Figure 3A), it may represent endogenous, unprocessed full length RipA. The same full length band was also observed in M. bovis BCG cell lysates (Figure S9). These data, along with the active domain overexpression analysis, support the idea that slow-growing mycobacteria process RipA less efficiently in order to keep this potentially lethal activity in check.
Bacteria rely on peptidoglycan (PG) for shape and structure. The prevailing view of PG remodeling requires the concerted action of synthetic enzymes ligating new subunits into the existing PG lattice followed by hydrolysis of the PG sacculus by autolysins to allow cellular expansion or division. This process is accomplished through the action of large holoenzyme complexes in the periplasm consisting of both PG synthetic and hydrolytic enzymes. Disruption of PG synthesis in these complexes can dysregulate cognate PG hydrolases, which can then become autolysins that lyse the cell [26]. Thus, the coordination and regulation of PG hydrolases is a critical process for the survival of the bacterium.
Here we find that RipA in M. tuberculosis and M. smegmatis can behave as an autolysin, resulting in the formation of spherical cells and lysis when overexpressed or dysregulated. Overexpression of a RipA dominant negative mutant not only causes loss of septal resolution and chaining but also leads to uncontrolled activity of endogenous PG hydrolases and lysis in a subset of cells. Thus, RipA requires downstream interactions to govern its correct function during septal resolution, as well as prevent lethal ectopic hydrolase activity. The relatively low amount of dominant negative RipA (about 10% of endogenous RipA) required for chaining suggests that the cell has finely tuned the amount of active RipA in the cell to near the level required for division; even loss of 10% of these active RipA species (which is manifest in a partial loss of endogenous RipA processing (Figure S6A) leads to a block in septal resolution. While it is clear that RipASm C408A overexpression can interfere with endogenous RipA activation, given its known interactions with two other PG remodeling enzymes that localize to the septum—RpfB and PBP1 [18], [24], [27]—it is possible the dominant negative mutant also incorporates into and inhibits functional PG remodeling complexes. A combination of these two activities may contribute to the RipASm C408A mutant's potency at inducing chaining at relatively low levels of induction.
Supporting the presence of regulatory RipA septal complexes, we showed chemical inhibition of peptidoglycan incorporating PBPs (of which the RipA binding partner PBP1 is a member) results in cell rounding and lysis. Loss of PG synthetic activity within a PG remodeling complex may allow cognate PG hydrolases (such as RipA) to become hyperactive and lyse the cell. We found that RipA depleted cells were specifically protected against meropenem-induced killing, but remained sensitive to other unrelated stresses. The depletion of the RipA likely affects the expression of RipB, which resides downstream in the same operon and has the same in vitro enzymatic specificity as RipA [14]. However, RipB is not essential for growth [28], and we have previously shown that RipA appears to be more phenotypically active than RipB in vivo. RipA, but not RipB, can complement the growth inhibition and cell chaining defects observed in the ripAB depletion strain [18]. While we cannot discount the possibility that RipB contributes to meropenem-mediated killing, it seems more likely that RipA is the main enzyme responsible for this lethal phenotype. Together, our data suggests that meropenem-induced killing is RipA dependent. However, we do observe a slight but significant difference in growth between RipA depleted cells in the presence and absence of meropenem (Figure 2C, lanes 2 and 4) that suggests there may also be some RipA independent growth inhibition (but not lysis) due to meropenem treatment. This may reflect the fact that meropenem can target several transpeptidases [21], [22]. Despite this, since RipA appears to mediate meropenem's bactericidal capacity, and thus appears to be a more attractive target for drug development, we would expect that a chemical activator of RipA might act synergistically with meropenem treatment. From these data alone, it may be possible that a RipA inhibitor would be contraindicated in combination with meropenem, as it would antagonize the effect of PBP blockade, but we have previously observed that RipA depletion can sensitize cells to carbenicllin, a β-lactam antibiotic that also targets various transpeptidases [18]. In previous assays, in contrast to meropenem, carbenicillin sensitization required long term RipA depletion—it had no bactericidal effect on cells depleted for RipA in the same time scale as our meropenem studies (Figure S1B). These data suggest that extended treatment with a RipA inhibitor may weaken cells enough to cause sensitivity to PBP inhibitors to which the cell was previously resistant. It would be interesting to determine whether RipA blockade can, in fact, synergize with existing PG targeting antibiotics in vivo.
Because of the threat of lethal autolysin activity, cells can control PG hydrolases through several, interconnected regulatory mechanisms. RipA is no exception, as we have found that in addition to protein interactions that modulate its function, RipA requires proteolytic activation. RipA exists primarily as smaller processed forms in M. smegmatis. Recent work with RipA in vitro has mapped a protease labile loop between a putative N terminal blocking domain and the C terminal p60 PG hydrolase domain [25]. The size of our truncated RipA species could contain the predicted size of the p60 domain itself after cleavage within this loop, but we were unable to determine the exact cleavage site(s) for RipA proteolytic activation in vivo using site directed mutagenesis—mutation of two pairs of highly scissile aspartate-proline peptide bonds [29] at DP301 and DP315 to alanines had no effect on the ability of RipA to be cleaved in M. smegmatis (Figure S5). This is consistent with the activation of Auto amidase in Listeria monocytogenes, which is also produced as a zymogen and becomes active only after proteolytic processing and removal of an N terminal inhibitory domain [30]. For Auto it was not possible to isolate single amino acid substitutions that abolish processing; instead, only deletion of the loop prevented proteolytic cleavage [30], which suggests that the activation loop is intrinsically labile and might be cleaved by many different proteases. Like Auto amidase, RipA's labile loop can be cleaved by many proteases in vitro [25], and thus may be a target of several proteases in vivo. This may explain why we see a smear of RipA truncated species in wildtype mycobacteria, as opposed to a single truncated band.
Given the work of Ruggiero et al [25], it was likely that RipA is produced as a zymogen in vivo, like Auto amindase. However, another recent report suggested a different effect of the N terminal domain in blocking RipA enzymatic activity [14], [25]. While both studies agreed that the N terminal domain appears to partially block the C terminal endopeptidase active site, the authors reached opposite conclusions as to whether the N terminus is inhibitory. Ruggiero et al [25] found that truncated RipA containing only the C terminal p60 domain was able to cleave purified PG, while full length RipA had minimal activity [25]. In contrast, Böth et al [14] showed that full length RipA was capable of degrading small synthetic PG fragments, and truncation of the N terminus produced no increase in enzymatic activity. However, in the latter work, the authors did not perform enzymatic digests using full length RipA on purified PG, as performed by Ruggiero et al. It is possible that the reported differences between these studies reflects the ability of small PG fragments to enter the RipA active site, despite partial occlusion by the N terminal domain, while access of larger substrates such as crosslinked and polymerized PG is blocked. Our results favor the zymogen model, as we have found that processing of the N terminal domain is required for full RipA enzymatic activation in vivo. Likewise, the lack of processing of RipATB in M. smegmatis likely accounts for the absence of its toxicity upon overexpression. While it is possible that full length RipA could serve some degradation function on smaller substrates in vivo, our results suggest that its main peptidoglycan remodeling activity requires removal of the N terminus, which contains a functional inhibitory domain.
Furthermore, using the less efficiently processed RipATB homologue, we showed that protein interactions are not only necessary for regulating functional septal complexes but also promote RipA proteolytic activation. Full length RipASm is toxic when overexpressed in M. smegmatis, but full length RipATB does not produce the same phenotype. However, when we bypassed processing and expressed the truncated RipATB active domain in M. smegmatis, we observed a full gain of toxicity. These results suggest that the interactions between RipASm and the cellular factors necessary for processing do not occur with the RipATB homologue. Since septation is a highly conserved process, these data may not necessarily indicate different RipA binding partners in slow and fast growing mycobacteria but rather that the M. smegmatis and M. tuberculosis binding partners have evolved together and may have higher affinities for one another.
Together, our work demonstrates that RipA regulation occurs at multiple levels post-transcriptionally. We did not see any transcriptional downregulation in cells transitioning into non-replicating conditions (Figure S4). In fact, there was a significant increase in RipA transcription during the transition to stationary phase, but the functional consequence of this observation remains unknown. Furthermore, while overexpression of the dominant negative RipASm C408A allele modulates processing of endogenous RipA (Fig S6A), this is due to competition for processing and not transcriptional feedback, even at high induction conditions (Figure S6B). This lack of transcriptional modulation is consistent with the observation that ripA expression has only limited variation across dozens of published experimental conditions (summarized on TBDB [31]), including general and antibiotic stresses. The only conditions under which ripA expression has been found to change are under non-replication conditions and, recently, when cells are blocked in cell division [32]. In the latter work, Plocinska et al found that ripA can be regulated by the MtrAB two-component system. Specifically, inhibiting septum formation prevents MtrB, which localizes at the septum, from activating the MtrA response regulator, leading to ripA downregulation. The authors proposed an interesting model in which ripA transcription could be upregulated by MtrB when it assembles at the division site; however, it remains unclear whether MtrAB regulation of ripA transcription occurs during normal growth or, instead, represents a stress response when cell division is inhibited. While the question of ripA transcriptional regulation during vegetative growth remains to be tested, our work suggests that post-translational mechanisms like processing may represent a key way of controlling RipA hydrolytic activity during growth. Therefore, we propose that protein-protein interactions help establish RipA function at the septum, where it is then aided in becoming proteolytically cleaved in the periplasm (Figure 8). After enzymatic activation, functional RipA can rely on both upstream and downstream protein interactions to help place it in the correct context during cell division—inhibition of cognate PG synthases can lead to dysregulated cell wall hydrolysis. The benefit of having multiple levels of RipA regulation is that the cell can exert a tighter control over RipA's activation and potential autolysin activity.
We found that RipA in M. tuberculosis is subject to less proteolytic activation than in M. smegmatis. The slow-growing mycobacteria like M. tuberculosis and M. bovis BCG might well have slower rates of PG hydrolysis and, consequently, reduced RipA activity. Indeed, we find significant amounts of full length RipATB in slow-growing mycobacterial lysates, a form that is not present in M. smegmatis lysates. Furthermore, while expression of the active domain of RipATB leads to severe growth inhibition with concomitant bulging, overexpression of full length RipATB has no such effect in M. tuberculosis, suggesting that slow-growing mycobacteria proteolytically activate less RipA than their fast-growing counterparts. The mechanism behind this additional control over RipA activation is not known, but may be at the level of expression or functional kinetics of the protease(s) responsible for RipA cleavage. In fact, there is an additional stretch of amino acids in RipASm compared to RipATB, which sits at the beginning of the N terminal inhibitory domain (Figure S7). Ultimately, an integrated mechanism for controlling PG hydrolases may represent a broad paradigm among cell wall degrading proteins. Multiple levels of regulation might be required to synchronize their activity to the cellular requirement while avoiding overactivity and toxicity. In support of this, expression of a dominant negative RipA allele at 10% of the endogenous RipA levels leads to abnormal chaining. Thus, M. smegmatis appears to carefully titrate the amount of processed RipA to nearly the minimum levels it requires for division.
Finally, beyond division mechanics, post-translational PG hydrolase regulation has the added benefit of inducing changes quickly in response to changing environmental conditions, especially in times of low transcription such as non-replicative conditions [33], [34]. The byproducts of PG hydrolysis can act as sensors for the bacterial environment, whether in vitro or within a host. For example, in B. subtilis, muropeptides have been found to be sufficient to induce spore resuscitation [35], [36] while in M. tuberculosis, RpfB, a lysozyme that is known to be a RipA interacting partner [27], is required for regrowth from both in vitro and in vivo non-replication states [37], [38]. The exact mechanism behind mycobacterial resuscitation remains unclear, but muropeptide-based signaling could play a major role. In fact, we found processed RipA species in the culture filtrates and recent work by Mir et al [39] demonstrated that the addition of muropeptides to dormant M. tuberculosis facilitated resuscitation, possibly through the binding and signaling of the essential mycobacterial integral membrane kinase, PknB. Thus, soluble PG remodeling proteins might play a role in fostering communication across a bacterial population.
In summary, this work has further defined two connected, but distinct, mechanisms to regulate the activity of RipA, a potential autolysin that is essential for septal resolution in mycobacteria. The complexity of this regulation, which involves protein interactions as well as proteolytic activation, underscores the importance of carefully coordinating cell wall hydrolysis during growth and division. By dissecting the molecular regulation of PG hydrolases, we gain fundamental insight into how the bacterial cell wall is dynamically maintained and also open up avenues for novel chemotherapeutics, especially against major human pathogens such as M. tuberculosis.
E. coli XL-1 Blue (Stratagene, Santa Clara, CA) were grown at 37°C in LB broth or agar and used for cloning. Selection was performed using kanamycin (50 µg/mL), hygromycin (100 µg/mL), ampcillin (100 µg/mL) or zeocin (25 µg/mL) when appropriate. Mycobacterium smegmatis mc2155 was grown at 37°C, unless otherwise indicated, in Middlebrook 7H9 broth supplemented with ADC (bovine albumin fraction V (Sigma)(5 g/L)-dextrose (2 g/L)-catalase (3 mg/L) and 0.05% Tween80. Selection of M. smegmatis was achieved by supplementation of kanamycin (25 µg/mL), hygromycin (50 µg/mL) or zeocin (25 µg/mL). M. tuberculosis H37Rv and M. bovis BCG were grown in liquid Middlebrook 7H9 broth and plated on Middlebrook 7H10 agar supplemented with OADC (oleic acid-albumin-dextrose-catalase) (BD Biosciences, Franklin Lakes, NJ). M. smegmatis in which the RipA endogenous promoter has been replaced by a tetracycline inducible promoter was previously constructed and characterized in [18].
Mtb RipA (Rv1477) and Msmeg RipA (MSMEG_3145) mutants were constructed through PCR stitching using the following primers:
Mtb C383A RipA Forward (CCGTCGGCTTCGACGCCTCAGGCCTGGTGTTG)
Mtb C383A Reverse (CAACACCAGGCCTGAGGCGTCGAAGCCGACGG)
Msmeg RipA C408A Forward (ACCGTCGGCTTCGACgcCTCGGGTCTGATG)
Msmeg C408A Reverse (CATCAGACCCGAGgcGTCGAAGCCGACGGT).
Msmeg RipA DP301AA DP315AA double point mutants were constructed by PCR stitching using the following primers:
Msmeg RipA DP315AA Forward (GCGATCCCGAGCGCGTTCGTCAGCGGTGcCgCCATCGCGATCATCAAC)
Msmeg RipA DP300AA Reverse (GAACGCGCTCGGGATCGCAGGCAGGGTCgCGgCCCACACGGCCCAGTT)
Secreted, RipA catalytic domain constructs were made using the M. smegmatis RipA secretion signal amino acids 1–51—Reverse primer: GAACCTgatatcGACGAGCGTGGCGAG. The RipATB active domain contained amino acids 332–472, while RipASm active domain contained amino acids 357 to 494.
Tetracycline inducible strains were created by cloning RipA genes into the Tet On plasmid, pSE100. For time-lapse microscopy, green fluorescent protein (GFP) was cloned downstream of RipASm C408A to create a transcriptional reporter. GFP was also cloned downstream in frame with RipASm to create a translational fusion. These inducible plasmids were then transformed into M. smegmatis in which the plasmid pMC1s, which encodes the tetR gene, had already been integrated at the L5 site.
Recombinant gene products were expressed using a published anhydrotetracycline inducible system [40]. Anhydrotetracycline induction was performed with 100 ng/mL of anhydrotetracycline unless otherwise indicated.
M. smegmatis with chromosomal ripA under the control of a tetracycline inducible promoter was grown in 7H9 ADC in the presence of 100 ng/mL anhydrotetracycline (aTc) to an OD of 0.2. The culture was split, the cells pelleted at 5000 rpm for 10 minutes and resuspended in 7H9 ADC with (RipA replete) or without (RipA depleted) aTc for 6 hours. Once depletion in the no inducer culture was confirmed by microscopic examination of the culture for chaining, 10 µg/mL meropenem were added and the cultures and cells grown at 37°C for 6 hours. After 6 hours of meropenem treatment, both RipA replete and depleted cultures were serially diluted and plated onto LB supplemented with hygromycin, kanamycin and 100 ng/mL aTc. Plates were incubated for 3 days at 37°C and colonies were counted. As a control, RipA replete and depleted cells were also treated with 0.08% SDS (w/vol), 0.08 µg/mL streptomycin (Sigma) or 500 µg/mL carbenicillin (Sigma).
M. smegmatis cells were grown overnight in 7H9 supplemented with dextrose (2 g/L), but not albumin or catalase at 37°C until mid log phase. This media should be made fresh for every experiment. Cells were pelleted and culture supernatants were precipitated with 10% TCA (tricholoroacetic acid) overnight at 4°C. Precipitates were pelleted at 15,000×g for 15 minutes at 4°C and washed once with ice cold acetone. The acetone wash was decanted, and the pellet was air dried at room temperature. Precipitated protein was resuspended in reducing SDS loading buffer at 65°C, for 10 minutes.
M. smegmatis cells was fractionated by French press three times. Unbroken cells and insoluble material were pelleted at 1000×g for 10 minutes. The supernatant was collected and insoluble cell wall material pelleted at 27,000×g at 4°C for 40 minutes. The remaining supernatant was centrifuged at 100,000×g for 1 hour at 4°C to pellet the membrane fraction, while te supernatant contains the soluble cytosolic fraction.
Rabbit polyclonal antibody was made from an affinity purified using a peptide derived from the Msmeg RipA epitope: NAGRKIPSSQMRRG (Genscript, Piscataway, NJ). Anti-RipA antibody was diluted to 1 mg/mL and used at a dilution of 1∶1000. Anti-FLAG antibody (Sigma, St. Louis, MO) was used according to manufacturer's instructions. Protein samples were mixed with 4x Laemmli SDS PAGE buffer (Boston BioProducts, Inc, Boston MA) and boiled for 5 minutes. M. bovis BCG and M. tuberculosis protein samples were boiled for 20 minutes. Proteins were separated on 12% Tris-glycine polyacrylamide gels, transferred to PVDF membrane (Pall Corp, Pensacola, FL), probed with anti-sera and developed with SuperSignal chemiluminescent reagent (Thermo, Pittsburg, PA). Densitometry on Western blot signal was performed using Multiguage software (Fujifilm).
For TMA-DPH (Invitrogen, Carlsbad, CA) staining, bacteria were centrifuged and media removed. Cell pellets were resuspended in 50 mM TMA-DPH in PBS and incubated in the dark for 10 minutes. Cells were also stained in FM4-64Fx (Invitrogen) at a concentration of 5 µg/mL in PBS for 10 minutes, and then fixed and stored in 4% paraformaldehyde. Samples were imaged using a Nikon TE-200E microscope with a 100x (NA1.4) objective and captured with an Orca-II ER cooled CCD camera (Hamamatsu, Japan). Shutter and image acquisition were controlled using Metamorph Software (Molecular Devices). Final images were prepared using Adobe Photoshop 7.0.
Four Gene Frames (Fisher Scientific) were stacked onto a glass slide and filled with Middlebrook 7H9 in low melting point agar, supplemented with 50 µg/mL of hygromycin and 100 ng/mL aTc. A glass coverslip was flattened atop the agar to create a smooth surface and then removed after the agar set. The agar pad was sliced into eighths and seven of the pieces were removed to provide an air reservoir. Onto the remaining pad, exponential phase M. smegmatis was pipetted and allowed to adsorb until the surface of the pad appeared dry. Finally, a glass coverslip was applied and the slide was imaged on the microscope in an environmental chamber warmed to 37°C (Applied Precision, Inc.).
Time-lapse images were acquired using a DeltaVision epifluorescence microscope with an automated stage enclosed with a 100x oil objective (Plan APO NA1.40). Cells were imaged every 10 minutes for up to 18 hours using brightfield and fluorescence illumination (461–489 nm; Applied Precision, Inc.) and images recorded with a CoolSnap HQ2 camera (Photometric). Focus was maintained using the software-based autofocus (Applied Precision, Inc).
M. smegmatis samples were collected at the indicated growth phases (log, OD600 = 0.5; early stationary, OD600 = 2; Stationary, 24 hours of OD600>7) and stored in RNA Protect Bacteria Reagent (Qiagen, Valencia, CA) at −80°C. The pellets were then mechanically disrupted by beadbeatting for three 1-minute cycles, and RNA isolated using the RNeasy Mini Kit (Qiagen), with one additional DNAse treatment (Qiagen) on the column before elution and a second DNAse digestion with Turbo DNase according to manufacturer's instructions (Ambion, Foster City, CA). Reverse transcription was carried out using the High Capacity cDNA Reverse Transcription kit (Applied Biosystem, Foster City, CA). Quantitative PCR reactions were set up in Power SyBr green PCR master mix (Applied Biosystems) and run and analyzed on a Step One Plus real time system (Applied Biosystems).
ripA expression was measured using the following intragenic primers: 5′ CAGATCGGTGTGCCCTACTC; 5′ GGCGAACATGTAGAGCATCAG; or against the 3′ UTR region: 5′ GCTCGAGGCCCCTTACAC; 5′ GGAGCGCAAAGTAATCCCATCAG
ripA expression was normalized to sigA levels, which utilized the following primers: 5′ AAGACACCGACCTGGAACTC; 5′AGCTTCTTCTTCCTCGTCCTC.
|
10.1371/journal.pntd.0000734 | Where Do We Go from Here? Prevalence of Trachoma Three Years after Stopping Mass Distribution of Antibiotics in the Regions of Kayes and Koulikoro, Mali | A national survey in 1997 demonstrated that trachoma was endemic in Mali. Interventions to control trachoma including mass drug administration (MDA) with azithromycin were launched in the regions of Kayes and Koulikoro in 2003. MDA was discontinued after three annual rounds in 2006, and an impact survey conducted. We resurveyed all districts in Kayes and Koulikoro in 2009 to reassess trachoma prevalence and determine intervention objectives for the future. In this paper we present findings from both the 2006 and 2009 surveys.
Population-based cluster surveys were conducted in each of the nine districts in Koulikoro in 2006 and 2009, whilst in Kayes, four of seven districts in 2006 and all seven districts in 2009 were surveyed. Household members present were examined for clinical signs of trachoma.
Overall, 29,179 persons from 2,528 compounds, in 260 clusters were examined in 2006 and 32,918 from 7,533 households in 320 clusters in 2009. The prevalence of TF in children aged 1–9 years in Kayes and Koulikoro was 3.9% (95%CI 2.9–5.0%, range by district 1.2–5.4%) and 2.7% (95%CI 2.3–3.1%, range by district 0.1–5.0%) respectively in 2006. In 2009 TF prevalence was 7.26% (95%CI 6.2–8.2%, range by district 2.5–15.4%) in Kayes and 8.19% (95%CI 7.3–9.1%, range by district 1.7–17.2%) in Koulikoro among children of the same age group. TT in adults 15 years of age and older was 2.37% (95%CI 1.66–3.07%, range by district 0.30–3.54%) in 2006 and 1.37% (95%CI 1.02–1.72%, range by district 0.37–1.87%) in 2009 in Kayes and 1.75% (95%CI 1.31–2.23%, range by district 1.06–2.49%) in 2006 and 1.08% (95%CI 0.86–1.30%, range by district 0.34–1.78%) in 2009 in Koulikoro.
Using WHO guidelines for decision making, four districts, Bafoulabe in Kayes Region; and Banamba, Kolokani and Koulikoro in Koulikoro Region, still meet criteria for district-wide implementation of the full SAFE strategy as TF in children exceeds 10%. A community-by-community approach to trachoma control may now be required in the other twelve districts. Trichiasis surgery provision remains a need in all districts and should be enhanced in six districts in Kayes and five in Koulikoro where the prevalence exceeded 1.0% in adults. Since 1997 great progress has been observed in the fight against blinding trachoma; however, greater effort is required to meet the elimination target of 2015.
| Trachoma, a blinding bacterial disease, is targeted for elimination by 2020. To achieve the elimination target, the World Health Organization (WHO) recommends member states implement the SAFE strategy; surgery, mass administration of antibiotics, promotion of hygiene and facial cleanliness and water and sanitation as environmental improvements. We present results from evaluation surveys conducted in 2006 and 2009 from the regions of Kayes and Koulikoro, Mali. Prevalence of active trachoma in 2006 was below baseline intervention thresholds in all surveyed districts and the national program stopped antibiotic distribution. The prevalence of trachoma in 2009 remained well below levels in 1998. However, in 8 of 13 districts compared, the prevalence of active trachoma was higher in 2009 than 2006. Three years of antibiotic intervention did not equate in all districts to a sustained reduction of active trachoma. No surveillance activities were implemented after stopping interventions. Surgical interventions may have reduced the burden of blinding trachoma but there is an ongoing need for surgeries specifically targeting affected women. Four districts meet the WHO criteria for resuming district-wide mass antibiotic distribution. A community-by-community approach to elimination may be needed in other districts. The promotion of facial cleanliness and good hygiene behavior should be reintroduced.
| Trachoma, a blinding bacterial disease of the conjunctiva, is targeted for elimination as a public health problem by the year 2020, yet an estimated 8.2 million people remain at immediate risk of blindness or visual impairment due to the disease [1]. To achieve the elimination target, the World Health Organization (WHO) recommends member states implement an integrated strategy of interventions known as SAFE: surgery to correct trachomatous trichiasis; mass administration of antibiotics to treat current trachoma infections and reduce the infectious reservoir; promotion of hygiene and facial cleanliness; and water and sanitation as environmental improvements aimed at interrupting transmission of the infection. Based on WHO guidelines, districts are categorized for intervention based on the prevalence of clinical signs of disease: trachomatous inflammation follicular (TF) in children aged 1–9 years and trachomatous trichiasis (TT) in adults aged 15 years and older [2], [3].
Following a national trachoma prevalence survey in 1997, The National Blindness Prevention Program in Mali initiated a trachoma control program. Mapping of trachoma in Mali identified trachoma to be of public health significance throughout the country, including the regions of Kayes and Koulikoro where the prevalence of TF in children less than 10 years of age was 42.5% and 33.5% respectively [4]. The highest levels of TT among women 15 years of age and older were observed in Kayes (3.3%) and Koulikoro (3.9%) [4]. From 2002 to 2006 all sixteen districts in Kayes and Koulikoro received SAFE interventions to control trachoma. The interventions implemented in each region are listed in Table 1. Interventions were conducted in several ways: trained ophthalmic nurses moved from village to village offering free trichiasis surgery; mass distribution of oral azithromycin and tetracycline ophthalmic ointment occurred in annual campaigns for three consecutive years in each district following pilot distributions in target areas; facial hygiene, latrine construction and use, and the utilization of water for hygiene were promoted over local and regional radio stations; and persons in each region were trained to deliver health education and promote behavior change. The number of doses distributed and population coverage with azithromycin by district and year is shown in Table 2. In 2006, after three years of intervention and in accordance with the WHO guidelines, an impact evaluation was conducted to assess the effect of the SAFE activities [2]. The Ministry of Health withdrew A, and support for F and E interventions from partner organizations was limited. The Ministry of Health concentrated efforts to scale up the SAFE strategy in other regions yet to initiate interventions. The purpose of this study was to re-evaluate the prevalence of trachoma three years after SAFE interventions were discontinued in Kayes and Koulikoro. Here we present the data from the first impact evaluation in 2006 and the recent 2009 evaluation. We also aimed to quantify any need for additional interventions.
These prevalence surveys were conducted in accordance with WHO guidelines as part of the ongoing effort of the Ministry of Health to eliminate blinding trachoma in Mali and were necessary to evaluate the impact of interventions. In addition to the Ministry of Health, the survey protocol was approved by the Emory University IRB under protocol 079-2006. Informed verbal consent and assent was received according to the principles of the Declaration of Helsinki. Written consent was not obtained in these surveys due to the low literacy rate, ranging from 3% in rural Mali to 38% in Bamako (Enquête Démographique et de Santé 2001). Emory University IRB approved the use of informed verbal consent. Oral informed consent was sought first from village chiefs before surveys were conducted in the randomly selected villages. Consent was then obtained from household heads of randomly selected households and finally oral consent was obtained from each adult examined and consent from a parent or caretaker was obtained to examine children. Verbal assent was obtained also from children 6–14 years of age. Survey participants were informed of the purpose of the trachoma examinations and their rights not to participate or to stop the examination at any time. Choosing not to participate did not affect any decision in determining the need for interventions. Verbal consent was documented on a standard survey data collection tool. All children presenting signs of TF or trachomatous inflammation intense (TI) were offered free tetracycline eye ointment and instructed to apply it twice daily for 6 weeks. Persons identified with TT were recorded, counseled, and offered free consultation and surgery with a trained TT surgeon.
Kayes Region is located in the extreme west of Mali bordering Mauritania to the north, Senegal to the west, and Guinea to the southwest (Figure 1). The region is divided into seven health districts with an estimated combined population of 1,763,987 persons (Mali National Demographic and Statistical Institute 2009 population projection). The primary ethnic groups are the Sarakole (Soninke) and Bambara.
Koulikoro Region is located in the western interior of Mali directly east of Kayes Region. It also borders Mauritania to the north and Guinea to the southwest. Koulikoro is divided into nine health districts with an estimated population of 2,072,185 persons (Mali National Demographic and Statistical Institute 2009 population projection). The primary ethnic groups are the Bambara and Malinke.
In both 2006 and 2009, population-based cross-sectional household surveys were conducted at the district level. Each survey in 2006 was done at least 6 months after the last round of antibiotic distribution following the implementation plan of the national program. Thus some districts in Kayes and Koulikoro were surveyed during the period between March and May and some during November and December. In 2009, all districts were surveyed during the period between March and May. Twenty villages (clusters) were selected from each district with a probability of selection proportional to the total population of the village. All villages with less than 5,000 total population from each district were eligible for selection. In 2006, 4 of 7 districts in Kayes (Bafoulabe, Diema, Kita and Nioro du Sahel) and all 9 districts in Koulikoro were assessed. In each of these districts, concessions (household compounds) were systematically selected using the random direction method [5]. All residents aged 1–9 years of age and 15 years of age and older from all households within selected concessions were examined for clinical signs of trachoma until approximately 60 qualifying children had been examined. In 2009, all 16 districts in the two regions were surveyed for a total of 320 clusters. Households within a cluster were randomly selected following the method of sketch mapping and segmentation which aimed to survey 24 households per cluster [6]. With the assistance of village leaders, survey teams drafted a list of all households, dividing households into segments of four. Village chiefs selected segments via lottery. All households in a selected segment were surveyed and all consenting persons over six months of age in each household were examined for trachoma. From the sampling methodology used in both surveys we assumed that the data was self-weighted.
Residents of selected households in 2009 were enumerated and designated as either present or absent. Absent was defined as not being physically present in the village on the day of the survey. Enumerated persons who were not at home, but in the village were found and recruited to the survey. Teams made at least one attempt on the same day to find residents marked absent during the first visit to the household. Absent residents in the 2006 survey were not enumerated. Clinical signs of trachoma were assessed using the WHO Simplified Grading System [3]. Examiners recorded the presence or absence of all trachoma grades in both eyes of survey participants using a ×2.5 binocular loupe and adequate light. The findings from the worst affected eye were reported. At examination in 2009, children were assessed for a clean face, defined as the absence of both ocular and nasal discharge. In 2009, each child 6–15 years of age was asked about their attendance in school, defined as public or private non-religious school. Attendance of Koranic schools or non-formal education was not assessed. Examined persons were asked about their participation in the most recent round of antibiotic distribution for trachoma control, defined as whether a person took oral azithromycin, applied tetracycline eye ointment, or did not participate. Estimates for participation in antibiotic distribution included only those persons present to give a response. Additionally in 2009, one adult respondent was interviewed in each household to determine the presence and use of a household latrine and the location of the main water source used by the household. The presence of a latrine was confirmed by direct observation and ‘use’ was defined as the observation of feces in the pit. The location of the water source was designated as within the household compound, within the village, or outside the geographical village boundaries as a proxy for distance and availability of water. Household interviews were not conducted in the 2006 surveys.
Prior to the surveys, ophthalmic nurses were trained to use the WHO Simplified Trachoma Grading System through repetitive grading of digital photographs in a classroom setting and assessment of individual patients in the field. In 2009, these exercises were followed by a formal inter-observer reliability test of trachoma grading against a standardized set of 50 slides presented on a computer and a field exam of 50 children in which SB, DS and JDK were considered the reference examiners. Reference grading was supplemented with digital photographs. Eight out of ten ophthalmic nurses met the criteria of achieving greater than 80% reliability and a kappa statistic of 0.6 and above for grade TF and were selected as examiners for the survey. Survey teams were trained to randomly select households within a cluster, conduct household interviews, and record findings on standardized forms. A survey team consisted of one data recorder and one ophthalmic nurse. Formal inter-observer reliability tests were not conducted for the examiners in 2006.
Data were double-entered, compared and corrected. Based on the survey design used, we adjusted confidence intervals for the prevalence estimates and odds ratios to account for correlation among the data due to clustering using SAS SURVEY procedures (SAS version 9.2, SAS Institute Inc) [7], [8]. Regional prevalence estimates accounted for population differences between districts. We calculated differences between prevalence estimates from 2006 and 2009 and tested the equality of the estimates using the Z statistic with α = 0.05.
The ultimate intervention goal considered for achieving blinding trachoma elimination is the presence of less than 1 TT case per 1,000 population [9]. We calculated the total backlog of persons with TT in need of surgery by multiplying the 2009 point estimate and confidence limits of the population prevalence of TT by the estimated total population to give a point estimate and lower and upper bounds of the total number of people to be operated. According to WHO guidelines, where district-level prevalence of TF in 1–9 year-old children exceeds 10% at baseline, A, F and E activities are warranted district-wide and thus the total population living in these areas is targeted [2]. Where SAFE activities have been implemented, all areas that remain above 5% TF prevalence among children should continue antibiotic distribution [2]. The target prevalence by which mass antibiotic interventions to control trachoma is not needed is below 5% TF [2]. We calculated the number of household latrines required to achieve goal 7c of the United Nations Millennium Development Goals (MDGs); halve by 2015 the proportion of people who do not have access to improved sanitation [10].
In 2006, a total of 29,179 persons were examined from 29,779 persons available in 2 528 selected concessions. The mean number of concessions per district was 194.5 with a range by district of 110 to 312 concessions. A mean of 9.7 concessions were surveyed per village (range by village 1–25). The mean number of households per concession was 1.9 (range by concession 1–17). In the four surveyed districts of Kayes, 4,168 adults over 14 years of age and 4,808 children 1–9 years of age were examined for clinical signs of trachoma. In Koulikoro, 9,679 adults and 10,524 children were examined. Among examined adults over 14 years of age, 68.9% were women and among examined children 1–9 years of age, 51.5% were girls.
The prevalence estimates of clinical signs of trachoma in Kayes and Koulikoro in 2006 are presented in the first three columns by district in Table 3. Among adults 15 years of age and older, the prevalence of TT in Kayes was 2.37% (95%CI 1.66–3.07%, range by district 0.30–3.54%) and in Koulikoro, 1.75% (95%CI 1.31–2.23%, range by district 1.06–2.49%). The prevalence of trachomatous scaring (TS) among adults was 10.33% (95%CI 8.6–12.0%, range by district 3.0–18.4%) and 4.18% (95%CI 3.5–4.8%, range by district 0.6%–9.2%) in Kayes and Koulikoro respectively (data not shown). Prevalence of trachomatous corneal opacity (CO) in Kayes was 0.38% (95%CI 0.14–0.61, range by district 0.0–0.9%) and 0.31% (95%CI 0.12–0.51%, range by district 0.0–0.9%) in Koulikoro. Women were more likely than men to have TT (OR = 1.61, 95%CI 1.16–2.23, p = 0.004). Adults 50 years and older were more likely to have TT than adults aged 15–49 years (OR = 6.73, 95%CI 4.99–9.07, p = <0.0001).
District-level prevalence of TF among children 1–9 years of age had reduced to below the 10% intervention threshold in all surveyed districts. Among children 1–9 years of age, the prevalence of TF was 3.9% (95%CI 2.9–5.0%, range by district 1.2–5.4%) in Kayes and 2.7% (95%CI 2.3–3.1%, range by district 0.1–5.6%) in Koulikoro. The prevalence of trachomatous inflammation intense (TI) among children aged 1–9 years of age was 1.0% (95% CI 0.6–1.5%, range by district 0.3–1.8%) in Kayes and 0.4% (95% CI 0.2–0.5%, range by district 0–2.0%) in Koulikoro. Active trachoma (TF and/or TI) prevalence was 4.53% (95%CI 3.3–5.7%, range by district 1.2–6.5%) in Kayes and 2.96% (95%CI 2.5–3.4%, range by district 0.2–5.6%) in Koulikoro.
In 2009, from all districts in both regions a total of 42,128 persons were enumerated in 7,533 households and 32,918 were examined. In Kayes, a total of 13,576 persons were examined for signs of trachoma out of 17,127 persons enumerated from 3,287 households for a response rate of 79.3%. In Koulikoro, 19,342 persons were examined out of 25,001 persons enumerated from 4,246 households (a response rate of 77.4%). The response rate in women was 83.1% (17,771/21,386) and 73.0% (15,147/20,742) in men. The majority of adult men unable to be examined were absent from the home at the time of the household visit. Children 1–9 years of age composed 33.9% of the total enumerated population. Adults 15 years of age and older were 51.0% of the total population. The proportion of enumerated children 1–9 years of age who were examined was 88.2% and 77.1% of enumerated adults were examined. Among examined adults over 14 years of age, 58.1% were women and among examined children 1–9 years of age, 49.7% were girls. Among children 6–15 years of age the proportion that reported attending school was 42.6% in Kayes and 54.1% in Koulikoro.
The prevalence estimates of clinical signs of trachoma in Kayes and Koulikoro for 2009 are presented by district in the last four columns of Table 3. The prevalence of TT in the total population of Kayes region was 0.69% (95%CI 0.53–0.85%, range by district 0.20–0.91%). In Koulikoro, TT prevalence in the total population was 0.56% (95%CI 0.43–0.69%, range by district 0.25–0.85%). Among adults 15 years of age and older, the prevalence of TT in Kayes was 1.45% (95%CI 1.10–1.79%, range by district 0.37–1.87%) and in Koulikoro, 1.10% (95%CI 0.84–1.35%, range by district 0.34–1.78%). The prevalence of trachomatous scaring (TS) among adults was 4.22% (95%CI 3.7–4.8%, range by district 0.3–5.3%) and 4.68% (95%CI 4.1–5.2%, range by district 1.4–8.1%) in Kayes and Koulikoro, respectively (data not shown). Prevalence of CO in Kayes was 0.11% (95%CI 0.03–0.19, range by district 0–0.42%) and 0.21% (95%CI 0.13–0.29%, range by district 0–0.73%) in Koulikoro. Odds of TT among adults 50 years of age and older were ten times higher than adults 15–49 years of age (OR = 10.61, 95%CI 7.62–14.78, p<0.0001). Women were nearly two times more likely to have TT than men (OR = 1.85, 95%CI 1.40–2.46, p<0.0001).
At the regional level, the prevalence of TF was 6.6% (95%CI 5.7–7.5%, range by district 2.5–15.4%) among children 1–9 years of age in Kayes and 8.7% (95%CI 7.5–9.9%, range by district 1.7–17.2%) in Koulikoro. The prevalence of TI among children aged 1–9 years of age was 1.5% (95% CI 1.1–1.8%, range by district 0.3–3.3%) in Kayes and 0.6% (95% CI 0.4–0.8%, range by district 0–1.9%) in Koulikoro. Active trachoma (TF and/or TI) prevalence was 7.34% (95%CI 6.4–8.3%, range by district 2.7–16.8%) in Kayes and 8.91% (95%CI 7.7–10.1%, range by district 2.0–17.9%) in Koulikoro.
A total of 7,533 households were surveyed (range by district 423–480). The mean number of persons living in each household was 5.2 (SD = 2.7, range by district 4.5–5.9) in Kayes and 5.9 (SD = 3.0, range by district 4.9–6.9) in Koulikoro. Indicators of uptake of the A, F and E components of the SAFE strategy are listed by district in Table 4. The proportion of examined household residents reporting taking azithromycin or using tetracycline eye ointment in the most recent round of distribution was 86.1% (95%CI 84.2–88.0, range by district 54.6–99.8%) in Kayes and 83.9% (95%CI 81.6–86.3%, range by district 59.9–96.8%) in Koulikoro. Among children 1–9 years of age, 76.5% (95%CI 74.3–78.7%, range by district 46.7–95.2%) and 75.0% (95%CI 71.8–78.1%, range by district 52.1–86.8%) in Kayes and Koulikoro, respectively, had a clean face at examination. Basic sanitation (a household latrine) was evident in over 80% of the households in 12 out of the 16 districts. The presence of a latrine with evidence of use was observed in 88.1% (95% CI 85.2–91.1%, range by district 50.4–100%) of surveyed households in Kayes and 87.2% (95%CI 84.5–89.9%, range by district 37.4–99.8%) in Koulikoro. A water source inside the compound was observed in 9.7% (95%CI 5.5–13.9%, range by district 0.6–17.2%) of surveyed households in Kayes and 3.2% (95%CI 0.5–6.0%, range by district 0–11.8%) reported having to travel outside the geographical boundaries of the village to collect water. In Koulikoro, 19.8% (95%CI 15.6–24.0%, range by district 0–31.9%) of households had a source of water within the compound and 5.6% (95%CI 3.5–7.7%, range by district 0–17.2%) reported having to collect water from a source outside of village boundaries. Overall, there was no difference in the prevalence of clean faces between children living in households with water access inside the compound and children in households where the water source was outside the compound; 76.3% compared to 75.7%, Z = 0.48, p = 0.633.
The regional estimates of prevalence of TT and CO among women 15 years of age and older from 2006 and 2009 are plotted in Figure 2 with the same estimates reported in 1997 for a comparison to baseline prevalence. In this age group, the difference in prevalence of TT between 2006 and 2009 for Kayes (Z = −1.33, p = 0.1829) and Koulikoro (Z = −1.78, p = 0.0744) regions was not statistically significant. There was no statistically significant difference in regional estimates of CO among adult women from 2006 to 2009 (Kayes Z = −1.01, p = 0.3117; Koulikoro Z = −0.44, p = 0.6626). However, among adults of both genders, the prevalence of TT in 2009 was less than the estimate in 2006 for both Kayes (Z = −2.06, p = 0.0396) and Koulikoro (Z = −2.79, p = 0.0052). The prevalence of CO among all adults between 2006 and 2009 did not differ in Kayes (Z = −1.21, p = 0.2245) or Koulikoro (Z = −0.55, p = 0.5838).
The regional prevalence of TF in 2009 was statistically greater than that observed in 2006 for both regions (Kayes Z = 8.13, p<0.0001; Koulikoro Z = 16.20, p<0.0001). The prevalence of TF for the region in 1997 is plotted with the district level estimates of TF from 2006 and 2009 in Figure 3. The differences in district level estimates between 2009 and 2006 with confidence intervals are listed in Table 5 along with Z statistic and p-values. The prevalence of TF observed in 2009 was statistically greater than that observed in 2006 for Bafoulabe, Nioro du Sahel, Banamba, Dioila, Fana, Kati, Kolokani and Koulikoro districts. The prevalence of TF in 2009 was the same or less than that observed in 2006 in the districts of Diema, Kita, Kangaba, Nara and Ouelessebougou. Regional estimates of TI among children from 1997, 2006 and 2009 are shown in Figure 4. Also for both regions, the prevalence of TI in this study was greater than that observed in 2006 (Kayes Z = 2.06, p = 0.0198; Koulikoro Z = 1.86, p = 0.0316). The prevalence of TI observed in 2009 was statistically greater than that observed in 2006 for Bafoulabe, Banamba, Dioila and Fana districts.
Based on 2009 estimates, the total number of persons with TT who remain in need of surgery in Kayes is 10 967 (lower and upper bounds: 7,144 to 14,123) and 10,726 (bounds: 9,932 to 16,487) in Koulikoro. TT prevalence among adults exceeded 1% in 11 of 16 districts warranting continued, enhanced efforts to provide surgery to affected patients. While TT surgery may not be a priority in Yelimane, Fana, Kangaba, Nara and Oulessebougou where TT among adults is less than 1%, eye care facilities with the capacity to operate presenting TT cases should exist.
Mass distribution of antibiotics should resume in Bafoulabe, Banamba, Kolokani and Koulikoro where the prevalence of TF exceeds 10% among children. Additionally, according to WHO guidelines, mass distribution of antibiotics should continue in areas where after three years of intervention, the prevalence of TF remains greater than 5% among children under 10 years of age [2]. This would include communities within all of the 12 other districts of Kayes and Koulikoro. If this WHO guideline is interpreted at the district level, 10 districts in Kayes and Koulikoro regions warrant ongoing mass distribution of antibiotics, targeting a total population of approximately 2,637,492 persons.
The promotion of facial hygiene and environmental improvements should resume in all districts. Access to water within village boundaries and household latrine coverage was not lacking in most districts. Fana, Dioila, Yelimane and Nara had the highest proportion of households reporting having to collect water outside village boundaries. An estimated 11,526 households in Kayes and 19,718 households in Koulikoro must collect water outside village boundaries. The construction and maintenance of water points could be targeted to communities where access to water is lacking. Kenieba in Kayes and Nara in Koulikoro had the lowest proportion of households with a latrine. To ensure that every household has access to basic improved sanitation, 41,712 latrines need to be built in Kayes and 33,928 in Koulikoro. Building half of these by 2015 would meet MDG 7c.
The national survey conducted in 1997 providing baseline regional prevalence estimates were very useful in establishing the widespread nature of trachoma in Mali. In response to results from the national survey, trachoma control interventions were initiated in Kayes and Koulikoro Regions. Interventions were focused mostly on the S and A components of SAFE. F and E interventions were implemented, but had less geographical coverage of the target population than S and A. Until 2006, monitoring of interventions was limited to program reports and did not include rigorous field evaluations. According to reports of azithromycin distributed, antibiotic coverage was not consistent between districts or years with reported district level coverage ranging from 20.9% to 108.6% after the pilot phase in 2002. Several districts failed to reach the desired minimum of 80% coverage of total population in any one year and only one averaged above 80% over three years. These inconsistencies in coverage were either due to problems with the distribution or estimates of the target population. For example, the total population registered prior to MDA may have exceeded the census estimate of total population where coverage was greater than 100%. An evaluation of antibiotic distribution in Southern Sudan demonstrated the limitations of using distribution reports alone to calculate population coverage, as population estimates and treatment records can lead to inaccurate coverage estimates [11]. In Kayes and Koulikoro, we have defined “distributed” as the total number of doses reported to have been given out to individuals during mass drug administration campaigns, and caution that they have not been validated with coverage surveys. The figures are those reported by the district to the national program.
The first impact assessment in 2006 found prevalence of TF among children to be below 5% in 9 of 13 districts and below 10% in all districts. The programmatic decision was made to focus the available resources to other endemic regions that had not initiated SAFE interventions. This resulted in stopping mass antibiotic distribution and limited ongoing promotion of facial cleanliness and environmental improvements through schools and radio. Follow-up on the progress of latrine construction and new water points targeted for trachoma control stopped. Surgical services to correct trichiasis were maintained. Between the surveys in 2006 and 2009, it appears as though clinical signs of active trachoma returned in eight out of the thirteen districts.
The current data have several possible interpretations. There may have been a true decline in active trachoma from baseline to the present and this decline is associated with interventions from 2003–2006. Although prevalence of active trachoma signs are higher in some districts now than observed in 2006, the prevalence remains well below the 34% and 42% TF reported in Koulikoro and Kayes respectively during the 1997 baseline survey. National programs do not have control groups and it is not possible to determine whether the decline is due to the intervention, or to a secular decline, as has been described elsewhere [12], [13], [14], [15]. Prevalence of active trachoma has been observed to decline in the absence of a trachoma control program in the dessert region of Kidal [16]. We may also consider that there has been an heterogeneous effect of the interventions with some districts showing a sustained reduction in the prevalence of TF (Diema, Kita, Fana, Kangaba, Nara and Oulessebougou) and others showing a rapid rebound after initial control (Bafoulabe, Nioro du Sahel, Banamba, Dioila, Kati, Kolokani and Koulikoro). Such random effects are assumed possible by chance at the community level according to a stochastic model of trachoma transmission [17]. Models also suggest that trachoma endemicity at baseline is predictive of return of infection after antibiotic intervention [18], yet we have no district-level estimates at baseline on which to make assumptions. Antibiotic coverage is an important factor in the return of infection after treatment and thus the elimination of trachoma [17], [18], [19]. It is possible that high-risk marginalized sections of the population are systematically missed in mass drug administration leaving them untreated and able to repeatedly reintroduce infection into treated communities. Coverage surveys performed immediately following the mass distribution campaigns at least once during the three years of intervention may have identified any such problem.
Alternatively, there may be no difference between prevalence estimates of active trachoma in 2006 and 2009 due to the differences between the survey methods used and season of assessment in some districts, although this is unlikely given the scale of the observed differences and that seasonality of trachoma in Mali has not been established. In the 2006 survey, household selection methods may have biased the samples in villages where only a few large concessions were selected. The starting points were markets or mosques, structures typically at the center of a community and often surrounded by more populated concessions. Some clusters in the 2006 survey were composed of persons examined from households within a few, very large concessions, rather than the randomly distributed sample of households obtained using the sketch mapping and segmentation in the 2009 surveys. The household sampling in 2009 was more similar to that used in 1997 where a systematic random sample was taken from a listing of households within clusters. Both evaluations began with training ophthalmic nurses in the WHO simplified trachoma grading system. However, in 2009, the grader's reliability to diagnose TF was assessed rigorously and nurses not meeting a certain criteria were excluded from serving as a grader. This type of field assessment should improve the validity and reliability of a grader's findings.
The observed reduction in the prevalence of TT may have been a direct effect of the ongoing surgical services provided to TT patients. The diagnosis of TT is straightforward and allows less room for subjectivity than TF since the grade is based on one or more lashes touching the eye, rather than 5 or more follicles greater than 0.5 mm in diameter in the central part of the tarsal conjunctiva [3]. The grader's ability to identify TT is assessed in the classroom using slides but not in the field reliability assessment [2]. It may have been possible that graders under diagnosed TT in the field, but the possibility of this type of misclassification should not have differed from 2006 and 2009. Additionally, a greater proportion of adult males were absent from the household than females. TT impairs vision and thus compromises mobility; therefore men with healthy eyes may be more likely to be absent and not examined. Only if the reverse is true, men present in the household are more likely to have unhealthy eyes, would any bias in the prevalence of TT in men have masked any gender difference in TT. In both impact evaluations, women were more likely to have TT than men, which is consistent with findings from a recent review on the association of gender and trichiasis [20]. The statistically significant difference observed between TT prevalence among adults of both genders, but not among adult females from 2006 to 2009, may suggest a gender disparity in benefit from ongoing surgical services with men being more likely than women to present for surgery. Eliminating the backlog of trichiasis patients needing surgery remains a priority in both regions. Surgical services may need modification to specifically target women.
The indicators for A, F and E uptake (Table 4) obtained from the household surveys have several limitations. Although antibiotic coverage obtained from personal reports from household residents appears high, these results should be interpreted with caution. Residents were asked whether they had taken azithromycin during the most recent mass distribution campaign, which was in 2006. It is unlikely that residents could recall specifically taking drugs for trachoma given that mass drug distribution campaigns for other NTDs had occurred in more recent years. Additionally, only responses from residents available to respond were taken. These residents may have been more likely to have been available to receive antibiotics during campaigns than those residents absent from the household at the time of the survey, potentially inflating the coverage estimate. Not surprisingly, these personal reports are higher than coverage estimated by district distribution reports (Table 2).
More than 75% of households surveyed in each district, except Nara and Keneiba, had access to a household latrine with evidence of use. The evidence of use was determined by the presence of feces in the pit, which may be incorrectly interpreted as latrine use by all persons within the household. A latrine will be categorized as ‘in use’ if only a proportion of the household is using it. The role of latrines in reducing trachoma transmission assumes that where latrines are used, no open human feces is available for flies to utilize as a medium for egg and larval development; reducing successive fly populations and reducing the number of fly to eye contact. However, if use of a latrine is limited to only certain groups or if certain groups choose not to use the latrine, open defecation will continue. Further evaluation may be needed to assess actual behavior and potentially explain conflicting outcomes of endemic trachoma in the presence of high sanitation coverage as seen in Bafoulabe and Banamba districts. Assessing behavior is also necessary in determining the influence of water on trachoma. In this survey, there was no association of a clean face and having access to a water source within the boundaries of the household compound or having access to a water source outside of village boundaries. Additionally, greater than 80% of children were observed to have a clean face in only four districts; indicating that the practice of face-washing has not been fully accepted and adopted among residents in the two regions. On the contrary, our findings may also indicate that the ability of F and E components to control trachoma may not be as effective as anticipated. However, a recent analysis of factors associated with active trachoma in Mali supports the utility of face washing and environmental improvements [21].
One of the criteria for the certification of the elimination of trachoma is to demonstrate the sustained reduction of prevalence of TF among children below 5% for a period of three years after interventions have ceased [8]. In only six districts did the point estimate of the prevalence of TF remain below 5% at the district level from 2006 to 2009. In the 2009 survey the prevalence of TF among children was above 10% in four districts and above 5% in another six districts. With the recent global expansion of mass distribution of antibiotics for trachoma elimination, national programs may soon face a need to prioritize a limited quantity of drug [22]. Given such circumstances, Mali is facing unique programmatic decisions. Currently, WHO guidelines suggest the district be the implementation unit, but for certification of elimination, no community must have more than 5% TF among children [2], [9]. This suggests a community-by-community approach to trachoma elimination even in districts where district-level estimates of TF prevalence are below 5%. There are no recommendations or guidelines as to how a country such as Mali should attempt to demonstrate each and every community throughout the vast landscape has reached the elimination target. An acceptable level of TF prevalence at which the risk of developing blinding trachoma has been eliminated is unknown if the acceptable level is not zero. TF is not closely correlated with the presence of Chlamydia trachomatis DNA on ocular swabs and is thought to linger in the absence of infection [23], [24], [25], [26]. However, TI is better associated with DNA positive ocular swabs and is also linked to increased likelihood of progression to scarring, so is considered a more severe form of the disease [23], [27]. Ocular Chlamydia infection may have been significantly reduced by the interventions as evidenced by prevalence of TI in 9 districts of less than 1 child per 100. TI is more closely correlated to current infection with Chlamydia trachomatis than residual TF and has been suggested as a potential marker of infection post treatment [28]. In this setting, microbiological supporting evidence of the presence of bacteria would be useful, yet no guidelines exist for the use of laboratory diagnostics on a programmatic scale and it is perceived that costs of adding such tests to impact evaluations are prohibitive. Using TI for a proxy of infection, the prevalence of TI in 5 of the 10 districts in Kayes and Koulikoro qualified to receive mass distribution of antibiotics based on TF, indicates that less than 1 per 100 children would receive trachoma-specific benefits from the antibiotic.
Achieving less than 5% TF at the district level is achievable and can be feasibly determined on a programmatic scale through the cluster random survey design as demonstrated in this study and in Ghana [29]. Not considering the differences in survey methodology, a district level prevalence of less than 5% TF after three continuous years of heavy antibiotic intervention did not equate in all districts to a sustained reduction of TF below 5%. No surveillance activities were implemented after stopping AFE interventions in these districts. Doing so may have identified resurgence in districts with an apparent rebound in active trachoma and allowed immediate intervention. Results from these surveys provide evidence in the setting of a national program that antibiotics alone are not enough to eliminate trachoma. An analysis of associations between the components of the SAFE strategy demonstrates clearly that changes in hygiene behavior and improved sanitation can have protective effects against active trachoma [30], which argues for equal emphasis on hygiene and environmental improvements. Indicators used in the 2009 survey suggest very high access to sanitation in the two regions, but the indicators fail to capture actual behaviors. The promotion of facial cleanliness and good hygiene behavior should be reintroduced in all districts of Kayes and Koulikoro. Surgical services to correct trichiasis should also be continued, but where and for how long to continue mass distribution of antibiotics is not as clear. Currently, the 4 districts with TF above 10% among children are priority for mass distribution of antibiotics. More guidelines from the international community are urgently required to help prioritize the limited quantity of donated antibiotic in addition to recommending appropriate evaluation methodology for determining when certification targets have been achieved.
|
10.1371/journal.pntd.0004578 | Improved PCR-Based Detection of Soil Transmitted Helminth Infections Using a Next-Generation Sequencing Approach to Assay Design | The soil transmitted helminths are a group of parasitic worms responsible for extensive morbidity in many of the world’s most economically depressed locations. With growing emphasis on disease mapping and eradication, the availability of accurate and cost-effective diagnostic measures is of paramount importance to global control and elimination efforts. While real-time PCR-based molecular detection assays have shown great promise, to date, these assays have utilized sub-optimal targets. By performing next-generation sequencing-based repeat analyses, we have identified high copy-number, non-coding DNA sequences from a series of soil transmitted pathogens. We have used these repetitive DNA elements as targets in the development of novel, multi-parallel, PCR-based diagnostic assays.
Utilizing next-generation sequencing and the Galaxy-based RepeatExplorer web server, we performed repeat DNA analysis on five species of soil transmitted helminths (Necator americanus, Ancylostoma duodenale, Trichuris trichiura, Ascaris lumbricoides, and Strongyloides stercoralis). Employing high copy-number, non-coding repeat DNA sequences as targets, novel real-time PCR assays were designed, and assays were tested against established molecular detection methods. Each assay provided consistent detection of genomic DNA at quantities of 2 fg or less, demonstrated species-specificity, and showed an improved limit of detection over the existing, proven PCR-based assay.
The utilization of next-generation sequencing-based repeat DNA analysis methodologies for the identification of molecular diagnostic targets has the ability to improve assay species-specificity and limits of detection. By exploiting such high copy-number repeat sequences, the assays described here will facilitate soil transmitted helminth diagnostic efforts. We recommend similar analyses when designing PCR-based diagnostic tests for the detection of other eukaryotic pathogens.
| With a growing emphasis on the mapping and elimination of soil transmitted helminth (STH) infections, the need for optimal and specific diagnostic methods is increasing. While PCR-based diagnostic methods for the detection of these parasitic organisms exist, these assays make use of sub-optimal target sequences. By designing assays that target non-coding, high copy-number repetitive sequences, both the limit of detection and species-specificity of detection improve. Using next-generation sequencing technology, we have identified high copy-number repeats for a series of STH species responsible for the greatest burden of disease. Using these repetitive sequences as targets in the design of novel real-time PCR assays, we have improved both the limits of detection and species-specificity of detection, and we have demonstrated this improved detection by testing these assays against an established PCR-based diagnostic methodology. Accordingly, these assays should facilitate mapping and monitoring efforts, and the generalized application of this approach to assay design should improve detection efforts for other eukaryotic pathogens.
| Estimated to infect more than one quarter of the world’s total population, the soil transmitted helminths (STH) are responsible for profound morbidity and nutritional insufficiency [1]. Concentrated in the world’s most impoverished locations, the results of widespread infection on economic capacity are equally burdensome. Yet despite the scope of such disease, and continuing efforts to improve treatment programs and integration strategies, reliable and accurate diagnosis of STH infections remains difficult, and resulting prevalence estimates remain imprecise [1–2].
In recent years, the interest in molecular diagnostic methods for the detection of gastrointestinal helminths has grown substantially. Largely, this escalation in interest has occurred in parallel with the belief that standard microscopy-based methodologies for the examination of stool samples are sub-optimal, leading to underrepresentation of infection [3–5]. Further complicating matters, rates of STH egg/larval excretion have been shown to vary considerably within sequentially collected stool samples originating from a single infected individual [6–7]. This variability in egg/larval count can result in false negative samples, particularly when non-amplification-based diagnostic methodologies are utilized [7]. Such underrepresentation of disease complicates programmatic efforts, making the accurate assessment of the effects of intervention difficult, and frequently leaving low-level infections undiagnosed [5, 8–9]. Additionally, microscopy-based diagnostic methods have been linked with pathogen misidentification due to the morphological similarities that exist between species [5, 10]. Because of such concerns, a number of conventional and real-time PCR-based assays have been developed with the objective of improving both species-specificity and limits of detection [4, 11–17]. These assays have proven valuable, and as global efforts to estimate the burden of disease caused by the soil transmitted helminths (STHs) continue to increase, the number of studies incorporating such assays has risen in response [3, 5, 9, 18–21]. To date, the target sequences for such assays have been ribosomal internal transcribed spacer (ITS) sequences, 18S or ribosomal subunit sequences, or mitochondrial genes such as cytochrome oxidase I (COI) [4, 11–14]. Ribosomal sequences have been selected as diagnostic targets because they are typically found as easily identified moderate copy number tandem repeats in nucleated organisms [22–25]. Similarly, multiple copies of mitochondrial targets are found in the vast majority of eukaryotic cells [26], making them attractive target choices. However, while effective, such diagnostic targets are often sub-optimal. This is particularly true in the case of nematodes and other multi-cellular organisms where species-specific, highly repetitive DNA elements frequently make up a substantial portion of the genome, and are often present at copy-numbers exceeding 1,000 per haploid genome [27–29]. Due to such overrepresentation, non-coding repetitive sequence elements have become the targets of choice for many PCR-based diagnostic assays for the detection of various helminth species [30–31]. However, the identification of such repeats has historically been complicated and labor intensive. This identification has relied on techniques such as restriction endonuclease digestion of genomic DNA, followed by gel electrophoresis and Sanger DNA sequencing or polyacrylamide slab gel sequencing [32–34]. However, the advent of next-generation sequencing (NGS) technologies and associated informatics tools has expedited the search for highly repetitive sequence elements [35–39], and greater confidence can be placed in the accuracy of the results of such searches. Furthermore, as ribosomal and mitochondrial sequences tend to demonstrate high degrees of conservation between species, species-specificity of detection is also improved through the targeting of unique, highly-divergent, non-coding repeat DNA elements.
Here we describe the development of a multi-parallel real-time PCR assay for the detection of five species of soil transmitted helminths (Necator americanus, Ancylostoma duodenale, Trichuris trichiura, Strongyloides stercoralis, and Ascaris lumbricoides). Using NGS-generated sequence data and the Galaxy-based RepeatExplorer computational pipeline [38–39], we have searched the genomes of each organism for highly repetitive, non-coding DNA elements in order to identify diagnostic targets capable of providing optimal limits of detection and species-specificity of detection. Using these targets to design small-volume, multi-parallel tests [4], we have created a platform that provides cost-minimizing implementation of only those assays appropriate for a specific geographic region based upon the infections present. While performing multiplex assays may provide labor and time savings in locations where many parasites are co-endemic, such assays result in considerable waste when used in areas harboring only one or a few of the target species. In such settings, the “pick-and-choose” nature of multi-parallel assays minimizes reagent waste, and by improving upon limits of detection, the species-specific platform we describe here should facilitate improved STH monitoring and mapping efforts. Since NGS-based repeat analyses allow for the selection of the most efficacious target sequences, this approach to assay design should be applied to the development of additional diagnostics tests for other eukaryotic pathogens.
For isolation of genomic DNA from N. americanus, A. duodenale, and T. trichiura, extractions were performed on cryopreserved adult worms in accordance with the “SWDNA1” protocol available on the Filarial Research Reagent Resource Center website (http://www.filariasiscenter.org/parasite-resources/Protocols/materials-1/). For N. americanus and A. duodenale, DNA extractions were conducted using a pool of approximately 10 adult worms. Both hookworm species belonged to strains originating in China. In the case of T. trichiura, extraction was performed using a single adult female worm of Ugandan origin. For S. stercoralis and A. lumbricoides, previously extracted genomic DNA was received from collaborators. S. stercoralis DNA was obtained from laboratory-reared worms originating from Pennsylvania, USA, and A. lumbricoides DNA was isolated from worms obtained from Ecuador.
For each parasite analyzed, raw sequencing reads were uploaded to the Galaxy-based RepeatExplorer web server [39]. Reads were processed according to the workflow in Fig 1, enabling the identification of high copy-number repeat DNA sequences for each organism. Promising repeat families were further analyzed using the Nucleotide BLAST tool (http://blast.ncbi.nlm.nih.gov/Blast.cgi) available from the National Center for Biotechnology Information (NCBI). Results from each organism were screened for repetitive DNA elements found to have high degrees of homology with elements of the human genome, common bacteria of the human microbiome, or other parasitic organisms likely to be found within the human gut. Had such sequences been identified as among the most repetitive, they would have been eliminated from further consideration as they would be expected to cause species-specificity challenges during downstream PCR assay development. However, no such conserved highly repetitive elements were identified. Following screening, sequences from each organism, putatively determined to be among the most highly repetitive, were utilized for further assay development (Fig 2).
Candidate primer and probe pairings for each organism, excluding A. lumbricoides, were designed using the PrimerQuest online tool (Integrated DNA Technologies, Coralville, IA), utilizing the default parameters for probe-based qPCR. The putative species-specificity of each primer pair was further examined using Primer-BLAST software (http://blast.ncbi.nlm.nih.gov/Blast.cgi). In the case of S. stercoralis the highest copy-number repeat (as determined by RepeatExplorer) was not selected as a target sequence, due to design difficulties associated with the extreme A-T richness of the repeat (A-T % = 80.25). As a result, a second repeat analysis was performed, selecting only for sequence reads with > 30% G-C content, and a second candidate sequence was selected based on these results. In the case of A. lumbricoides, RepeatExplorer analyses of two different sequencing runs performed from two distinct libraries both resulted in the identification of ribosomal and mitochondrial sequences as the most highly repetitive. For this reason, sequences from an existing, proven, primer and probe set targeting the ITS1 region were selected for further analysis [14, 16]. With the exception of the previously published A. lumbricoides probe, all probes were labeled with a 6FAM fluorophore at the 5’ end, and were double quenched using the internal quencher ZEN and 3IABkFQ (IOWA BLACK) at the 3’ end (Integrated DNA Technologies). This fluorophore-quencher combination was chosen as comparative testing of each probe revealed improved Ct values and greater ΔRn values using this chemistry when compared to typical TAMRA quenching (Fig 3). Primer and probe sets for each organism can be found in Table 1.
A panel of 79 blindly-coded patient samples, obtained in Timor-Leste as part of a previously described study [42], was tested using the newly described multi-parallel Smith assays, as well as the previously described, multiplex real-time PCR detection methodology (QIMR assay) (Table 2, S1 Table). As samples were patient-obtained and no true “gold standard” exists for the detection of the various STH infections examined here, it is difficult to definitively determine whether increased sample positivity is a result of improved assay detection limits or non-specific, off-target amplification. For this reason, the comparative performances of each assay were assessed through calculations of positive, negative, and overall agreement [43]. For the detection of N. americanus, a positive agreement (PA) of 100% and a negative agreement (NA) of 61% were calculated. This resulted in an overall agreement (PO) of 85% (Kappa 0.658). Use of the Smith assay resulted in the detection of 60 positive samples, while the QIMR assay resulted in the detection of 48 positives. All 48 QIMR-positive samples were among the 60 positive samples detected using the Smith methodology.
For the detection of A. lumbricoides, a PA of 100%, an NA of 82%, and a PO of 91% (Kappa 0.822) were seen. The Smith assay for A. lumbricoides detection resulted in the identification of 47 positive samples, while the corresponding QIMR assay resulted in 40 positives. Again, all 40 QIMR-positive samples were among the 47 Smith-positive samples which were identified.
Detection of Trichuris gave a PA of 71%, an NA of 88% and a PO of 85% (Kappa 0.580). Sample examination using the Smith assay identified 18 positive extracts, while examination with the QIMR assay identified 14 positives. However, only 10 positives were common to both assays, with 8 samples identified as positive only by the Smith assay, and 4 samples demonstrating the presence of parasite DNA using only the QIMR methodology. Amplification in control reactions demonstrated that the QIMR assay, but not the Smith assay, would provide for the detection of the closely related parasite Trichuris vulpis, a whipworm species common to canines, but also known to cause zoonotic infection [48–49]. As Trichuris ssp. including T. vulpis, Trichuris suis, and Trichuris ovis have a wide geographic distribution with increased prevalence in tropical and sub-tropical locations [50–51], the four QIMR-positive, Smith-negative samples were sequenced to determine the identity of the Trichuris species present within these samples. BLAST analysis indicated that two of the samples contained DNA from the ruminant parasite T. ovis (E values = 0.0). Unfortunately, two independent trials failed to produce usable sequence for the remaining two samples, after which both sample stocks had been exhausted, making further examination impossible.
Examination of all 79 samples for the presence of S. stercoralis resulted in the detection of only a single positive sample. This single sample was identified using both the Smith and QIMR assays. Sample examination for the presence of Ancylostoma resulted in the identification of 22 Ancylostoma ssp. positive samples using the QIMR methodology. However, not a single A. duodenale-positive sample was identified using the Smith assay. As the zoonotic parasite Ancylostoma ceylanicum has been suspected of causing human infection in Timor-Leste [52], a previously described, semi-nested PCR-RFLP assay was employed to discriminate infection with A. duodenale from infection with A. ceylanicum [47]. In this assay, an MvaI restriction digest of PCR product is indicative of the presence of A. ceylanicum, while digestion with Psp1406I is indicative of A. duodenale. 21 of the 22 Ancylostoma ssp. positive samples were digested by MvaI, identifying the infections as A. ceylanicum in origin. Two independent PCR trials (four replicates) failed to amplify the remaining Ancylostoma ssp.-positive sample, preventing a definitive determination of the identity of the parasite in that sample.
Because a sizeable panel of field-collected samples was analyzed using the two different real-time PCR methodologies discussed here, a comparison of Ct values was conducted for all samples testing positive for a given parasite by both the Smith and QIMR methods (S1 Table). All 10 samples demonstrating positive results for T. trichiura when tested by both assays showed lower Ct values using the Smith methodology (mean difference in Ct value = 7.86 +/- 2.46). Examination for N. americanus resulted in a similar pattern, with all 48 samples testing positive by both methodologies possessing lower Ct values when tested using the Smith assay (mean difference in Ct value = 4.94 +/- 1.22). In the case of A. lumbricoides, Ct values were lower using the QIMR methodology for 38 of 40 samples demonstrating positive results for both assays. However, at 0.896 +/- 0.767, the mean difference in Ct values was low. For S. stercoralis testing, only a single positive sample was identified. This sample possessed a lower Ct value when tested using the Smith assay. As no samples tested positive for Ancylostoma using the Smith assay (QIMR-positive samples were demonstrated to be A. ceylanicum), a Ct comparison could not be made.
In light of their impact on global health, the importance of optimal and species-specific diagnostic methods for the detection of soil transmitted helminths cannot be overestimated. While current molecular assays making use of ribosomal and mitochondrial targets have vastly improved the diagnosis of STH infection, these targets are frequently sub-optimal, potentially leaving low-level infections undiagnosed. Furthermore, such sequences may lack the species-specificity required to discriminate between different species of the same genus. In contrast, assays targeting high copy-number repetitive sequences improve upon assay detection limits, as many eukaryotic pathogens contain large numbers of such non-coding repeat DNA elements. Accordingly, by coupling the high throughput nature of NGS with the Galaxy-based RepeatExplorer computational pipeline, a cost effective, accurate, and expedited methodology for the identification of high copy-number repeat DNA elements was developed. Through the design of real-time PCR primer/probe pairings that uniquely target such repetitive sequences in a species-specific manner, diagnostic accuracy and limits of detection are improved dramatically when compared with microscopy-based diagnostic techniques and PCR-based diagnostics targeting mitochondrial or ribosomal sequences. Utilizing this strategy, we have successfully identified novel target sequences for the detection of N. americanus, A. duodenale, T. trichiura, and S. stercoralis. Furthermore, we have demonstrated the consistent detection of genomic DNA from each target organism at quantities of 2 fg or less, and have presented evidence to suggest improved limits of detection and species-specificity relative to an established and validated PCR diagnostic methodology [Llewellyn, 2016]. Although further testing utilizing “spiked” samples containing known quantities of eggs/larvae is currently underway, 2 fg of DNA is far less than the quantity present within a single fertilized egg or L1 larvae of each species [53–55] (Table 3). In principle, we have therefore demonstrated the potential of these assays to detect a single egg within a tested patient stool sample.
While the high copy-number nature of non-coding repetitive sequence elements makes them attractive diagnostic targets, such elements also frequently demonstrate rapid evolutionary divergence [56–57]. This divergence increases the diagnostic appeal of these sequences, as divergence reduces the risk for non-specific, off-target amplification, a characteristic essential for the development of species-specific PCR assays capable of discriminating between closely related organisms. Accordingly, while additional testing against genomic DNA from a growing panel of closely related parasites will continue to be used to evaluate the species-specificity of each selected primer/probe set, we have successfully demonstrated that each Smith assay does not amplify off-target templates from any other parasite species included within this multi-parallel platform. Furthermore, by employing a semi-nested PCR-RFLP tool, we were able to successfully demonstrate that our assay for the detection of A. duodenale does not amplify the closely related parasite A. ceylanicum. In contrast, the previously published primer/probe set employed for comparative testing was unable to distinguish between these two species, resulting in consistent off-target amplification of A. ceylanicum DNA. Similarly, while our T. trichiura assay failed to amplify four samples containing genetic material from Trichuris ssp., the comparative QIMR assay again demonstrated non-specific, off-target amplification for at least two of these samples, as sequence analysis demonstrated the presence of DNA from the ruminant parasite T. ovis. Taken together, these findings support the notion that improved assay species-specificity results from non-coding, repeat-based PCR assay design. Of note, to our knowledge, this is the first example of T. ovis potentially serving as a causative agent of zoonotic infection. However, as sheep are considered a major agricultural commodity of Timor-Leste [58], the possibility exists that individuals testing positive for T. ovis may have ingested intestinal material from an animal harboring infection, making it conceivable that the T. ovis DNA present was not the result of zoonotic infection. Given that T. ovis is not known to cause human infection, further exploration of this possible zoonosis is warranted.
Attempting to design a non-coding, repetitive DNA sequence-based assay for the species-specific detection of A. lumbricoides presented a unique set of challenges. A. lumbricoides, like many species of Ascaridae, discards large portions of its highly repetitive, non-coding genomic DNA during embryonic development. This process, known as chromosome diminution, eliminates the presence of such DNA elements from post-embryonic somatic cells [59–61]. Presumably for this reason, two separate repeat analyses, performed on two distinct library preparations, failed to identify any repetitive sequences with copy numbers greater than ribosomal and mitochondrial targets. Accordingly, a previously described primer/probe set targeting the ITS1 ribosomal region was chosen for inclusion in our multi-parallel platform [14, 16]. In order to improve diagnostics for this parasite, further analysis of A. lumbricoides using DNA extracted from eggs alone (before chromosome diminution) will be undertaken.
In addition to the potential detection limit improvements and species-specificity gains realized when diagnostically targeting non-coding repetitive DNA sequences, designing multi-parallel assays provides another unique set of advantages over previous design strategies [4]. By reducing the number of tests required, multiplex assays can provide labor and reagent savings over alternative diagnostic measures when used in environments that harbor the full complement of organisms targeted by the assay [62–63]. However, as the geographic distribution of STH species is not uniform, the use of multi-parallel assays makes it possible to select only the assays appropriate for a given location, reducing primer/probe costs associated with testing for unnecessary targets [4]. By running these assays as “small-volume” 7 μl reactions, reagent use is minimized, resulting in cost savings. Furthermore, as multi-parallel reactions are run independently, this enables the development of new assays for new pathogens and their subsequent addition to the testing platform without the complex re-optimization of assay conditions required for multiplex PCR assays.
While reagent costs associated with performing molecular diagnostic testing are higher than costs associated with conducting traditional microscopy-based diagnostics, expenses associated with molecular techniques are declining as improved reagents and enzymes have allowed reaction volumes to decrease, minimizing reagent needs [4, 64]. Furthermore, reagent improvements have increased the practicality of sample pooling, a practice already adopted by many tropical disease surveillance and diagnostic efforts [65–69]. Such pooling allows for cost-reducing high-throughput screening of stool samples [70–71]. Thus, while the total cost associated with performing a duplicate Kato-Katz thick smear under field conditions has been estimated at $2.06 [72] and we estimate the total cost associated with the duplicate testing a single stool sample using all five multi-parallel assays to be approximately $10, the pooling of as few as five samples would render small volume, multi-parallel PCR testing more cost effective than Kato-Katz testing. Furthermore, molecular diagnostic accuracy and reliability provide increased clarity of results [64], allowing for the implementation of more informed and effective treatment and control strategies. Such improvements in efficiency result in greater programmatic gains, drastically reducing long-term costs and expenses of control or elimination programs.
One profound shortcoming which hampers STH diagnostic development is the lack of a reliable gold standard for detection [8]. While still used in many clinical, mapping, and research efforts, microscopy-based methodologies are known to lack both adequate limits of detection and species-specificity of detection [3–5, 10, 64]. Similarly, while currently available molecular methods have greatly improved upon many of the shortcomings inherent to microscopy, the use of sub-optimal ribosomal or mitochondrial targets possessing relatively high degrees of conservation can result in both false-negative, and off-target, false-positive results. Thus, a gold standard of detection is sorely needed. Unfortunately, without a definitive method for assigning positive/negative status to an unknown sample, distinguishing improved limits of detection from false-positive amplification can be difficult. Nonetheless, comparative assay testing remains an important aspect of designing any diagnostic test. As such, we believe the evaluation of Timor-Leste patient samples presented in this paper provides strong evidence for improved limits of detection when utilizing the newly described Smith assays. While strain-specific genetic differences arising within divergent geographic isolates could present detection challenges, testing on a limited number of patient-derived samples from Argentina and Ethiopia aimed at providing evidence for the global applicability of these multi-parallel assays is currently underway. Additional studies to further validate these assays on a variety of geographic isolates will continue.
In all instances, and for all parasites excluding Ancylostoma and Trichuris (where off-target amplification of A. ceylanicum and T. ovis by the QIMR assay was demonstrated), each Timor-Leste patient sample that provided a positive QIMR assay result also demonstrated positivity with the corresponding Smith assay. Furthermore, all N. americanus, T. trichiura, and S. stercoralis samples that were positive by both assays exhibited lower Ct values for the Smith assay results. These findings strongly suggest improved limits of detection for the Smith assays, and support our contention that samples returning Smith assay positive results, but QIMR assay negative results, are likely low-level positives escaping detection by the sub-optimal PCR platform. This conclusion is further supported by the finding that the Smith assays do not show off-target amplification of any other STH parasites, human DNA or E. coli DNA.
As both the QIMR and Smith assays for the detection of A. lumbricoides make use of the same previously published primer/probe combination [14, 16], comparative assay testing for this parasite provided results which were more difficult to interpret. As increased reaction volumes are known to frequently improve detection limits for an assay, likely due to the large volume nature of the QIMR assay (25 μl vs. 7 μl for Smith), 38 of 40 samples returning positive results for both testing platforms demonstrated lower Ct values when examined using the QIMR method. Interestingly, despite this tendency for QIMR testing to result in lower Ct values, seven samples identified as positive using the Smith assay were found to be QIMR-negative. In contrast, not a single sample was found to be QIMR-positive and Smith-negative. As the QIMR assays are multiplexed, one explanation for this apparent contradiction is that the multiplex methodology failed to detect A. lumbricoides in a subset of samples that were positive for multiple STH parasites (S1 Table). Such failures are known to occur in multiplex reactions, particularly when primer concentrations are suboptimal, as reagents are utilized for the amplification of a more prevalent target, preventing the amplification of the lower copy-number target sequences within the sample [73]. Alternatively, while the results of our assay specificity testing present compelling evidence to the contrary, the possibility of false positive amplification cannot be definitively ruled out.
Non-coding repetitive DNA elements are found in nearly all eukaryotic organisms. Such sequences are typically highly divergent, and frequently exist in high copy-number. These characteristics make them ideal molecular diagnostic targets, particularly for the detection of pathogens such as the STHs, which remain an underdiagnosed, poorly mapped global health concern. By applying next-generation sequencing technology to the challenge of repeat DNA discovery, we have designed highly specific multi-parallel PCR assays with improved limits of detection over existing diagnostic platforms. We believe that these assays will greatly aid in the global efforts to map STH infection, facilitating accurate disease prevalence estimates. Furthermore, we intend to apply this approach to molecular target discovery of other parasitic organisms and NTDs, as optimal limits of detection and species-specificity of detection are vital to all diagnostic efforts. This is particularly true when implementing diagnostics in climates of decreasing disease prevalence. Accordingly, as NTD elimination efforts continue to progress, optimized assays will play an increasingly critical role in the detection of sporadic and focal infections and the monitoring for disease recrudescence.
|
10.1371/journal.pgen.1001288 | Pathogenic VCP/TER94 Alleles Are Dominant Actives and Contribute to Neurodegeneration by Altering Cellular ATP Level in a Drosophila IBMPFD Model | Inclusion body myopathy with Paget's disease of bone and frontotemporal dementia (IBMPFD) is caused by mutations in Valosin-containing protein (VCP), a hexameric AAA ATPase that participates in a variety of cellular processes such as protein degradation, organelle biogenesis, and cell-cycle regulation. To understand how VCP mutations cause IBMPFD, we have established a Drosophila model by overexpressing TER94 (the sole Drosophila VCP ortholog) carrying mutations analogous to those implicated in IBMPFD. Expression of these TER94 mutants in muscle and nervous systems causes tissue degeneration, recapitulating the pathogenic phenotypes in IBMPFD patients. TER94-induced neurodegenerative defects are enhanced by elevated expression of wild-type TER94, suggesting that the pathogenic alleles are dominant active mutations. This conclusion is further supported by the observation that TER94-induced neurodegenerative defects require the formation of hexamer complex, a prerequisite for a functional AAA ATPase. Surprisingly, while disruptions of the ubiquitin-proteasome system (UPS) and the ER–associated degradation (ERAD) have been implicated as causes for VCP–induced tissue degeneration, these processes are not significantly affected in our fly model. Instead, the neurodegenerative defect of TER94 mutants seems sensitive to the level of cellular ATP. We show that increasing cellular ATP by independent mechanisms could suppress the phenotypes of TER94 mutants. Conversely, decreasing cellular ATP would enhance the TER94 mutant phenotypes. Taken together, our analyses have defined the nature of IBMPFD–causing VCP mutations and made an unexpected link between cellular ATP level and IBMPFD pathogenesis.
| Inclusion body myopathy with Paget's disease of bone and frontotemporal dementia (IBMPFD) is a progressive autosomal dominant disease, characterized by the adult onset of muscle degeneration, abnormal bone metabolism, and drastic behavior changes. IBMPFD is caused by specific mutations in the highly conserved VCP gene, an ATPase known to participate in numerous cellular functions. Because of its diverse functions, it has been difficult to decipher how VCP mutations cause this debilitating disorder. To understand how these specific mutations in VCP lead to IBMPFD, we have developed a Drosophila IBMPFD model by introducing analogous mutations in TER94, the fly VCP homolog. We show that TER94 carrying these specific mutations can disrupt the fly muscle and nervous systems, similar to the symptoms of IBMPFD in humans. These phenotypic similarities suggest that information gained from our analysis of TER94 will enhance our understanding of how VCP mutations cause IBMPFD. By subjecting our fly IBMPFD model to various physiological and genetic manipulations, we have uncovered a novel link between the disease progression and cellular ATP level. Thus, in addition to establishing a fly model for further analysis of this disease, our finding should suggest new therapeutic strategies for IBMPFD.
| IBMPFD is a progressive autosomal dominant disorder, characterized by the adult onset of muscle degeneration, abnormal bone metabolism, and drastic behavior changes. This disease has been linked to mutations in VCP (known as p97 in mouse or CDC48 in yeast), a hexameric AAA (ATPase associated with various cellular activities) ATPase known to participate in numerous cellular events [1], including ubiquitinated protein processing [2], [3], homotypic membrane fusion [4]–[6], nuclear envelope reconstruction [7], and cell cycle regulation [8]. Despite these advances in understanding VCP functions, it is unclear which of the aforementioned VCP roles are critical for causing IBMPFD.
VCP protein contains an N-terminal CDC48 domain and two ATPase domains (D1 and D2). In VCP hexamers, the D1 and D2 domains of each monomer align in a head-to-tail manner [9], [10]. It is thought that ATP hydrolysis causes major conformational changes to the hexamer [11], [12], thus generating the mechanical force required for VCP function. The ATPase activity of VCP appears to be mediated mainly by D2, whereas D1 contributes to heat-induced activity [13]. In support of this, mutations disrupting the residues required for the ATP-binding and ATP-hydrolysis in D2 have been used to dominantly interfere with endogenous VCP function [14]–[16]. The N-terminal CDC48 domain has been shown to bind cofactors and ubiquitin. As VCP is implicated in numerous processes, the cooperation with different cofactors may account for its functional diversity. For instance, VCP associates with p47 in mediating homotypic Golgi membrane fusion [6], [17], with p37 in ER and Golgi biogenesis [18], and with Ufd1/Npl4 in ERAD [19] and nuclear envelope reassembly [7].
Nearly all of the VCP mutations implicated in IBMPFD are located in the N-terminal CDC48 domain, the N-D1 linker, and the D1 ATPase domain. The R155 residue in the CDC48 domain has been mutated into different amino acids (C, H, and P) in 14 familial IBMPFD cases [20]. Mutations disrupting R191 (in N-D1 linker) and A232 (linker-D1 junction) are also implicated in familial IBMPFD cases, and it has been shown that individuals carrying the VCPA232E mutation exhibited more severe symptoms [20]. It is peculiar that the D2 domain, while essential for VCP function in vitro, has not been disrupted by any of the currently known IBMPFD-causing mutations (Figure S1). It is possible that, as VCP functions as hexamers, mutations in the D2 domain will dominantly deplete the pool of functional VCP hexamers, resulting in early demises of heterozygous individuals. Alternatively, the integrity of the D2 domain may be required for VCP mutants to cause IBMPFD.
To decipher the mechanistic link between IBMPFD and VCP mutations, efforts have been made to establish animal models expressing VCP mutants. Overexpression of VCPR155H in mice caused accumulation of ubiquitinated proteins, implying that UPS is an underlying cause for IBMPFD [21]. However, recent reports showed that cells expressing VCPR155H also exhibited impaired ERAD [22] and autophagy [23], [24], indicating that the expression of IBMPFD-causing VCP mutants could hamper multiple cellular pathways. Redistribution of TAR DNA-binding protein-43 has also been implicated as a cause for VCP-induced toxicity [25]. Biochemical analysis showed that IBMPFD-causing VCP mutants have elevated ATPase activity [15], although the significance of this finding is unclear.
Drosophila contains a single VCP homolog TER94 [26], which shares ∼83% protein sequence identity with human VCP. Of the twelve known VCP amino acid substitutions in IBMPFD, nine residues are conserved in TER94 (Figure S1), suggesting that Drosophila is a suitable model for IBMPFD. Here we show that expression of TER94 carrying mutations analogous to those implicated in IBMPFD mutants could disrupt muscle integrity and cause progressive neurodegenerative defects. Genetic evidence suggests that IBMPFD-causing VCP mutations are dominant active alleles. Mutational analysis shows that the ability of TER94 to form hexamers is essential for the mutant proteins to induce neurodegenerative defects, further suggesting that the IBMPFD-causing VCP mutations are not simple loss-of-function mutations. Using reporters specific for ERAD and UPS, we found the disruptions of these pathways are unlikely to be the underlying cause for the neurodegenerative defects in our model. Instead, the TER94-dependent neurodegenerative defects correlate with cellular ATP reduction, and could be suppressed by increasing cellular ATP level and enhanced by decreasing cellular ATP level. These observations, along with earlier report that IBMPFD-linked VCP mutants have elevated ATPase activity, suggest that depleting cellular ATP is a contributing factor for VCP mutant–induced tissue degeneration.
To establish a Drosophila model for IBMPFD, we introduced amino acid substitutions at three residues in TER94: R152H, R188Q and A229E, to simulate the human VCP mutations R155H, R191Q and A232E respectively (Figure 1A). The R152H mutation is expected to affect the N-terminal CDC48 domain, whereas R188Q and A229E are located in the linker 1 (L1) region and L1-D1 junction respectively. As human versions of these Drosophila alleles have been linked to IBMPFD, they will be referred as “IBMPFD mutants” hereafter. To investigate the importance of ATPase activity in VCP function, we also generated flies expressing mutant TER94 defective in ATP-binding (K248A & K521A; K2A for short) or ATP-hydrolysis (E302Q & E575Q; E2Q for short) (Figure 1A). As disruption of the nucleotide hydrolysis cycle of protomer can affect the configuration and function of VCP hexamer [14], expression of these ATPase mutants is expected to dominantly impair the activity of endogenous TER94.
While IBMPFD affects multiple tissues, the most prevalent pathology is myopathy. To ask if our model could recapitulate this defect, we utilized 24B-GAL4 [27] (an early driver active in myoblasts) and Mhc-GAL4 (a muscle-specific driver active from larval stage onward) to examine the effect of mutant TER94 expression on muscle development and maintenance respectively. The muscle fibers, visualized using mCD8-GFP (a membrane-bound GFP), were organized in segmental fashion in wild type at late embryonic stage (Figure 1B). In 24B>TER94wt embryos, mCD8-GFP localization appeared comparable to wild type, indicating that muscle fibers developed normally up to this stage (Figure 1C). In contrast, muscle fibers in embryos expressing TER94 IBMPFD mutants appeared disorganized and loss of GFP signals was seen in some segments (Figure 1D–1F). Among the three IBMPFD mutants, there appeared to be a difference in phenotypic severity, as the disruption of muscle fiber development seemed strongest in 24B>TER94A229E, coinciding with the observation that patients bearing VCPA232E allele had more severe symptoms [20]. This difference in phenotypic severity was not caused by differences in TER94 transgene expression, as quantitative Western blots showed comparable levels of TER94 proteins in lines used in our analysis (Figure S2). Furthermore, TER94 expression from transgenes inserted at identical genomic location (generated by integrase-mediated transformation [28] to eliminate positional effect) showed that TER94A229E caused the strongest photoreceptor degeneration among the three IBMPFD mutants (Figure S3; see below). Compared to IBMPFD mutants, the phenotype in 24B>TER94K2A was even more severe, as developing muscle fibers were completely absent (Figure 1G). This strong phenotype in 24B>TER94K2A is consistent with the hypothesis that mutations disrupting the ATPase activity are more debilitating than the disease alleles.
To determine the effect of TER94 alleles on mature muscles, adult flies expressing TER94 mutants under the control of Mhc-GAL4 [29] were subjected to flight tests. Those lacking functional muscles, as caused by mutant TER94 expression, were expected to perform poorly in this simple behavior assay. Indeed, flies expressing all TER94 mutants tested showed disrupted flight behavior (Figure 1H). Interestingly, Mhc>TER94wt also exhibited impaired flight (Figure 1H), suggesting that excessive TER94 activity could disrupt muscle function. To ensure that the flightless phenotype was caused by muscle defect, phalloidin staining was performed to determine the integrity of indirect flight muscles (IFMs, Figure 1I–1L). In Mhc>lacZ control, typical pattern of myofibril with clear A-bands was seen (arrowheads in Figure 1J), indicating that IFMs were normal. In contrast, myofibril from Mhc>TER94A229E showed disrupted sarcomeres without repetitive A-bands (compare Figure 1J to Figure 1L), suggesting that the flightless phenotype was caused by structural defects in IFMs. To monitor TER94 localization in IFMs, we stained Mhc>lacZ and Mhc>TER94A229E tissues with an αVCP antibody (Cell Signaling) that could recognize Drosophila TER94 (Figure S2). Interestingly, while endogenous TER94 was localized diffusely in Mhc>lacZ, Mhc>TER94A229E myofibrils contained TER94-positive inclusion-like structures (arrows in Figure 1L), reminiscent of the rimmed vacuoles found in IBMPFD patient's muscles [1], [30].
To ensure that this flight impairment was caused by muscle degeneration, temperature-sensitive GAL80 (GAL80ts) was used to prevent GAL4 from activating TER94 expression until adulthood. Mhc>tub-GAL80ts >TER94A229E flies raised at 25°C exhibited normal flight, demonstrating that muscles had formed normally at permissive temperature. However, the same flies lost their ability to fly after being shifted to 29°C for 10 days. In comparison, such temperature shift did not affect the flight ability of Mhc>tub-GAL80ts >LacZ flies (data not shown). Together, our results suggest that expression of the TER94 IBMPFD mutants can affect both the development and maintenance of muscles.
In addition to myopathy, mutations in VCP have been implicated in frontotemporal lobar degeneration [31]. To examine the effect of VCP mutations in brain, we used elav-GAL4, a driver active in all neuronal cells, to express TER94 IBMPFD mutants. In addition, UAS-mCD8-GFP was included to label the plasma membrane of Elav-positive neurons. While mCD8-GFP localization remained normal in elav>LacZ and elav>TER94wt brains (Figure 2A and 2B), elav>TER94 IBMPFD mutant brains had a midline-crossing phenotype in the β/γ lobes of the mushroom body (arrows in Figure 2C–2E). The penetrance of this defect was lowest for elav>TER94R152H (22%, n = 18), but higher for elav>TER94R188Q (80%, n = 15) and elav>TER94A229E (62%, n = 21). The midline-crossing phenotype was seen in newly eclosed elav>TER94 IBMPFD mutant adults (data not shown), indicating that expression of TER94 IBMPFD mutants in neuronal cells may disrupt axonal guidance during brain development.
The midline-crossing phenotype observed in elav>TER94 IBMPFD mutants is reminiscent of those described for linotte, a Drosophila memory mutant [32]. We thus performed olfactory learning tests to see if elav>TER94 IBMPFD mutants had learning deficit. elav>TER94R152H, which had the lowest penetrance of midline-crossing defects, did not exhibit detectable deficit in olfactory learning. On the other hand, elav>TER94R188Q, the mutant that exhibited the highest penetrance of midline-crossing defects, had the most severe learning deficit (Figure 2F). elav>TER94A229E flies showed intermediate decline in both the midline-crossing assay and the learning tests (Figure 2F).
To further understand the effect of TER94 IBMPFD mutants on neuronal cells, we used GMR-GAL4 driver, which is active in photoreceptor cells (R cells) in the developing eye. The development and maintenance of R cells in fly eye has been a powerful model for analyzing genes contributing to human neurodegenerative diseases [33]. As shown in Figure 2G and 2H, external morphologies of GMR>lacZ (a negative control) and GMR>TER94wt eyes were normal. Phalloidin staining showed that GMR>lacZ and GMR>TER94wt retina both had normal arrangement of rhabdomeres (the light-sensing organelles in R cells) (Figure 2Gi and 2Hi), indicating the presence of normal complement of R cells. In contrast, the rhabdomere arrangement was disorganized in animals expressing TER94 IBMPFD mutants under the control of GMR-GAL4. Similar to muscle formation, expression of TER94R152H had the weakest effect on eye formation, as GMR>TER94R152H eyes actually appeared normal externally (Figure 2I). However, the rhabdomeres in GMR>TER94R152H retina were noticeably disorganized (Figure 2Ii), and many clusters contained fewer rhabdomeres, indicative of loss of R cells. On the other hand, GMR>TER94R188Q and GMR>TER94A229E both exhibited noticeable eye roughness (Figure 2J and 2K). Furthermore, the severe defect in their rhabdomere organization suggested that both GMR>TER94R188Q and GMR>TER94A229E retina also lost significant numbers of R cells (Figure 2Ji and 2Ki). In addition to disruption in rhabdomere organization in tangential sections, GMR>TER94R152H, GMR>TER94R188Q, and GMR>TER94A229E retinas all showed reduced thicknesses and disorganizations in the longitudinal sections (Figure 2Iii–2Kii). This defect further strengthens the notion that expression of TER94 IBMPFD mutants could interfere with R cell development.
To understand the effect of VCP mutations on R cell maintenance, we used Rh1-GAL4 to express TER94 IBMPFD mutants. Rh1-GAL4 becomes active in the outer R cells (those with large rhabdomeres; R1-6 in Figure 3A), but not in the inner R cells (those with small rhabdomeres; R7 in Figure 3A), at late pupal stage. Thus, this driver will allow us to circumvent the potentially detrimental effect of TER94 on development and focus on its effect on mature neurons, and ask if the neurodegenerative defect is cell autonomous. Retina from young Rh1>TER94 (in all tested transgenic alleles) adults all showed normal rhabdomere organization (Figure 3B–3E). However, in the retina of 28 day-old TER94 IBMPFD mutants, rhabdomere staining was disorganized and significant loss of R cells was evident (Figure 3H–3J). It should be noted that only the outer R cells suffered degeneration, indicating that the effect of TER94 on neurodegeneration is cell autonomous. Furthermore, the fact that older flies, but not the young ones, displayed loss of photoreceptors demonstrates that expression of TER94 IBMPFD mutants could cause progressive degeneration of R cells.
The presence of ubiquitinated inclusions, which also contain VCP mutant proteins, has been suggested as the underlying mechanism of IBMPFD pathogenesis [21], [34]. To test whether TER94 associates with aggregates in our model, retina of various genotypes were stained with αVCP antibody to monitor its localization. We reasoned that, if the formation of VCP-containing aggregates causes IBMPFD, a strong correlation between the extent of aggregate formation and the severity of neurodegenerative defects should be observed.
While rhabdomere organization was normal in young Rh1>TER94 (both wild type and IBMPFD mutants; Figure 3B–3E), scattered large structures with intense TER94 staining could be seen (arrowheads in Figure 3B–3E). Immunostaining with FK2 antibody (specific for polyubiquitin) suggests that most of these structures did not contain polyubiquitinated proteins (data not shown). In 28 day-old flies, progressive degeneration of R cells was observed, along with these TER94-containing structures (Figure 3G–3J). However, while the TER94-induced degeneration could be suppressed by increasing cellular ATP level (see below), the level of large TER94-positive structures was unaffected (compare Figure 9E to Figure 9Ei, and Figure 10D to Figure 10Di). Similarly, while R cell formation was unaffected in GMR>TER94wt retina, large TER94-positive structures were detected (arrows in Figure 2Hi). Thus, in this fly IBMPFD model, it appears that the phenotypes of large TER94-positive structures and TER94-induced neurodegeneration could be uncoupled.
It has been suggested that impaired UPS, caused by VCP mutations, results in the deposition of proteinaceous aggregates and IBMPFD [21]. To test whether the expression of TER94 IBMPFD mutants disrupts UPS, UAS-CL1-GFP (kindly provided by Dr. Paul Taylor) [35] was co-expressed with UAS-TER94A229E (the IBMPFD mutant that exhibited strongest muscle and R cell defects) in the eye discs. To demonstrate that CL1-GFP could monitor proteasome function, GMR>CL1-GFP eye discs were treated with lactacystin, a proteasome inhibitor. As shown in Figure 3K and 3L, 1 mM of lactacystin could elicit robust GFP signal, whereas DMSO alone had no effect, indicating that this reporter responded to disruption of proteasome function. Furthermore, GFP intensity increased in GMR>CL1-GFP when one copy of the 20S proteasome α1 subunit gene (Prosα1l(2)SH2342) was mutated (Figure 3N). However, although this reporter appeared to be sensitive, no GFP signal was detected in GMR>TER94A229E larval eye discs (Figure 3P). In pupal GMR>TER94A229E eye where large TER94-containing structures and R cell disruption were evident, CL1-GFP signal remained undetectable (compare Figure 3Q to Figure 3R, and Figure 3Qi to Figure 3Ri). Thus, expression of TER94 IBMPFD mutants does not appear to cause UPS impairment in our model.
Impairment of ERAD, caused by VCP mutants, has also been suggested as a mechanism for IBMPFD pathogenesis [22]. To test whether expression of TER94 IBMPFD mutants disrupts ERAD, we generated transgenic flies carrying UAS-CD3δ-YFP, a well-established ERAD reporter in cell culture and mouse [36], [37]. To test if CD3δ-YFP could be a potent ERAD reporter in fly, GMR>CD3δ-YFP eye discs were treated with 5 mM dithiothreitol (DTT), a reducing agent capable of eliciting ER stress. While no YFP signal was seen in untreated tissues, intense YFP signal was observed in DTT-treated GMR>CD3δ-YFP eye discs (Figure 4A and 4Ai). Moreover, GFP intensity increased in GMR>CD3δ-YFP when Sip3, the Drosophila homolog of a key ERAD component Hrd1, was knockdown by dsRNA-mediated RNA interference (RNAi) (Figure 4B). It is worth noting that expression of toxic polyglutamine (Q108), which does not have apparent role in ERAD, did not elicit CD3δ-YFP signals (Figure 4Bi). These data demonstrated that CD3δ-YFP, a mammalian T-cell receptor subunit, is specific and capable of detecting ERAD in Drosophila.
Consistent with the notion that TER94 has a role in ERAD, robust YFP signals were detected in GMR>TER94K2A and GMR>TER94E2Q larval eye discs (Figure 4C and 4Ci). However, in larval eye discs expressing TER94 IBMPFD mutants, the CD3δ-YFP signal was nearly undetectable (Figure 4D–4F and data not shown). Subsequent examinations of two pupal stages revealed that GMR>TER94A229E retina exhibited aberrant arrangement of R cells, whereas GMR>TER94wt had normal complement of R cells. Yet the CD3δ-YFP signals in both GMR>TER94A229E and GMR>TER94wt were comparable (compare Figure 4Fi to Figure 4Di and 4Ei, and Figure 4Fii to Figure 4Dii and 4Eii). This lack of correlation between the CD3δ-YFP signal and the photoreceptor defects suggests that impaired ERAD is not a major factor for the R cell loss.
To independently confirm this finding, we used xbp1-EGFP (kindly provided by Dr. Hermann Steller) to detect the unfolded protein response (UPR) in TER94-expressing eye discs [38]. Similar to CD3δ-YFP, robust xbp1-EGFP signal was detected when eye discs were treated with DTT (Figure S4B) or expressing TER94K2A (Figure S4G and S4Gi). In contrast, no detectable xbp1-EGFP signal was seen in GMR>TER94 IBMPFD mutants (Figure S4D, S4E, S4F). These observations suggest that overexpressing TER94 IBMPFD mutants does not trigger ERAD impairment or UPR in our fly model.
VCP is known to act as hexamer, and a recent report showed that p97R155P and p97A232E are capable of forming hexamer in vitro [15]. To ask if the ability of these IBMPFD mutants to disrupt R cells requires hexameric formation, we made monomeric forms of TER94 (mTER94) by mutating R356 and R359, two positively charged arginine residues in D1 domain, to negatively charged glutamic acids (Figure 1A). Although these mutations were chosen based on literature [39], blue native polyacrylamide gel electrophoresis (BN-PAGE) was performed to ensure that these mTER94 mutants were indeed monomers. In GMR-GAL4 extract under native conditions, αVCP antibody detected a band migrating at ∼676 kDa, corresponding to the hexameric TER94 (Figure 5A). In extracts from GMR>TER94wt, GMR>TER94A229E, and GMR>TER94R188Q, the intensity of this ∼676 kDa band became elevated, indicating that TER94 proteins made from the transgenes were capable of forming hexamers (Figure 5A). In contrast, in extracts from GMR>mTER94wt, GMR>mTER94A229E, and GMR>mTER94R188Q, the intensity of ∼676 kDa band was reduced to the GMR-GAL4 level (Figure 5A). Immunoblots of the same extracts in SDS-PAGE detected ∼95 kDa monomer in mTER94 lines with intensity that was significantly higher than endogenous protein in GMR-GAL4 control (Figure 5B), demonstrating that these mTER94 proteins were expressed, but incapable of forming hexamers. Although overexpressed mTER94 proteins were seen in SDS-PAGE, we did not detect mTER94 in the native PAGE in several attempts (Figure 5A and data not shown). It is possible that the epitope is masked in native mTER94 proteins.
To ask whether the monomeric TER94 IBMPFD mutants could still elicit neurodegenerative defects, we expressed these transgenes with Rh1-GAL4 and examined R cell organization. In young (5-day old) Rh1>mTER94A229E or Rh1>mTER94R188Q, the number and arrangement of R cells was completely normal (Figure 5C–5F). In 28 day-old flies, the R cells remained normal in Rh1>mTER94A229E or Rh1>mTER94R188Q eyes; however, the R cells were severely degenerated in eyes expressing TER94R188Q or TER94A229E (compare Figure 5Ci to Figure 5Di, Figure 5Ei to Figure 5Fi, and Figure 5G). These observations suggest the ability to form hexamers is essential for TER94-dependent R cell degeneration.
While the inheritance of IBMPFD is dominant, the nature of these VCP mutations remains unclear. To distinguish whether these mutations were dominant-active or dominant-negative, we examined the effect of altering TER94 gene dose on the eye phenotypes of TER94 IBMPFD mutants. We reasoned that if TER94 IBMPFD mutations are dominant-active, the rough eye phenotypes of GMR>TER94 IBMPFD mutants should be suppressed by inactivating one copy of the endogenous TER94 and enhanced by overexpressing wild type TER94. Conversely, if TER94 IBMPFD mutations act as dominant negatives, the rough eye phenotypes of GMR>TER94 IBMPFD mutants should be enhanced further by inactivating one copy of the endogenous TER94 and suppressed by overexpressing wild type TER94. As shown in Figure 6, the rough eye and R cell defects of GMR>TER94R152H, GMR>TER94R188Q, and GMR>TER94A229E were suppressed by inactivating one copy of TER94 and noticeably enhanced by overexpressing TER94wt transgene. To demonstrate the specificity of these genetic interactions, we tested the effect of TER94wt overexpression on GMR>TER94K2A, a known dominant negative. GMR>TER94K2A animals died at larval or early pupal stage (possibly due to GMR-GAL4 leaky expression elsewhere); however, this lethality could be rescued by overexpressing TER94wt (Figure 6J and 6Ji). Together, these results strongly suggest that IBMPFD mutations are dominant actives, and TER94 IBMPFD mutants and TER94K2A confer cytotoxicity through distinct mechanisms.
Hexamers consisted of VCP disease proteins have recently been shown to possess increased ATPase activity in vitro [15] and in vivo [40]. As IBMPFD seems to preferentially affect tissues with high energy-demand, we speculated that this elevated ATPase activity of VCP mutants might contribute to IBMPFD pathogenesis by depleting cellular ATP. To test whether TER94 IBMPFD mutants could deplete ATP, cellular ATP levels from thorax (eyes were not used because eye pigment would interfere with the ATP assay) of flies expressing TER94 IBMPFD mutants driven by hs-GAL4 driver were measured. After three cycles of induction by heat shocks, hs>TER94wt and hs>TER94R152H did not show significant reduction in ATP level when compared to hs>LacZ control (Figure 7). However, significant reduction in ATP level was seen in hs>TER94A229E and hs>TER94R188Q, the alleles associated with strong defects in muscles and photoreceptors. While this reduction in ATP level appears modest, it is worth mentioning that mouse kidney cells undergo apoptosis when cellular ATP is depleted to ∼70% of normal level [41]. Thus it is entirely possible that similar level of ATP reduction could cause muscle and R cell degeneration in flies. In any case, our results are consistent with the notion that TER94 IBMPFD mutants could alter cellular ATP level.
If this ATP depletion contributes to the neurodegenerative defects of TER94 IBMPFD mutants, it should be possible to suppress the phenotypes by boosting cellular ATP production or reducing energy consumption. To test this, we subjected TER94 flies to dietary restriction (DR), a regimen known to boost energy production by either increasing the number of electron transport chains during mitochondrial biogenesis [42], or enhancing the metabolic adaptation to reduce overall energy expenditure [43]. Freshly eclosed Rh1>TER94 flies were raised on either normal food or DR food (reduction of two-third of sucrose and yeast). In support of the notion that energy expenditure plays a role in IBMPFD pathogenesis, the progressive degeneration of R cells was markedly mitigated for all tested TER94 IBMPFD mutants raised under DR condition for 10 days (data not shown). This diet-dependent suppression of R cell degeneration was still seen in most of the TER94 IBMPFD mutants raised under DR condition for 20 days (Figure 8). The only exception was Rh1>TER94A229E, which had the most severe R cell degeneration and showed only mild response to DR condition at this age (Figure 8). This suppression appears specific to IBMPFD, as DR had no evident effect on expanded polyglutamine-induced neurodegeneration (Figure 8E–8F).
To independently verify this energy expenditure idea, we took advantage of R cell physiology to assess the role of ATP in IBMPFD. It is known that fly R cells consume at least 5-fold more ATP under illuminated condition than in the dark [44]. Freshly eclosed (<12 h) Rh1>TER94 and controls flies were raised under a normal 12 hours light/12 hours dark cycles (L/D, light intensity ∼520 lux), constant light (L/L), or completely dark (D/D) conditions, and the presence of the R cells in these flies were compared. The progressive neurodegeneration of R cells was easily seen in TER94 IBMPFD mutant-expressing flies raised under the L/D condition (Figure 9). This loss of R cells was rescued when these flies were raised under the D/D condition (compare Figure 9C to Figure 9Ci, Figure 9D to Figure 9Di, Figure 9E to Figure 9Ei, and Figure 9I). Although this darkness treatment rescued the R cell defect in all TER94 IBMPFD mutants, it failed to restore the degeneration caused by overexpressing TER94K2A (compare Figure 9F to Figure 9Fi, and Figure 9I) or two expanded polyglutamine disease models (compare Figure 9G to Figure 9Gi, Figure 9H to Figure 9Hi, and Figure 9I) [45]–[47]. This suggests that the reduction of ATP expenditure is not a universal antidote for neurodegeneration, but specific to these TER94 IBMPFD mutants. The R cell degeneration was strongly enhanced under L/L condition for all TER94 IBMPFD mutants (data not shown). However, L/L condition also caused mild R cell degeneration in control and flies expressing TER94wt, likely due to retinopathy induced by continuous light exposure [48]. In any case, data from both DR and light conditioning experiments support energy expenditure as a pathophysiological mechanism in this Drosophila IBMPFD model.
To further strengthen our energy expenditure hypothesis, we used genetic approaches to modulate ATP levels in fly IBMPFD model. As the electron transport chain of mitochondria generates the majority of ATP in animal cells, we found that RNAi knockdown of components in the catalytic core of F1 ATP synthase of complex V could significantly enhance R cells degeneration in Rh1>TER94A229E flies (Figure S5). Conversely, reducing ATP consumption through RNAi knockdown of two ATPases, ATP-citrate lyase [49] or a putative human copper transporter DmATP7 [50], could suppress Rh1>TER94A229E-induced R cell degeneration (Figure S5), further supporting the notion that ATP level is a key to the disease.
To genetically boost cellular ATP production, we took advantage of an observation that RNAi knockdown of mitochondrial phosphatase PTPMT1 (a protein tyrosine phosphatase localized to the mitochondrion) could markedly increase ATP production in mouse cells [51]. We thus expressed RNAi transgenes that target plip, the Drosophila homolog of PTPMT1, to analyze its effect on our disease model (Figure 10A). To avoid off-target effect, two independent plip-RNAi constructs v104774 and v47624, which target partially overlapped plip mRNA sequence, were used [52]. Knockdown with v104774 and v47624 both decreased the level of plip mRNA (Figure 10A) and elevated cellular ATP level (Figure 10B), although the increase in ATP level was higher with v47624. Expression of either v104774 or v47624 in outer R cells using Rh1-GAL4 had no effect on R cells (data not shown). However, expression of these plip-RNAi lines could significantly suppress the R cell degeneration in RH1>TER94R188Q and RH1>TER94A229E (compare Figure 10C to Figure 10Ci and Figure 10Cii, Figure 10D to Figure 10Di and Figure 10Dii, and Figure 10E). This suppression of TER94 IBMPFD mutants by plip-RNAi is specific, as knockdown of plip had no significant effect on the R cell degeneration in Rh1>TER94K2A or mutant MJDtr-Q78 (Figure 10E). To determine whether the suppression of TER94-dependent R cell degeneration correlates with ATP level, we measured ATP content in flies that co-express plip-RNAiv47624 with TER94R188Q or TER94A229E. Compared to TER94R188Q and TER94A229E alone, co-expressing plip-RNAiv47624 showed elevated ATP content (Figure 10F), further supporting a role of cellular ATP level in IBMPFD mutant-induced neurodegeneration.
In this study, we introduced IBMPFD-causing mutations in Drosophila TER94 to elucidate the pathogenesis of VCP mutants. Expression of these TER94 IBMPFD mutants in mature muscle cells causes loss of muscle tissues. Similarly, expression of these TER94 mutants in mature photoreceptor cells causes progressive neurodegeneration. Moreover, the observation that TER94A229E consistently had the strongest phenotypes correlates well with the allelic strength of its counterpart in human. Thus, although the fly model differs from human IBMPFD in the sense that TER94 mutant proteins are overexpressed, the fact that the expression of these TER94 mutants could recapitulate phenotypic and genetic features of IBMPFD suggests that our fly model is an appropriate system to analyze VCP mutations.
Using early muscle- and R cell-specific drivers, we show that overexpression of TER94 IBMPFD mutants during development can interfere with the formation of muscle and neuronal cells. Thus, although IBMPFD is an adult onset disease, the cytotoxic effect of IBMPFD-causing VCP mutant proteins may not be restricted to mature muscle and neuronal cells. It is likely that if expressed at high level, these VCP mutant proteins can cause deterioration of other cell types and manifest their cytotoxicity at earlier stages. Given the onset age of IBMPFD varies broadly (range 20 to >60 yrs) [20], [53], it seems plausible that subtle defects may have occurred in the development of muscle and neuronal tissues in IBMPFD individuals.
What is the nature of these IBMPFD-causing VCP mutations? The autosomal dominant inheritance of IBMPFD suggests that these mutations are not simple loss-of-function mutations. One possible scenario is that TER94 IBMPFD mutants interfere with wild type TER94 by forming non-functional hexamers. We showed that expression of TER94K2A, a known dominant negative, could disrupt muscles and R cells. However, while TER94K2A expression readily elicits ERAD and UPR responses, expression of TER94 IBMPFD mutants has little effect on these processes, suggesting that TER94K2A and TER94 IBMPFD mutants cause cytotoxicity through distinct mechanisms. Indeed, several results argue that TER94 IBMPFD alleles are dominant actives. First, disruptions of flight and R cell organization, phenotypes associated with TER94 IBMPFD mutants, could be mimicked by elevated expression of TER94wt. Moreover, IBMPFD-causing VCP mutants are known to possess elevated ATPase activity [15], [40] and we demonstrated that hexameric formation is critical for the IBMPFD mutants to disrupt photoreceptors. Most importantly, we showed that the eye defects of GMR>TER94 IBMPFD mutants could be enhanced by additional wild type TER94 expression. Taken together, these data argue strongly that IBMPFD-causing VCP alleles are gain-of-function mutations.
It is well documented that VCP participates in UPS and ERAD [19], [54]. The phenotypes of TER94 mutants defective in ATP-binding or ATP-hydrolysis also demonstrated that this AAA ATPase is indispensible for ERAD in Drosophila. The key question, however, is whether the disruption of UPS or ERAD is the cause that VCP mutations induce IBMPFD. Using reporters capable of monitoring UPS and ERAD, we showed that expression of TER94 IBMPFD mutants had little, if any, effect on both processes. The disconnect between the ability of TER94 IBMPFD mutants to induce neurodegeneration and their ability to impact UPS and ERAD is surprising because expression of pathogenic VCP mutants has been shown to disrupt ERAD and caused the accumulation of ubiquitinated aggregates in mice [21], [22], [34]. It is not clear why this difference exists between the two model systems. One possible explanation is that the reporters we used were not sensitive enough. Although we cannot formally exclude this possibility, CL1-GFP was capable of detecting a 50% reduction of gene dose in the 20S proteasome α1 subunit gene, suggesting that the UPS reporter is sensitive.
It should also be mentioned that in mammals, expression of VCP mutant does not always result in disruption of UPS and ERAD. Analysis of primary IBMPFD myoblasts (VCPH155C) and cells transfected with IBMPFD mutants did not reveal obvious increase of polyubiquitinated aggregates [55]. In cells expressing VCP mutant R155H or A232E, UbG76V-YFP and CD3δ-YFP reporters did not detect impairment to UPS and ERAD respectively [56]. Moreover, biochemical analysis showed that wild type and three disease proteins had identical binding affinity to cofactors Ufd1, Npl4, and Ataxin-3 [55]. In support of this, we have been unable to detect any genetic interaction between TER94 IBMPFD mutants and the VCP cofactors Ufd1, Npl4, and Eyc (Chang and Sang, unpublished data). It is possible that VCP mutant proteins can cause IBMPFD through multiple mechanisms, and in the Drosophila model, impairment of UPS and ERAD is not the main cause for TER94-induced R cell degeneration.
All VCP mutants implicated in IBMPFD hold increased ATPase activity [15], [40]. In addition, the symptoms of IBMPFD appear to manifest in tissues with high energy-demands. These observations prompt us to examine whether there is a mechanistic link between the TER94 mutant-induced neurodegenerative defects and the cellular ATP level. We show that expression of TER94A229E and TER94R188Q could reduce cellular ATP level. More strikingly, the neurodegenerative defects in TER94 IBMPFD mutant expressing eyes could be modified by perturbations in ATP level. Increasing ATP level by dietary restriction, dark conditions, or plip knockdown could all suppress the TER94 IBMPFD mutant-induced R cell degeneration. We further demonstrated that plip-dependent suppression of TER94A229E-induced neurodegeneration coincides with increase in cellular ATP level. Conversely, decreasing ATP level could enhance these degenerative defects. This phenotypic modulation by ATP level is specific, as these treatments had no impact on the degeneration caused by ATP-binding defective TER94 and polyglutamine expansion-containing proteins.
Several other observations lend support to this energy expenditure hypothesis. We showed that TER94 IBMPFD mutant-induced neurodegeneration requires hexameric formation, which is necessary for a functional ATPase. As the model stipulates that the ATPase activity would be essential for the pathogenesis, it makes sense that no IBMPFD-causing mutation resides in D2. Taken together, these results suggest that elevated ATPase activities may have a prominent role in IBMPFD pathology. It is imaginable that pathogenic VCP mutants may spend excessive amount of energy in performing their functions, thereby gradually deteriorating other energy-dependent pathways and causing progressive tissue degenerations.
In addition to improve our understanding of how VCP mutations cause IBMFPFD, another major goal of establishing a Drosophila IBMPFD model is to facilitate the identification of targets for designing therapeutic agents. Our demonstration that RNAi-mediated knockdown of plip could suppress TER94-induced neurodegeneration clearly suggests that this is a viable strategy. Furthermore, our demonstration that dietary restrictions and illumination conditions could modify TER94 phenotypes suggests a potential avenue of treating IBMPFD using environmental approach.
The Drosophila TER94 cDNA clone (GM02885), obtained from the Drosophila Genomics Resource Center (Indiana, USA), was subcloned into the transformation vectors pUAST and pattB-UAST as an EcoRI-XhoI fragment. Mutations in the TER94 cDNA were introduced using QuikChange (Stratagene), according to instructions from the manufacturer, and primers used for the site-directed mutagenesis of TER94 are listed in Table S1. CD3δ-YFP cDNA was purchased from Addgene [37], and subcloned into pUAST as an EcoRI-NotI fragment. All constructs were verified by sequencing prior to transgenic fly production. Flies carrying pUAST-based transgenes were generated by P element-mediated transformation, and flies carrying pattB-UAST-based constructs were generated by phiC31-mediated integration (Bestgene, California, USA).
Flies were raised in standard cornmeal food at 25°C in 12 hours light/12 hours dark cycles unless otherwise noted. 24B-GAL4 and elav-GAL4c155 were obtained from the Drosophila Stock Center (Bloomington, Indiana, USA). GMR-GAL4 has been described previously [46]. Rh1-GAL4 and UAS-mCD8-GFP stocks were originally obtained from Dr. Larry Zipursky. An enhancer trap line l(2)SH2342 that allelic to CG18495 in which encodes Proteasome α1 subunit was acquired from Szeged Drosophila Stock Center (Hungary). All transgenic RNAi lines were provided by the Vienna Drosophila RNAi Center. The UAS-CL1-GFP was a generous gift from Dr. Paul Taylor. Dr. Hermann Steller kindly provided the UAS-xbp1-EGFP stocks. Standard genetic markers and balancer chromosomes were used to generate specific genotypes.
Analyses of eye morphology using scanning electron microscopy and whole-mount preparation of fly eyes were performed as described previously [46]. Primary antibodies used were the following: anti-VCP (1∶100, Cell Signaling), anti-Elav (1∶20, Developmental Studies Hybridoma Bank, DSHB), anti-LaminDm (1∶20, DSHB). FITC, Cy3, and Cy5 conjugated secondary antibodies (Jackson ImmunoResearch Laboratories) were used at 1∶100 dilutions. F-Actin enriched rhabdomere was labeled by Rhodamine-conjugated phalloidin (Sigma). Whole-mount brain preparation was performed as described previously [57]. The mushroom body was examined as a projected image that covered the whole structure. Zeiss LSM-510 or 710 confocal microscopes were used for collecting all fluorescent images. Photoshop CS was used for image processing. For experiments that comparing fluorescent-labeled probes among different genotypes, the sample preparation and image processing was performed in the same procedure and setting.
For the learning test, the classical T-maze paradigm (Pavlovian) conditioning procedure was followed [58]. Briefly, flies were trained by exposure to electroshock paired with one odor of either 3-octanol (OCT) or 4-methycyclohexanol (MCH) for 1 minute, and subsequent exposure to the second odor (either OCT or MCH) without electroshock. Immediately after training, learning was measured by allowing flies to choose between the two odors used during training. Avoidance of the odor previously paired with electroshock produced a performance index. For each test, more than a hundred flies from each genotype were analyzed.
For measuring flight ability, a classical flightless assay was adopted [59]. 30–40 flies aged to 3 to 5 days were gently empted into a 2-liter graduated cylinder that coated with vacuum pump oil through a funnel at the top. Flies with normal flight ability were trapped in oil on the wall at higher cylinder marks, whereas flightless flies landed on the lower level or fell to the bottom. The numbers of flies stuck on the cylinder wall was counted into 10 divisions with higher index that indicates normal flight behavior and vise versa.
Protein preparation for SDS-PAGE from Drosophila heads followed the procedures described previously [46]. For the BN-PAGE experiment, the proteins extraction and separation followed manufacture's direction (Invitrogen). Primary antibodies were used as the following dilutions: anti-VCP (1∶3000, Cell Signaling) and anti-β-Actin (1∶5000, Abcam). Secondary antibodies conjugated with HRP (Jackson ImmunoResearch Laboratories) were used in 1∶5000 dilutions. All loading controls were prepared by stripping off the reagents from the original membrane and then re-immunoblotting with anti-β-Actin following the standard procedures. ImageJ was used to quantify bands intensity.
A naïve examiner who was blinded to the genotype counted the number of rhabdomeres in each unit eye. At least six individual eyes were scored in each experimental group. Data plotting and statistics were processed using Prism (GraphPad software) or SigmaPlot 10 (Systat software).
Adult eyes from GMR>LacZ, GMR>plip-RNAiv47624 and GMR>plip-RNAiv104774 were dissected for RNA isolation using TRI Reagent (Sigma). 2 µg RNA was used for reverse transcription (SuperScript First Strand, Invitrogen) following the manufacture's direction. Subsequent PCR amplification was performed with about 1.5 µg cDNA and specific primer pairs for plip (forward: 5′-CATGTTCGCACGCGTTTC-3′, reverse: 5′-GGTCATGATTTCGTCTCCAC-3′) and internal control G3PDH (forward: 5′-CCACTGCCGAGGAGGTCAACTA-3′, reverse: 5′-GCTCAGGGTGATTGCGTATGCA-3′). The amplification was done for 24 cycles (95°C 30 sec, 52°C 45 sec, 72°C 1 min).
Drosophila ATP assay was performed as following a previous report [60]. Briefly, the fly thoraces were dissected and homogenized in 6 M guanidine hydrochloride in extraction buffer (100 mM Tris and 4 mM EDTA, pH7.8) to denature endogenous ATPases. The supernatant fraction of the homogenate was collected after centrifugation at 16,100 g and then diluted in 1/750 with extraction buffer. The protein concentration was determined using the Bradford protein assay system (Bio-Rad). The ATP level was determined by ATP determination kit (Invitrogen) and measured by Victor 3 plate reader (PerkinElmer). The relative ATP level was then calculated by dividing the luminescence readout with the protein concentration and then normalized to control.
|
10.1371/journal.ppat.1002314 | Deep Molecular Characterization of HIV-1 Dynamics under Suppressive HAART | In order to design strategies for eradication of HIV-1 from infected individuals, detailed insight into the HIV-1 reservoirs that persist in patients on suppressive antiretroviral therapy (ART) is required. In this regard, most studies have focused on integrated (proviral) HIV-1 DNA forms in cells circulating in blood. However, the majority of proviral DNA is replication-defective and archival, and as such, has limited ability to reveal the dynamics of the viral population that persists in patients on suppressive ART. In contrast, extrachromosomal (episomal) viral DNA is labile and as a consequence is a better surrogate for recent infection events and is able to inform on the extent to which residual replication contributes to viral reservoir maintenance. To gain insight into the diversity and compartmentalization of HIV-1 under suppressive ART, we extensively analyzed longitudinal peripheral blood mononuclear cells (PBMC) samples by deep sequencing of episomal and integrated HIV-1 DNA from patients undergoing raltegravir intensification. Reverse-transcriptase genes selectively amplified from episomal and proviral HIV-1 DNA were analyzed by deep sequencing 0, 2, 4, 12, 24 and 48 weeks after raltegravir intensification. We used maximum likelihood phylogenies and statistical tests (AMOVA and Slatkin-Maddison (SM)) in order to determine molecular compartmentalization. We observed low molecular variance (mean variability ≤0.042). Although phylogenies showed that both DNA forms were intermingled within the phylogenetic tree, we found a statistically significant compartmentalization between episomal and proviral DNA samples (P<10−6 AMOVA test; P = 0.001 SM test), suggesting that they belong to different viral populations. In addition, longitudinal analysis of episomal and proviral DNA by phylogeny and AMOVA showed signs of non-chronological temporal compartmentalization (all comparisons P<10−6) suggesting that episomal and proviral DNA forms originated from different anatomical compartments. Collectively, this suggests the presence of a chronic viral reservoir in which there is stochastic release of infectious virus and in which there are limited rounds of de novo infection. This could be explained by the existence of different reservoirs with unique pharmacological accessibility properties, which will require strategies that improve drug penetration/retention within these reservoirs in order to minimise maintenance of the viral reservoir by de novo infection.
| In the majority of HIV-1 positive patients, antiretroviral therapy (ART) effects a sustained reduction in plasma viremia to below detectable levels. Despite this, replication competent viruses persist and fuel viremia if antiretroviral treatment is interrupted. This viral persistence stands in the way of viral eradication through ART. While this ability to persist in the face of therapy is generally considered to be attributable to a reservoir of latently infected cells, there is debate as to how this reservoir is maintained and in particular, whether there is replenishment of the reservoir by low level, residual replication. Novel antiviral agents targeting the viral integrase offer tools to explore the viral reservoirs that persist in the face of ART and we have shown that raltegravir perturbs these reservoirs as evidenced by an accumulation of episomal DNA upon rategravir intensification (Buzon et al., 2010). Through “deep sequencing” technology, we have longitudinally analyzed the genotypes of HIV episomes and integrated HIV DNA to evaluate whether they represent interrelated sequences or whether they have distinct origins. Statistical methods showed molecular compartmentalization, among and within episomal and integrated HIV-1 DNA samples, and suggest that episomal DNA in PBMC originates from a cellular/anatomic reservoir that is not revealed by sequencing of proviral DNA in PBMC in this study. These, and other data, suggest that ongoing replication, which can be blocked by adding raltegravir, occurs from proviruses that are genetically distinguishable from those detected at >1% frequency in these circulating blood cells.
| In the majority of HIV-1 infected individuals antiretroviral therapy (ART) is able to sustain suppression of plasma viral load to undetectable levels (<50 copies HIV RNA/ml plasma) for sustained intervals. However, viremia resumes if treatment is interrupted. Therefore, HIV-1 is able to persist in the face of suppressive ART. In addition low-level residual viremia has been detected with ultrasensitive assays that are able to measure down to several copies of HIV RNA/ml plasma [1], [2]. It has been suggested that low level viremia in ART-suppressed patients represents release of viral particles by long-lived latently infected CD4+ T-cells [3], [4], [5] or virions produced as a result of low-level, residual viral replication [6], [7], [8], [9], [10], [11]. The nature of this residual viremia, remains poorly understood, mainly because the very low number of virions in plasma limits its molecular characterization [3], [12], [13].
Intensification protocols employing integrase inhibitors have been used to probe the viral reservoirs that persist in ART-suppressed patients. When viral integration is inhibited, the linear viral genome, which is the precursor to the integrated provirus, is converted to episomes [14], [15]. Although sequences gleaned from episomal DNAs could be present in both productive and non-productive infections, integrase inhibition specifically results in increased episome formation and since linear cDNA is a product of reverse transcription, increases in episomal cDNA in blood cells after starting raltegravir indicates de novo infection and blocked integration. Because of the dynamic nature of episomes, they harbor a higher percentage of contemporary sequences as compared to proviral sequences that contain a higher percentage of archival sequences. Therefore, although episomes are dead-end products of viral replication, sequences contained within them will also be observed in functional viral genomes. As a consequence, characterization of the nature of episomal HIV-1 DNA during raltegravir intensification of a suppressive HAART regimen could provide new insights into the molecular diversity and compartmentalization of the viral reservoirs that persist in the face of suppressive therapy. We previously reported that raltegravir intensification of HAART-suppressed patients affected HIV-1 replication and immune dynamics in a large percentage of these patients [16], [17]. In order to gain further insight into the molecular diversity, population structure, and compartmentalization of replicative viral forms under suppressive HAART, episomal and integrated HIV-1 DNA samples were longitudinally analyzed by deep sequencing after intensification with raltegravir. We found signs of molecular compartmentalization distinguishing episomal and integrated HIV-1 DNA populations in PBMC, suggesting that proviruses in a cellular/anatomical compartment other than those cells may give rise to stochastic release of replication-competent virus during HAART.
The study sample comprised two participants from our previously reported raltegravir-intensification study [16], [17] who had plasma viremia below 50 HIV-1 RNA copies per ml for two years on stable HAART. Subjects were selected based on sample availability. Reverse-transcriptase genes from episomal and integrated HIV-1 DNA were specifically amplified and analyzed at weeks 0, 2, 4, 12, 24 and 48 following raltegravir intensification. Only viral sequences present in ≥1% of the virus population were considered for further analysis. The median number and interquartile range of episomal HIV-1 DNA clonal sequences for patients 1 and 2 was 3,063 (2,017–3,473) and 4,040 (2,106–6,131), respectively. For integrated HIV-1 DNA in patients 1 and 2, the median was 2,645 (1,570–3,592) and 2,889 (2,247–4,184), respectively.
We constructed a phylogenetic tree for each patient to assess if episomal and integrated HIV-1 DNA sequences belonged to different genetic populations or to one intermixed population. We used a neighbor-joining approach, as implemented in MEGA4 [18], to construct a phylogenetic tree for each patient with the best evolutionary model found in jModeltest v0.1.1. Phylogenetic trees did not show a clear cluster differentiation of both DNA forms (Figs. 1 and 2), suggesting a lack of population structure between both DNA samples, at least at the sequence composition level. Even when sequences do not clearly group into separate branches, statistical analysis can still reveal differences in sequence diversity between different HIV populations [12], [19], [20]. Therefore, to better assess possible compartmentalization, we performed a population structure test based on the analysis of molecular variance (AMOVA) on pairwise genetic distances and percentages of the presence of each clone in each population [21]. This test showed different ratios of population structure (FST) between episomal and integrated sequences (Table 1), which all were statistically significant (P<10−6) in both patients. As different tests of population structure can yield contradictory results, we performed a recommended conservative analysis [22] using a complementary compartmentalization test. We applied the Slatkin-Maddison test [23], which is based only on tree topology comparison. Consistent with the AMOVA test, the results of the Slatkin-Maddison analysis showed only a few migration events between episomal and integrated samples: 12 out of 55 possible migration events, and 9 out of 46 possible migration events in patients 1 and 2 respectively (P = 0.001) (Table 1). In addition, we obtained similar results when a longitudinal point-by-point comparison was performed between both DNA viral forms, except for three samples where the migration events had a high probability of being random (Table 1). Interestingly, two of these three samples were taken at the study baseline, i.e. right before HAART intensification with raltegravir. Overall, these results point to statistically significant compartmentalization between episomal and integrated DNA samples and suggest that they belong to different viral populations.
We next assessed whether longitudinal episomal and integrated DNA sequences had a temporal structure. Firstly, we constructed a separate neighbor-joining phylogenetic tree for each patient and each viral DNA form (Fig. 3 and 4). Phylogenies showed evidence of a temporal structure within episomal and integrated viral DNA forms across different time-points. Temporal structure was more evident in the episomal samples of patient 2 (Fig. 4a) and the integrated samples of patient 1 (Fig. 3b). However, a clear sign of temporal structure was difficult to observe in the remaining phylogenetic trees (Fig. 3a and 4b). Therefore, we used the AMOVA and Slatkin-Maddison tests to again assess the presence of temporal population structure within each viral DNA form. AMOVA showed that all longitudinal comparisons within episomal and integrated samples were significantly different (P<10−6) indicating different temporal population structures (Tables 2–5). Statistically significant Slatkin-Maddison results were partly consistent with those detected by AMOVA (Tables 2–5). Discrepancies between both tests might be due to the large number of sequences with the same haplotype (sometimes present in both compartments). In fact, the performance of the Slatkin-Maddison test might be limited when there is a combination of relatively short sequence, high depth and low within-patient diversity [23], as in this case.
Our results revealed that in both episomal and integrated HIV-1 DNA samples, distinct genetic populations appeared at different time-points, suggesting that the appearance of each viral DNA form in blood could be the result of stochastic mobilization of different HIV-infected cells. This effect has been observed for residual viremia [12] and for different populations of CD4+ T-cells [7]. Furthermore, we did not observe any signs of evolution across longitudinal samples within episomal DNA or within integrated viral forms (Fig. 3 and 4). We found temporal variation but no evidence of continued evolution. Of note, patient 1's viruses harbor the mutation M184I in the reverse transcriptase of the integrated HIV sequences from weeks 0 and 2 (but not from weeks 4 or 24), which is associated to resistance to lamivudine and emtricitabine. This patient was under a regimen containing tenofovir, lamivudine, lopinavir, ritonavir and raltegravir. In contrast episomal sequences from weeks 0, 2 and 4 were wild type, which suggest that 2LTR circles were generated in a different cellular/anatomical compartment that is possibly less accessible to lamivudine.
A population structure can occur at two levels: (i) different composition at the sequence level and (ii) different proportions of specific haplotypes. Therefore, the percentage of each clonal sequence of each DNA sample was represented (Fig. 3c–d and Fig. 4c–d). The results show that some haplotypes were shared among and within HIV-1 viral DNA forms, but unique haplotypes were also found. Moreover, when sequences were shared, the percentage of each haplotype was different between samples. This observation, together with the low molecular variability found in each sample (Table 1), suggests that patients under suppressive HAART have a limited variability in viral DNA sequences and that the presence and relative proportion of each clonal sequence might determine whether a population structure exists.
Previous reports have shown evidence of compartmentalization between residual plasma viremia and proviruses in fractionated and unfractionated PBMC [12]. However, this is the first time that episomal cDNA and integrated HIV-1 DNA genomes have been extensively compared. We found a statistically significant compartmentalization between episomal and integrated DNA samples in PBMC suggesting that, as with residual plasma viral RNA, episomal HIV-1 DNA forms are genetically distinct from proviral genomes and that they encompass two different genetic populations. In addition we have shown that in both episomal and integrated HIV-1 DNA samples, distinct genetic populations appeared, in a non-chronologic manner, at different time points. Longitudinal, non-chronological population structure between and within samples was detected in both circular episomal and proviral HIV-1 DNA viral forms. One explanation for our findings is that episomal and proviral sequences were generated in distinct cell types or anatomical compartments, possibly with different pharmacological penetration profiles. The detection of both DNA forms could result from stochastic mobilization from tissues to blood of a few infected cells, as their low molecular variance suggests. However, the labile nature of episomal DNA and its specific dynamics after raltegravir intensification [16] implies that the infection events that generated them occur in a pharmacologically privileged site (because the infections that generated them are occurring in the face of RT inhibitors) yet that is still accessible to raltegravir. A recent report shows that ileum may support ongoing productive infection in some patients on HAART, even if the contribution to plasma RNA is not discernible. In fact, raltegravir intensification contributed to a decrease in the cell-associated HIV RNA in this anatomic site relative to other gut sites or PBMC, suggesting that gut sites differ with respect to penetration by antiretroviral drugs, immunologic environments or the composition of CD4+ T cell populations [24]. Alternatively, our results might also suggest that cells containing these transiently-increased episomes might result from new infections with replication-competent viruses originating from rare proviruses not detectable in peripheral blood mononuclear cells, and that these episomes have had their integration blocked by raltegravir.
The lack of population structure or evolution in integrated HIV-1 genomes is best explained by the fact that proviruses are predominantly defective and archival. Although the provirus is the molecular precursor for all virions, only a very small percentage of proviruses are replication competent and only a small percentage of these would exist in a latent state- one capable of producing replication-competent virions. Therefore, temporal structure might simply be a result of continuous seeding of new cells over time. In this regard, multiple monotypic HIV-1 sequences have been observed across the uterine cervix and in blood, presumably as a result of the proliferation of cells harboring proviruses [25]. Although the origin of these cells with integrated HIV-1 DNA in our study remains unknown, forthcoming genotypic analysis across cell subpopulations in blood and tissues may cast light on this issue. Memory CD4+ T cells are thought to be a stable reservoir of HIV infection [26], [27], [28], [29], [30], [31]. The transient increases in 2LTR circles in PBMC from patients during early HAART in the absence of raltegravir, has recently been associated with the redistribution of 2LTR-enriched memory CD4+ T-cells from lymphoid tissues due to short-term decreases in immune activation [32]. In our study, changes in immune activation occurred more slowly. As such, distinct mechanisms appear to account for the changes in 2LTR circles. Interestingly, it has also recently been shown that the majority of a highly specialized subset of antigen-specific memory CD4+ T cells in mice were found to reside in a resting state in the bone marrow, surviving in close proximity to IL-7-secreting stromal cells [33]. Coincidently, IL-7 also induces homeostatic proliferation of human central memory CD4+ T cells without causing viral reactivation [34]. Therefore, the different population structure within integrated DNA, which is not coincident with the episomal DNA population, could be explained by the re-activation and mobilization of memory CD4+ T-cells from different niches in bone marrow without subsequent viral production, and could be a consequence of cellular rather than viral dynamics. Previous studies have examined rebounding viremia following treatment interruption and suggested that emerging viral variants result from the stochastic reactivation of different HIV-1 infected cells [4], [35], [36], [37]. Finally, infectious events during HAART can occur in multiple, temporary, small and locally scattered bursts [38] consistent with our observed genotypic compartmentalization. In addition, we observed that the cell-associated HIV proviral DNA remained relatively stable (Fig. S1) despite the large shifts in sequence populations observed during the study. This would not only mean that newly infected cell populations had undergone expansion but that other populations contracted at the same time. It is tempting to speculate that these dynamics might be consistent with the continuous trafficking between blood and tissues (preferentially lymphoid tissue) of stochastically activated antigen-specific memory CD4+ T-cells. We believe that it is unlikely that the results obtained reflect limited sampling, because we detected different proportions of shared haplotypes at different longitudinal time points (Fig. 1c,d and Fig. 2c,d). Sequence diversity restrictions due to sampling limitations would be reflected by either a completely different population structure or by invariable shared haplotypes in different samples. Moreover, at least in some time points, there is quite a large number of unique haplotypes (>1%), suggesting good depth.
Proviral HIV sequences are currently thought to be representative of archival HIV infection in an infected patient. Based on this hypothesis, sources of residual viremia other than CD4+ T-cells have been postulated as long-lived viral reservoirs [3]. Our observation that longitudinal detection of proviral genomes is dynamic in patients on HAART is important because it points to some limitations in the conclusions drawn from cross-sectional studies comparing HIV sequences in plasma and circulating T-cells.
Our results collectively suggest the presence of a chronic viral reservoir in which there is stochastic release of infectious virus and in which there are limited rounds of de novo infection. This could be explained by the existence of a limited cellular/anatomic reservoir in which de novo infection continues during HAART because some antiretroviral drugs do not effectively inhibit replication in this compartment. However, evidence that episomes transiently increase after raltegravir intensification suggests that this cellular/anatomic reservoir may be accessible to raltegravir, in contrast to other drugs. If proven in future work, the concept that ongoing replication during successful HAART originates from proviruses that are not detectable in peripheral blood mononuclear cells has important implications for the design of strategies aimed at viral eradication or functional cure. It indicates the need to further define the limited and covert cellular/anatomic reservoir in which ongoing HIV replication may occur during suppressive HAART.
The study was approved by the Germans Trias i Pujol hospital review board and informed consent was obtained in writing from study participants.
We extensively analyzed longitudinal samples from 2 HIV-infected patients whose plasma viral load had been suppressed to <50 HIV-1 RNA copies/ml for 2 years on a stable HAART regimen. Both patients had participated in a previously reported raltegravir-intensification study [16], [17] were intensification of a three-drug suppressive HAART regimen resulted in a specific and transient increase in episomal DNA in a large percentage of patients. The original study was designed to compare populations of episomal and integrated HIV-1 DNA and plasma viral RNA in 5 patients with detectable episomal DNA before raltegravir intensification. Although plasma viral load assays employed 7 ml of plasma, we were unable to amplify viral RNA sequences for the majority of time points nor amplify episomal and integrated HIV-1 DNA from the majority of longitudinal samples in 3 patients. For this reason, structure comparisons between episomes and integrated HIV-1 DNA were only possible for the 2 subjects shown in this study. Episome and proviral DNA dynamics for both patients are shown in Fig. S1. HAART regimens included lopinavir, ritonavir, lamivudine, tenofovir and raltegravir for patient 1 and efavirenz, emtricitabine, tenofovir and raltegravir for patient 2. None of the included patients had previously been exposed to integrase inhibitors. Peripheral blood mononuclear cells (PBMC) and plasma samples included in this study encompassed weeks 0, 2, 4, 12, 24 and 48 after raltegravir intensification.
A median of 6×107 PBMC were obtained at weeks 0, 2, 4, 12, 24 and 48 after intensification and purified by Ficoll centrifugation and resuspended in 350 µl of P1 buffer (Qiaprep miniprep kit, Qiagen). 250 µl of cell suspensions were used for extrachromosomal HIV-1 DNA extraction (Qiaprep miniprep kit, Qiagen) using a modification for the isolation of low-copy-number plasmids. Total cellular DNA was purified from 100 µl of cell resuspension with a standard protocol (QIAamp DNA Blood Kit, Qiagen) as previously described [16].
Analysis of HIV genomes from a sample containing a low copy number of HIV, such as PBMC from patients with undetectable viral load, can result in a high probability of resampling. The probability of resampling is related to both the number of target molecules during the amplification step and the number of sequenced clones. Therefore the higher the input of target molecules in the PCR and the higher the number of sequenced clones, the less likely the probability of resampling [39]. In order to avoid resampling, we extracted episomal and integrated DNA from a median of 6×107 PBMC to increase the number of input molecules during the first PCR. In addition, only samples with individual clonal sequences higher than 1,500 after deep sequencing were considered for further analysis.
We used a two-step PCR to amplify the RT region of episomal and integrated HIV-1 DNA. Primers Aluf and LA7 for integrated DNA and Jct f and LA7 for episomal DNA were used as previously described [10]. Nested PCR amplification of the RT region (codons 150 to 250) was performed as part of the DS protocol (see below). Nested PCR of background controls with primers DR pol f and DR pol r [10], in parallel to 454 amplification, was carried out to ensure that nested PCR was specific for integrated and episomal DNA viral forms.
Pooled, purified PCR products were used as template to generate a single amplicon covering codons 150 to 250 from the RT region. The amplicon library was generated in triplicate during 20 cycles of PCR amplification (Platinum Taq DNA Polymerase High Fidelity, Invitrogen, Carlsbad, CA) followed by pooling and purification of triplicate PCR products using magnetic beads (Agencourt AMPure Kit (Beckman Coulter, Benried, Germany) to eliminate primer-dimers. The number of molecules was quantified by fluorometry (Quant-iT PicoGreen dsDNA assay kit, Invitrogen, Carlsbad, CA). The quality of each amplicon was analyzed by spectrometry using a BioAnalyzer (Agilent Technologies Inc., Santa Clara, CA). Deep Sequencing (DS) was performed in-house on a 454 Life Science/Roche platform. The error rate of the in-house DS technique, as inspected with 992 pNL43 clonal sequences obtained with DS under the same conditions as those used for patient samples, was 0.07% (0.13%), which is close to previous reports [40]. This mismatch rate corresponds to a variability rate of 1.69×10-5, within the range of expected PCR error. The 99th percentile of mismatches would establish the threshold for nucleotide errors in 0.61%. Therefore, we decided to include for further analysis only patient clonal sequences present at ≥1% of the viral population.
454 DNA amplicon sequences were aligned with an HXB2 reference sequence using Muscle v3.7 [41] and an independent alignment for each DNA, time-point and patient was built. In order to increase the number of sequences for further analysis and to avoid sequencing errors produced at the end of the sequencing run, we extracted from codon 50 to 209 from each of the sequences obtained with DS. Technical errors of the DS technique drive the introduction of indeterminations (introducing N instead of A, T, G or C) into the sequences. These indetermination were substitute by gaps. An in-house method was used to merge clonal sequences into unique sequences. Only clonal sequences present at ≥1% of the clonal population were used. Alignments are available upon request.
jModeltest v0.1.1 [42] was used to infer the best phylogenetic model to explain the alignment sequence evolution. This program is able to implement a discrete gamma distribution (Γ) which models the heterogeneity rate among sites. A neighbour-joining approach, as implemented in MEGA4 [18], was used to construct a phylogenetic tree with the best evolutionary model found in jModeltest v0.1.1.
In order to detect differences in sequence composition between episomal and integrated DNA at different time-points, we performed two analyses, (i) an analysis of molecular variance (AMOVA) as implemented in the Arlequin software package [21], which is based on pairwise genetic distances and percentage of presence of each clone at each population, and (ii) a tree based topology method, the Slatkin-Maddison test [23] as implemented in HYPHY [43]. The Slatkin-Maddison tests involve estimating the number of migrations between populations, and determining whether the estimated number of migrations is less than expected if there were no compartmentalization. As the maximum possible number of migrations depends on the number of sequences analyzed from each compartment, a randomization test is performed to estimate a p value to assess the significance of compartmentalization. This approach has previously been used to study differences in sequence diversity between different HIV populations [12], [19], [20]. AMOVA is a genetic distance-based test where the frequency of a sequence variant (haplotype) i in the organ j, xij, can be expressed as xij = x + ai + bij, where ai and bij are episomal or integrated DNA and the haplotype within-DNA specific effects, respectively. These two factors have associated variances and that can be described as total variance among haplotypes as . The FST index measures the population differentiation. This value is defined as the ratio between , and it can be estimated from the usual partition of total variance into its components in a nested analysis of variance (ANOVA) [44]. We carried out AMOVA analysis by computing FST with a distance matrix obtained from Arlequin program using the best evolutionary model found by jModeltest.
|
10.1371/journal.ppat.1003350 | Global Expression Profiling of Transcription Factor Genes Provides New Insights into Pathogenicity and Stress Responses in the Rice Blast Fungus | Because most efforts to understand the molecular mechanisms underpinning fungal pathogenicity have focused on studying the function and role of individual genes, relatively little is known about how transcriptional machineries globally regulate and coordinate the expression of a large group of genes involved in pathogenesis. Using quantitative real-time PCR, we analyzed the expression patterns of 206 transcription factor (TF) genes in the rice blast fungus Magnaporthe oryzae under 32 conditions, including multiple infection-related developmental stages and various abiotic stresses. The resulting data, which are publicly available via an online platform, provided new insights into how these TFs are regulated and potentially work together to control cellular responses to a diverse array of stimuli. High degrees of differential TF expression were observed under the conditions tested. More than 50% of the 206 TF genes were up-regulated during conidiation and/or in conidia. Mutations in ten conidiation-specific TF genes caused defects in conidiation. Expression patterns in planta were similar to those under oxidative stress conditions. Mutants of in planta inducible genes not only exhibited sensitive to oxidative stress but also failed to infect rice. These experimental validations clearly demonstrated the value of TF expression patterns in predicting the function of individual TF genes. The regulatory network of TF genes revealed by this study provides a solid foundation for elucidating how M. oryzae regulates its pathogenesis, development, and stress responses.
| Rice blast disease, caused by Magnaporthe oryzae, destroys rice crop enough to feed 60 million people every year and has served as a model pathosystem for understanding host-parasite interactions. However, little is known about how M. oryzae globally regulates and coordinates its gene expression at the whole-genome scale. We analyzed the expression patterns of 206 M. oryzae genes encoding transcription factors (TFs) under 32 conditions, including infection-related developmental stages and various abiotic stresses, using quantitative real-time PCR. We focused on identifying the TF genes that are induced during the two most important infection-related morphogenetic changes; conidiation and infectious growth in rice. We identified 57 conidiation-specific TF genes and functionally characterized ten of them. Our data also showed that infectious growth in planta and oxidative stress responses in vitro involve largely overlapping groups of TFs. Comprehensive TF expression data and functional validation provided new insights into the regulatory mechanism underpinning pathogenicity and stress responses in M. oryzae. These data will also serve as a guide in studying the role of individual TF genes and the coordination of their expression in controlling development, pathogenicity, and abiotic stress responses in M. oryzae.
| Fungal pathogenesis requires well-orchestrated regulation of multiple cellular and developmental processes in response to diverse stimuli from the host and the environment. Transcription factors (TFs) function as key regulators of such processes. Identification of TF genes, which typically represent 3–6% of the predicted genes in eukaryotic genomes, has been greatly facilitated by genome sequencing [1]. High-throughput methods for gene expression analysis have enabled studies on how TF genes are globally regulated under diverse conditions [2]–[4]. A combination of these approaches has uncovered putative roles and potential interactions of TFs in animals and plants [3], [5]. Although DNA microarrays have been successfully used to study global gene expression patterns, this approach may not be sensitive enough to accurately analyze low-abundance transcripts, including those from many TF genes [6]. Quantitative RT-PCR (qRT-PCR) has been shown to be five times more sensitive than microarrays [4], serving as an effective means for accurate quantification of TF transcripts.
The rice blast fungus Magnaporthe oryzae, one of the most devastating pathogens of rice and related grass species, undergoes sequential developmental changes to successfully infect host plants and complete the disease cycles. These processes include conidiogenesis, conidial germination, appressorium formation, penetration peg formation and infectious growth. Extensive studies have been performed to identify and characterize the genes that participate in these developmental changes and pathogenicity in M. oryzae [7]–[11]. Recent functional analyses of several M. oryzae TF genes demonstrated their critical roles in processes such as conidiation (COS1, MoHOX2, MoHOX4, and COM1; [12]–[14]), appressorium formation (MoHOX7, MoLDB1, and Con7p; [12], [15], [16]), infectious growth (Mig1, Mstu1, MoHOX8, and MoMCM1; [12], [17]–[19]), oxidative stress (Moatf1; [20]), and light regulation (Mgwc-1; [21]). However, how M. oryzae TF genes are globally regulated and coordinated at the transcriptional level has not been studied. To address this knowledge gap, we analyzed expression patterns of 206 TF genes under 32 conditions, including infection-related developmental stages and various abiotic stresses, using qRT-PCR.
To test the utility of expression profiles for predicting the role of individual TF genes in development and pathogenicity, mutants of selected TF genes were characterized. The resulting data clearly demonstrated their value. All the data from this study are publicly available through the Fungal Transcription Factor Database (http://ftfd.snu.ac.kr/magnaporthe), an online platform designed to systematically identify and catalog TF genes in fungi [22].
The data extraction pipeline of FTFD identified 495 putative TF genes (4.5% of the 11,054 protein-coding genes in M. oryzae) using the InterPro terms associated with DNA-binding motifs. The proportion of TF genes in the total proteome of 23 other fungal and Oomycetes species ranged from 2.4% (Laccaria bicolor) to 6.4% (Rhizopus oryzae) (Table S1). Interestingly, 26 genes (5.3% of the TF genes) belonging to 9 different TF families appeared to be M. oryzae–specific based on the lack of orthologs in other species, which was determined using basic local alignment search tool (E<10−50) and InParanoid algorithm [23] (Table S2).
According to the InterPro classification [24], 495 M. oryzae TF genes were grouped into 44 families with the following four families dominating (Figure 1): fungal-specific Zn2Cys6 (141 genes; 28.5%), C2H2 zinc finger (89 genes; 18.0%), HMG (48 genes; 9.7%), and OB-fold (47 genes: 9.5%). Furthermore, 49 genes possessed more than one DNA-binding domains; among these, 29 of 35 homeodomain-like TF genes belonged to six different families. TFs with multiple DNA-binding domains are not unique to M. oryzae and have been detected in animals and plants [1], [25].
A few genes, such as tubulins, actins, and elongation factors, have been used as references for normalizing M. oryzae gene expression data generated using RT-PCR or qRT-PCR [12], [20], [26]–[32]. To identify the most stable reference gene under all the conditions used in our study, we evaluated seven candidate genes: β-tubulin [12], [31], [32], actin2 [20], [29], [30], glyceraldehydes-3-phosphate dehydrogenase (GAPDH) [28], cyclophilin (CYP1) [26], [27], elongation factor1β (EF1β), α-tubulin, and ubiquitin extension protein (UEP1) (Table S3).
One of the widely used methods for identifying stably expressed genes is to calculate the cycle threshold (Ct). These seven genes showed a relatively narrow range of Ct mean values across all conditions (Figure S1A and B). To evaluate the stability of gene expression, we employed the GeNorm software [33]. Under all conditions tested, these candidate genes exhibited a high degree of expression stability with relatively low M values (less than 0.1), which are far below the default limit of M≤0.15 [33] (Figure S2A). For all samples, the most stable gene was β-tubulin with M value of 0.049, indicating that β-tubulin can be used as a stable reference gene under multiple conditions (Figure S2 B). To further validate the results obtained using GeNorm, we also employed Normfinder [34] and BestKeeper [35], which showed almost identical patterns (data not shown).
We analyzed the expression patterns of 206 M. oryzae TF genes at multiple developmental stages and under various stress conditions that M. oryzae likely encounters during infection of host plants. These genes were chosen mainly based on their predicted significance and belong to 10 families, including one dominant and well-conserved family in fungi, plants, and animals (Zinc finger proteins [36]), two fungal specific families (Zn2Cys6 and APSES [37]), and those that are known to be involved in development (Homeobox [12] and bHLH [38]), cell differentiation (Myb [39]), and cell cycle (Forkhead [25]) (Table 1). The conditions analyzed included: (A) three developmental stages (conidiation, conidial germination, and appressorium formation); (B) two in planta infection stages at 78 hours post inoculation (hpi) and 150 hpi; and (C) 26 abiotic stress conditions (Table S4).
The quality of RNA samples was evaluated using two pathogenicity genes with well- known expression patterns. The expression patterns of MPG1 [40], a developmentally regulated gene, and DES1 [26], which is up-regulated in the early stage of infection and under H2O2 stress, were consistent with published data (Figure S3A and B). We analyzed the abundance of transcripts of 206 TF genes under 32 conditions, and fold changes relative to levels in vegetatively grown mycelia were calculated using the 2−ΔΔCt method [41].
Through a hierarchical clustering based on gene expression patterns, 185 of 206 TF genes were categorized into 4 groups with distinct expression patterns (Figure 2A). Group I contained 47 genes that were up-regulated preferentially at all infection-related developmental stages and under carbon (C)-starvation conditions and included the previously characterized TF gene MoHOX7, which regulates appressorium formation [12]. Genes in Group II (39), including Mgwc-1 [21], MoCRZ1 [9], and Mstu1 [18] were induced preferentially by abiotic stresses. Group III contained 63 genes that were activated mainly at 78 and 150 hpi and under C-starvation and abiotic stresses caused by methyl viologen, H2O2, MnCl2, Congo red, FeSO4, and uric acid. None of the TF genes in this group have been characterized. Group IV consisted of 36 genes that were up-regulated by abiotic stresses, but not during 3 developmental stages, and included COS1 [14] and MoHOX1 [12].
The number of TF genes with significantly altered expression varied widely depending on the conditions (Figure 2B). Most TF genes were up-regulated (>2-fold) in response to treatment with methyl viologen (191, 92.7%) and H2O2 (119, 57.8%). More than 50% of the TF genes were up-regulated during conidiation and/or in conidia (112, 54.4%), host infection at 78 hpi (139, 67.5%) and 150 hpi (141, 68.4%). In contrast, less than 20% of the TF genes were induced in response to changes in nutrient conditions (i.e., minimal medium, nitrogen starvation, and thiamine treatment) and pH (4 and 8). Under ionic stress, MnCl2 induced the expression of most genes, whereas LiCl caused the down-regulation of the 47.3% of the genes (Figure 2B). Less than 20% of the genes were down-regulated in most of the conditions tested, except conidial germination (43, 20.8%), appressorium formation (54, 26.1%), LiCl (100, 48.3%), and 4 min UV irradiation (103, 49.8%) (Figure 2B).
To identify TF genes that potentially control infection-related fungal development, we analyzed TF expression patterns during conidiation and/or in conidia, conidial germination, and appressorium formation. We identified 127 genes (61.7%) that were up-regulated during at least one of these developmental stages (Figure 3A). Expression of 70 genes was up-regulated at a single stage only: 57 (conidiation and/or in conidia), 5 (conidial germination), and 8 (appressorium formation). MoHOX2, a previously reported conidiation-specific TF gene [12], belonged to the first group. Thirty-one genes were found to be up-regulated at all three stages, and interestingly and included MGG_00021.6, a gene that is present exclusively in M. oryzae (Table S5).
To colonize host plants successfully, pathogens must overcome host-generated, defense-associated compounds such as reactive oxygen species (ROS) [42], [43]. To test the potential correlation between infectious growth in planta and oxidative stress responses, we compared the expression profiles under these conditions (Figure 3B). During infectious growth, 139 (67.5%) and 141 (68.4%) genes were up-regulated at 78 hpi and 150 hpi, respectively with 117 (71.8%) being up-regulated at both time points. Treatment with H2O2 or methyl viologen up-regulated 117 genes (71.8%), in which 61.5% of them (72) were also induced during in planta proliferation (Figure 3B). To further analyze this correlation, PCA was conducted with the data from five infection-related conditions and oxidative stresses caused by H2O2 and methyl viologen. The data from 78 hpi and 150 hpi and these oxidative stress conditions were separated from those collected during conidiation and/or in conidia, conidial germination, and appressorium formation (Figure 3C), further supporting a close relationship between infectious growth and oxidative stress responses.
To validate the functional significance of these 72 genes during infectious growth and oxidative stress responses, we retrieved mutants in four genes, ATMT4413 (MGG_06279.6, Zn2Cys6 family), ATMT0047A6 (MGG_04951.6, Zn2Cys6 family), ATMT0662D4 (MGG_04521.6, GATA family), and ATMT0334A5 (MGG_06434.6, Myb family), from a M. oryzae T-DNA insertion mutant library [44]. Compared to wild-type strain, three of the mutants (ATMT4413, ATMT0047A6, and ATMT0662D4) with an insertion upstream of the open reading frame (ORF) showed increased sensitivity to 2.5 mM H2O2 (Figure 4A). These mutants also exhibited impaired infectious growth in rice, resulting in decreased virulence. However, one mutant (ATMT0334A5), with a T-DNA insertion at the 206 bp downstream from the stop codon of MGG_06434.6, was insensitive to 2.5 mM H2O2 and was nearly identical with wild-type strain KJ201 in terms of infectious growth and virulence (Figure 4A). Because all four mutants had a T-DNA insertion outside of ORF, we hypothesized that the phenotypes observed, except that of ATMT0334A5, were most likely caused by reduced expression of the tagged genes. To test this hypothesis, we examined their expression using qRT-PCR. The level of transcripts from the disrupted gene in the mutants in ATMT4413, ATMT0047A6, and ATMT0662D4 was reduced to 60%, 20% and 50%, respectively, of the corresponding wild-type level (Figure 4B). These results supported a strong correlation between expression profiles and function and suggested the involvement of largely overlapping sets of TFs in controlling pathogenicity and ROS stress responses.
Two members of the fungal-specific APSES family, MoAPS1 (MGG_09869.6) and MoAPS2 (MGG_08463.6) (Figure S4) are up-regulated specifically during conidiation and/or in conidia (Figure 5A). Deletion of these genes (Figure S4C and S4E) caused a significant reduction in conidiation. In addition, the ΔMoaps1 and ΔMoaps2 mutants showed reduced vegetative growth (Figure 5C) and infectious growth in rice sheath cells (Figure 5D), resulting in 50% reduction in virulence. However, conidial germination and appressorium formation were normal (Figure 5B). All of the mutant phenotypes of ΔMoaps1 and ΔMoaps2 were restored by genetic complementation.
To further validate the utility of predicting functional roles based on expression profiles, we studied T-DNA insertion mutants of eight additional conidiation-specific TF genes (see Table S6). All eight mutants were defective in conidiation or conidial morphology with some additional defect in conidial germination, appressorium formation or pathogenicity (Figure S5). Conidiation of four mutants, ATMT0094A6 (MGG_06243.6, Zn2Cys6 family), ATMT0104A6 (MGG_02474.6, C2H2 family), ATMT0068B3 (MGG_01426.6, Myb family), and ATMT 0349D2 (MGG_02755.6, GATA family), was significantly reduced, and one previously reported mutant, (ATMT0651A4 (MoHOX2)) [12], did not produce any conidia. The remaining three mutants, ATMT0052B2 (MGG_06355.6, Zn2Cys6 family), ATMT0591D1 (MGG_09263.6, Zn2Cys6 family), and ATMT0034B1 (MGG_06507.6, C2H2 family), produced abnormally shaped conidia. Taken together, the phenotypes of both groups of mutants strongly support the value of expression patterns of TF genes in predicting their functions.
In M. oryzae, conidiogenesis is generally divided into four stages: (A) generation of conidiophores; (B) formation of a single-celled young conidium at the tip of conidiophore; (C) maturation of a three-celled conidium; and (D) multiplication of conidia in a sympodial manner [45]. To investigate expression patterns of these 57 genes at these stages, we collected samples at four different time points after induction of conidiation (Figure S6A). The time point at 0 h corresponded to submerged mycelial cultures in liquid CM which inhibits conidiogenesis [45], [46]. No conidia were observed at 6 h after induction of conidiation. Whereas, one to three-celled conidia were detected (5.3±3.1×104 conidia/plate) at 12 h. After 18 h, many of typical three-celled conidia were detected (26.7±1.5×104 conidia/plate). Finally, conidia were produced abundantly (756.7±20.5×104 conidia/plate) at 24 h time point (Figure S6B). These observations were illustrated in Figure S6C.
To test whether these samples were suitable for stage-specific gene expression profiling during conidiogenesis, we examined expression patterns of three well known conidiogenesis-related genes, COS1 [14], CON7 [16], and ACR1 [47]. Fold change in expression was calculated by dividing the expression level at 6 to 24 h by that at 0 h. Expression of all three genes increased during conidiation and/or in conidia. Increased COS1 transcripts were first detected at 6 h. Levels of Con7 and ACR1 transcripts increased (≥2 fold) after 12 h. In particular, the amount of ACR1 transcripts at 24 h was 17 times higher than that at 0 h (Figure S6D). These results are consistent with data in previous studies [14], [16], [47], supporting that our samples were suitable for detailed gene expression analyses during conidiogenesis.
All 57 conidiation-specific TF genes showed increased transcripts (≥2 fold) at more than one stage (Table S6). Seven genes (MGG_07319.6, MGG_00139.6, MGG_02447.6, MGG_07681.6, MGG_09263.6, MGG_01833.6, and MGG_06243.6) showed increased transcript levels at all four time points compared with that at 0 h, while 21 genes increased transcripts at only one of the time points (one gene at 6 h, one at 12 h, 10 at 18 h, and nine at 24 h). The rest of the genes had increased transcripts at two to three time points (three at 6 h, 12 at 18 h, 14 at 12 h, 18 h, and 24 h, nine at 12 and 18 h, one at 18 h and 24 h, and two at 12 h and 18 h). This data clearly showed differential expression of all 57 conidiation-specific TF genes conidiogenesis, suggesting their involvement in this process.
To investigate the regulatory network controlling the expression and interactions of these 57 genes during conidiation and/or in conidia, we examined their expression in six TF gene deletion mutants. These mutants showed conidiation-related phenotypes such as no conidial production (ΔMohox2 [12]), smaller conidia (ΔMohox4 [12]), and reduced conidial production (ΔMoaps1(this study), ΔMoaps2 (this study), ΔMoleu3 [48], and ΔMonit4 [48]). We compared gene expression profiles of these 57 genes in the six mutants with those in KJ201 to determine if and how their gene expression was affected by each mutation (Figure 6). Sixteen genes (Figure 6) were not affected by any of the mutations.
Among the remaining 41 genes, TF116 (MGG_02474.6, C2H2 family) and TF192 (MGG_03711.6, Zn2Cys6) were down-regulated in all mutants, suggesting that their expression requires the mutated genes, whereas three genes, including TF035 (MGG_07319.6, GATA type), TF220 (MGG_06243.6, Zn2Cys6), and TF269 (MGG_09829.6, Zn2Cys6), were up-regulated in all mutants. Expression of several genes were up- or down-regulated only in one mutant: TF094 (MGG_00373.6, C2H2) and TF150 (MGG_06507.6, C2H2) in ΔMohox2; TF206 (MGG_04951.6, Zn2Cys6), TF260 (MGG_09263.6, Zn2Cys6), TF231 (MGG_07131.6, Zn2Cys6) in Δ Mohox4; TF241 (MGG_07681.6, Zn2Cys6), TF246 (MGG_08094.6, Zn2Cys6), and TF268 (MGG_09825.6, Zn2Cys6) in ΔMoaps1 ;TF271 (MGG_09950.6, Zn2Cys6), MoFOK1, MoHOX3 in ΔMoAPS2; TF263 (MGG_09312.6, Zn2Cys6), TF117 (MGG_02505.6, C2H2), and MoHOX8 in ΔMonit4. In addition, expression of TF134 (MGG_02845.6, C2H2), TF008 (MGG_10837.6, bHLH), and TF276 (MGG_10528.6, Zn2Cys6) seems to require both MoHOX2 and MoHOX4, while MoHOX1 requires only MoAPS2 and is down-regulated in ΔMoaps1, ΔMoleu3 and ΔMonit4. Based on the results shown in Figure 6, we developed a model for the regulatory network controlling the expression of conidiation-specific TF genes (Figure 7).
Advances in tools for analyzing global gene expression profiles have facilitated the identification of genes potentially associated with specific processes and the characterization of regulatory networks controlling their expression. To test whether expression patterns of TF genes under diverse conditions help predict the functional roles of individual genes and potential regulatory interactions among them, we analyzed expression of 206 M. oryzae TF genes under 32 conditions using qRT-PCR. Expression profiles and functional validation of several genes selected based on their expression patterns clearly demonstrate the value of TF gene expression patterns in predicting their function. This comprehensive expression data of TF genes, publicly available through FTFD, will serve as a new community resource in analyzing the functions of and potential interactions among individual TF genes.
Previous studies based on microarrays [49], [50], SAGE [51], or RNA-seq [52] revealed many genes that potentially play important roles under specific conditions in M. oryzae. However, despite the biological significance of TF genes, relatively few have been characterized in M. oryzae and their regulation and genetic interactions have not been systematically investigated. In this study, we adopted qRT-PCR to address this deficiency. This method is labor intensive but has been shown to be robust in accurately quantifying TF transcripts [4]. We have identified differentially expressed TF genes under 32 conditions with most of them being up-regulated under at least one of these conditions (Figure 2).
Conidiation in plant pathogenic fungi, including M. oryzae, plays a central role in their life and disease cycles and epidemics. However, little is known about the molecular changes underpinning conidiation in M. oryzae. The developmental complexity of conidiation was suggested by the fact that 8.5% of the protein-coding genes in M. oryzae are differentially expressed during conidiation and/or in conidia based on a whole-genome microarray experiment [46]. Approximately 25% of the predicted genes are differentially expressed during conidiation in Neurospora crassa [53] and that ∼1,000 genes in Aspergillus nidulans are involved in conidiation [54]. Thus, it is likely that a relatively large numbers of TF genes are involved in controlling and coordinating the expression of many genes that participate in producing conidia. Our analysis revealed that more TF genes were up-regulated during conidiation and/or in conidia (112 genes) than during conidial germination (51 genes) and appressorium formation (52 genes). However, most of the genes induced during conidial germination and appressorium formation were also induced during conidiation and/or in conidia, suggesting that the same general transcription regulators probably control multiple developmental changes.
In total, 57 genes were considered conidiation-specific. These 57 genes were differentially expressed at one or more stages of conidiation, including conidiophore formation, conidia formation, and multiplication of conidia in a sympodial manner (Figure S6). The importance of many of these genes (41 out of 57) in conidiation was implied by their modified expression in one or more mutants that are defective in conidiation. Compared with the patterns observed in the wild-type strain KJ201, three genes were up-regulated while two genes were down-regulated in all the mutants during conidia production and/or in conidia. We hypothesize that these TFs act as major regulators of transcription throughout conidiation. These genes are interesting candidates for functional studies via mutagenesis. Results from this gene expression analysis in the multiple mutant backgrounds led to a model for a regulatory network controlling the expression of conidiation-specific TF (Figure 7). This model will serve as a useful roadmap in studying the regulation of conidiation.
Interestingly, most of the TF genes induced by oxidative stresses were also induced during in planta growth (72 genes, Figure 3B); this finding is consistent with the accumulating evidence suggesting that fungal pathogens must overcome plant-generated ROS for successful invasion [20], [26], [42], [55]. Our results also indicate that in vitro oxidative stress conditions mimic those that the fungus encounters in planta, and that in planta invasion and in vitro oxidative stress responses share common transcriptional regulatory factors.
Nitrogen starvation is known to be one of the important environmental cues for appressorium formation and in planta growth of M. oryzae [50]. Donofrio et al [50] reported that one GATA family TF gene, NUT1 (MGG_06050.6), was highly up-regulated in both nitrogen starvation condition and inside infected rice, suggesting NUT1 is a global nitrogen regulator. We also found that 13 TF genes were up-regulated in response to nitrogen starvation as well as during host infection (data not shown). Moreover, one of the M. oryzae specific TF gene (MGG_00021.6, Zn2Cys6) and one Myb family TF gene (MGG_06898.6) showed up-regulation at all three developmental stages, two infection stages, and nitrogen starvation, suggesting that these TF genes function as general regulators controlling multiple processes in M. oryzae.
One of the most important outcomes of this study is demonstrating the value of expression data in predicting the putative function of individual TF genes. Those TF genes induced during conidiation and/or in conidia were used to test their value. MoHOX2, which plays a critical role in conidial production [12], was identified as a conidiation-specific TF gene. Further, T-DNA insertional mutants in seven of these genes were defective in conidiogenesis. Targeted mutagenesis of two fungal-specific TF genes of the APSES family, which are up-regulated during conidiation and/or in conidia, also caused defects in conidiation. In a second test involving four mutants in the TF genes induced both during infection and under oxidative stress also showed that the mutants displayed increased sensitivity to oxidative stress and severely reduced infectious growth in rice (Figure 4A). Results from both tests strongly supported the predictive value of expression patterns in functional studies.
Considering that similar TF expression profiles were observed between in planta infectious growth and oxidative stress, a high throughput in vitro assay system that screens for mutants defective in growth under oxidative stress can serve as a surrogate platform for quickly identifying candidate pathogenicity genes. Metal ions, such as MnCl2 and FeSO4, induced expression of many TF genes. The effect of metal ions in fungal biology and pathogenicity is not clearly understood. However, a recent study suggested that ferrous ion is required for the normal function of the DES1 gene in M. oryzae [26]. In mammals, manganese ion induces apoptosis by causing endoplasmic reticulum stress and mitochondrial dysfunction [56], [57]. Comprehensive expression profiles of TF genes in the presence of metal ions or other abiotic stresses will help decipher not only how fungal responses to such stresses are controlled at the transcriptional level, but also their roles in fungal biology and pathogenicity.
Functional characterization of fungal genes requires a well-standardized platform that assays diverse phenotypes. However, only a few phenotypes, such as mycelial growth, reproduction, and pathogenicity, have been evaluated in gene functional studies with filamentous fungi [44], [58], [59]. When mutants of N. crassa in 103 TF genes were evaluated, only less than half of the resulting mutants exhibited clear phenotypes [59], which can be attributed to overlapped functions among TFs, limited phenotype assays, or a combination of both. In clusion of 26 abiotic stress conditions to profile expression patterns has helped the establishment of a novel phenomics platform for large-scale gene functional studies in M. oryzae and other pathogenic fungi. This platform will help systematically decipher the functional roles of TF genes in fungal development, pathogenicity, and abiotic stress management.
Annotated genomes of 21 fungal and 2 Oomycete species (Table S1) were used to compare of the number and types of TF genes. Putative TF genes in version 6 of the M. oryzae genome (http://www.broadinstitute.org/annotation/fungi/magnaporthe) were identified using the annotation pipeline in FTFD which annotates fungal TFs based on the InterPro database using DNA binding motifs [22]. To identify M. oryzae specific TF genes (orphan genes), a combination of BLAST matrix [60] and InParanoid algorism [23] was used. We applied a cutoff e-value of less than 10−50 for protein similarity for BLAST matrix searches and the default parameter for InParanoid.
M. oryzae KJ201 (wild-type strain) and all mutants used in this study were obtained from the Center for Fungal Genetic Resource (CFGR) at Seoul National University, Seoul, Korea. All strains were grown at 25°C for 14 days on oatmeal agar. Conidia and germinated conidia were harvested as described previously [61], and appressoria were collected 6 h after dropping a conidial suspension (5×104 conidia/ml) on a hydrophobic surface. For infected plant samples, after inoculating rice seedlings (3–4 leaf stage) with 20 ml of a KJ201 conidial suspension (1×105 conidia/ml), leaves were collected at 78 hpi and 150 hpi.
Prior to exposing fungal cultures to various types of stress, cultures of 100 ml liquid CM (complete medium) inoculated with 1 ml of a conidial suspension (5×104 conidia/ml) were incubated at 25°C for 4 days in an orbital shaker (120 rpm). The resulting mycelia were harvested using a 0.45-µm filter, washed with sterilized distilled water, transferred to fresh liquid CM and minimal medium (MM) [62] as a control, and CM or MM containing each treatment (Table S4) for 4 h culture. All mycelial samples were harvested from three replicates of three biological repeats, immediately frozen using liquid nitrogen, and stored at −80°C until processed.
For harvesting samples at different time points during conidiogenesis, a previously described procedure [46] was slightly modified. Actively growing wild-type mycelia were inoculated into liquid CM, and incubated at 25°C on a 120 rpm orbital shaker for 4 days. The resulting mycelia were fragmented using spatula and pressed through two-layers of cheese cloth. The mycelia were collected using two-layers of miracloth (Calbiochem, California, USA) and washed three times with one liter of sterilized distilled water. After resuspending the harvested mycelia in 10 ml sterilized distilled water, 400 µl of the suspension was spread on each 0.45 µm pore cellulose nitrate membrane filter (Whatman, Maidstone, England) placed on V8-Juice agar plate. The plates were incubated at 25°C with constant light. The whole tissue on the membrane filters was collected at 0 h, 6 h, 12 h, 18 h, and 24 h after inoculation by disposable scraper (iNtRON Biotechnology, Seoul, Korea). All samples were harvested from three replicates of three biological repeats, immediately frozen using liquid nitrogen, and stored at −80°C until processing.
Total RNA was extracted using an Easy-Spin Total RNA Extraction Kit (iNtRON Biotechnology, Seoul, Korea), and 5 µg of RNA was reverse-transcribed to cDNA using the Prom-II Reverse Transcription System (Promega, Madison, WI, USA) according to the manufacturer's instructions. The resulting cDNA preparations were diluted to 12.5 ng/µl and kept at −20°C. A total of 206 primer pairs were designed using the 3′-end exon region of the target genes (GC contents = 45–55% and Tm = 60) (Table S7).
qRT-PCR reactions were performed using a MicroAmp Optical 96-Well Reaction Plate (PE Biosystems, Foster City, CA, USA) and an Applied Biosystems 7500 Real-Time PCR System. Each well contained 5 µl of Power 2× SYBR Green PCR Master Mix (Applied Biosystems, Warrington, UK), 2 µl of cDNA (12.5 ng/µl), and 15 pmol of each primer. The thermal cycling conditions were 10 min at 94°C followed by 40 cycles of 15 s at 94°C and 1 min at 60°C. All amplification curves were analyzed with a normalized reporter threshold of 0.1 to obtain the threshold cycle (Ct) values.
To identify an appropriate reference gene for normalizing the expression levels of individual TF genes, GeNorm v.3.4 [33], Normfinder [34] and BestKeeper [35] were used. Expression levels of the chosen reference gene, β-tubulin, were measured in more than two replicates for each PCR run, and their average Ct value was used for relative expression analyses.
To compare the relative abundance of target gene transcripts, the average Ct value was normalized to that of ß-tubulin for each of the samples as 2−ΔCt, where −ΔCt = (Ct of the target gene – Ct of ß-tubulin). Fold changes of transcripts in samples representing developmental stages and infectious growth relative to those in mycelial samples in liquid CM were calculated as 2−ΔΔCt, where −ΔΔCt = (Ct of the target gene –Ct of ß-tubulin) test condition - (Ct of the target gene – Ct of ß-tubulin) CM [41]. qRT-PCR was conducted twice with three replicates, and all data are presented. The fold changes of transcripts from various stress-exposed mycelial samples compared to those in untreated samples (CM or MM) were calculated as 2−ΔΔCt, where −ΔΔCt = (Ct of the target gene –Ct of ß-tubulin) treated condition - (Ct of the target gene – Ct of ß-tubulin) untreated condition.
Pearson's correlation coefficient and Spearman's rank were used to measure the similarity between gene expression profiles and the similarity between samples, respectively. A heat map of the clustered genes and samples was generated by complete linkage. A principle component analysis (PCA) was conducted to reduce the dimensions and to understand the relationships between the TF genes and the experimental conditions. PCA was performed using SPSS software v.12.0 (SPSS Inc., Chicago, IL, USA).
To build a model for the regulatory network controlling the expression of conidiation-specific TF genes based on their expression patterns in six TF gene deletion mutants, we used NodeXL (http://nodexl.codeplex.com).
Assays for measuring the sensitivity to exogenous oxidative stress were performed on CM agar amended with 2.5–5 mM H2O2 or methyl viologen. Radial colony growth was measured on day 6 after inoculation. Infection assays with rice sheath and 3-week-old rice seedlings were conducted as described previously [63].
Gene disruption (Fig. S4B and D) and fungal transformation were conducted as described previously [61]. Putative mutants were confirmed by Southern blot analysis. Vegetative growth, pigmentation, conidiation, conidial size, conidial germination, appressorium formation, and infection assays on onion epidermis, rice sheath cells, and rice seedlings were conducted as described previously [12], [63].
|
10.1371/journal.pcbi.1003196 | Causes and Consequences of Hyperexcitation in Central Clock Neurons | Hyperexcited states, including depolarization block and depolarized low amplitude membrane oscillations (DLAMOs), have been observed in neurons of the suprachiasmatic nuclei (SCN), the site of the central mammalian circadian (∼24-hour) clock. The causes and consequences of this hyperexcitation have not yet been determined. Here, we explore how individual ionic currents contribute to these hyperexcited states, and how hyperexcitation can then influence molecular circadian timekeeping within SCN neurons. We developed a mathematical model of the electrical activity of SCN neurons, and experimentally verified its prediction that DLAMOs depend on post-synaptic L-type calcium current. The model predicts that hyperexcited states cause high intracellular calcium concentrations, which could trigger transcription of clock genes. The model also predicts that circadian control of certain ionic currents can induce hyperexcited states. Putting it all together into an integrative model, we show how membrane potential and calcium concentration provide a fast feedback that can enhance rhythmicity of the intracellular circadian clock. This work puts forward a novel role for electrical activity in circadian timekeeping, and suggests that hyperexcited states provide a general mechanism for linking membrane electrical dynamics to transcription activation in the nucleus.
| Daily rhythms in the behavior and physiology of mammals are coordinated by a group of neurons that constitute the central circadian (∼24-hour) clock. Clock neurons contain molecular feedback loops that lead to rhythmic expression of clock-related genes. Much progress has been made in the past two decades to understand the genetic basis of the molecular circadian clock. However, the relationship between the molecular clock and the primary output of clock neurons—their electrical activity—remains unclear. Here, we explore this relationship using computational modeling of an unusual electrical state that clock neurons enter at a certain time of day. We predict that this state causes high concentration of calcium ions inside clock neurons, which activates transcription of clock genes. We demonstrate that this additional feedback promotes 24-hour gene expression rhythms. Thus, we propose that electrical activity is not just an output of the clock, but also part of the core circadian timekeeping mechanism that plays an important role in health and disease.
| The conventional theory of neuronal information processing is based on action potential (AP) firing [1], [2]. While signaling through APs is a ubiquitous form of neuronal communication throughout the nervous system, it is not the only mechanism through which neurons may signal. In particular, neurons that receive input that induces large inward currents (hyperexcitation) may display depolarization block, and be unable to fire APs due to voltage–gated sodium channel inactivation. For example, antipsychotic drugs can trigger depolarization block in midbrain dopamine neurons [3]–[5]. Depolarization block is also a feature of many mathematical models of neuronal dynamics [6], [7], including the canonical Hodgkin-Huxley [8] and Morris-Lecar [9] models. Furthermore, large inward currents can induce depolarized electrical states with low amplitude membrane oscillations (DLAMOs). Such depolarized states occur in intrinsically photosensitive retinal ganglion cells in the presence of bright light [10]. DLAMOs and depolarizing block occur spontaneously—not as the result of external stimulation—in neurons of the hypothalamic suprachiasmatic nuclei (SCN) [11]–[13]. Spontaneous depolarization block has also been reported in cerebellar nuclear neurons [14], [15]. These various depolarized states add complexity to the repertoire of neuronal communication.
In the SCN, which function as the central mammalian circadian (∼24-hour) pacemaker [16], depolarization block and DLAMOs occur only in a subset of SCN neurons and mostly during the latter half of the day [13]. Here, we study the ionic mechanisms that underlie these depolarized states in SCN neurons. We also seek to determine general principles for how neurons can spontaneously enter such states and the physiological role(s) they may play. Electrical activity of SCN neurons is not only important for sending timekeeping signals to other cells, but also for transmitting information from the external world to their intracellular molecular circadian clocks [17]. Thus, understanding the electrophysiology of SCN neurons is essential for understanding circadian timekeeping in mammals [18], and may also yield general insights into mechanisms for signaling from synapse to gene [19].
Our approach uses mathematical modeling in combination with experimental validation. We find that DLAMOs are caused by the interplay of L-type calcium and calcium-activated potassium (KCa) currents. During depolarized states, we predict that intracellular calcium concentration reaches high (but physiological) levels. We propose that these daily elevated calcium levels activate clock gene transcription during the day, which in turn increases the expression of KCa and potassium leak currents to hyperpolarize the membrane at dusk and night. We show that this additional feedback loop between membrane excitability and gene expression can promote rhythmicity of the intracellular circadian clock.
Our new computational model of a SCN neuron extends the model of Belle et al. [13] by incorporating a L-type calcium current, a KCa current, and intracellular calcium dynamics. The model is able to produce repetitive firing of APs in the absence of externally applied current (Iapp = 0), consistent with the spontaneous firing behavior of SCN neurons (see Brown and Piggins [18] for review). As in the original version of the model [20], the periodic solutions modeling repetitive AP firing arise when the quiescent steady state becomes unstable due to Hopf bifurcation [21]. The improved model more closely replicates the biophysical properties of SCN neurons as individual APs are now followed by an appropriate after hyperpolarization (AHP). The AHP amplitude is mediated by KCa currents and partly regulates the daily pattern of action potential firing frequency observed in SCN neurons [22], [23]. Another feature of SCN neurons are low amplitude membrane oscillations in the presence of TTX [11], [12]. Since these TTX-induced oscillations (TTXLAMOs) are seen in dissociated SCN cells [24], we concluded that they are intrinsic and unmasked by the post-synaptic effects of TTX. In the original version of the model [13], [20], simulation of the post-synaptic effects of TTX (setting sodium conductance (gNa) to 0) did not produce these oscillations, but instead resulted in a steady-state voltage (data not shown). In contrast, in our revised model, calcium and potassium currents interact to produce oscillations in the absence of sodium current (Figure 1B), consistent with the TTXLAMOs previously reported in the literature (Figure 1A). Additionally, in agreement with experimental data, simulating the application of the L-type calcium channel blocker nimodipine abolished TTXLAMOs (Figure 1A,B). An interesting feature seen both in the model and experimental data was that blocking calcium channels leads to a depolarization rather than a hyperpolarization of SCN cells. This results from a reduction in the amplitude of KCa currents, an important parameter that determines the resting membrane potential (RMP) of some SCN neurons [13].
A subpopulation of SCN neurons, specifically those expressing detectable levels of the Period1 gene (Per1::eGFP+ve neurons), are at a depolarized RMP during much of the afternoon [13], and can show spontaneous DLAMOs. In our model, reducing KCa conductance (gKCa) can depolarize the membrane and transition a cell from generating APs to DLAMOs (Figure 2B), consistent with the effect of KCa channel blockers on SCN neurons in vitro (Figure 2A). We hypothesized that DLAMOs occur through a similar mechanism to TTXLAMOs. According to this hypothesis, calcium currents would be larger than sodium currents in the depolarized states since most of the TTX sensitive sodium channels will be inactivated. Thus, we expected that application of TTX would have little effect on DLAMOs, whereas application of nimodipine would inhibit these oscillations [12].
We tested this hypothesis by simulating a neuron in a state producing spontaneous DLAMOs (Figure 3A). When the post-synaptic effect of TTX application was simulated (sodium conductance (gNa) set to 0), in neurons that otherwise would show DLAMOs, very little change in the neuron's behavior was seen. However, simulation of nimodipine application (L-type calcium conductance (gCaL) set to 0) abolished all oscillations. To test this experimentally, we recorded from 42 Per1::eGFP+ve neurons that spontaneously exhibited DLAMOs during the projected day. Validating the model's predictions, we find that application of TTX had only a subtle effect on the oscillations, while nimodipine application abolished them (Figure 3B). This provides pharmacological evidence that DLAMOs require L-type calcium current, similar to the membrane oscillations seen in the presence of TTX [12], [24].
The fact that DLAMOs require L-type calcium current suggests that they may be sensitive to factors affecting calcium homeostasis in SCN neurons, since inactivation of L-type current is primarily calcium-dependent [25]. To explore this using our model, we simultaneously varied gKCa and the basal calcium concentration near the membrane (bs). Mathematically, the transition to DLAMOs occurs through a Hopf bifurcation from a depolarized steady state. For a given basal calcium concentration, DLAMOs are seen once gKCa exceeds a minimal value (Figure S1). As basal calcium is increased, less gKCa is required for DLAMOs to be seen (Figure S2). However, if gKCa is too small, then a Hopf bifurcation does not occur—increasing bs instead leads to saddle-node bifurcation and a transition from the depolarized steady state to a hyperpolarized steady state (Figures S3 and S4). Thus, the onset of DLAMOs in a population of SCN neurons may be heterogeneous, in accordance with heterogeneity in the balance of calcium homeostasis and KCa channel expression across cells. These observations lead to the following testable predictions. If a SCN neuron is in a depolarized steady state, raising extracellular Ca2+ to a sufficient level may induce DLAMOs. Once the cell is exhibiting DLAMOs, raising extracellular Ca2+ further will induce a hyperpolarized steady state. If elevating extracellular Ca2+ does not have the predicted effects, it is an indication that KCa channel expression is very low in that cell. In such a cell, application of a synthetic KCa channel opener such as NS004 or NS11021 [26] should initiate DLAMOs.
To better understand the ionic currents underlying the electrical behaviors of SCN neurons, we first considered the contribution of sodium (INa), calcium (ICa), potassium (IK) and calcium-activated potassium (IKCa) currents during simulated AP firing (Figure 4A). The model AP is characterized by rapid activation of INa, quickly followed by activation of IK and inactivation of INa. Likewise, activation of IKCa follows activation of ICa, but these currents are slower and lower in magnitude. Since the simulated ionic currents closely resembled the currents measured during AP clamp experiments [24], we then used the model to predict the ionic contributions driving the low amplitude membrane oscillations observed in SCN neurons. We found that both TTXLAMOs and DLAMOs occurred via a balance between calcium, potassium, and calcium-activated potassium currents (Figure 4B,C). Calcium current activation preceded the opening of potassium currents and dominated the rising phase of the oscillations. As calcium enters the cell it activates KCa channels, which contribute to the falling phase of the oscillations. In simulations of TTXLAMOs, sodium currents were set to zero to reflect the post-synaptic effect of TTX. In DLAMOs, the sodium current was naturally minimal due to the depolarized state of the neuron.
A key difference between TTXLAMOs and DLAMOs was the mean calcium current. In TTXLAMOs, the mean calcium current was small and at times near zero (Figure 4B). However, in DLAMOs, the mean (and minimal) calcium current was substantially different from zero (Figure 4C). We sought to predict the effect of this increased calcium influx on intracellular calcium levels. Our minimal model for intracellular calcium (Cac) dynamics (see Materials and Methods) requires the estimation of two parameters: a clearance rate (1/τc) and a factor for converting calcium current to concentration (kc). Since the intracellular calcium concentration ([Ca2+]i) is typically tightly regulated, we hypothesized that in order to avoid accumulation of calcium ions in the cytosol, the rate of calcium clearance would be higher at depolarized membrane potentials. To test this hypothesis, we measured [Ca2+]i in ten SCN neurons following 200 ms duration depolarizing voltage steps from a holding potential of −60 mV, and estimated τc by fitting [Ca2+]i to an exponential decay. The means of the peak [Ca2+]i after voltage steps to −20 mV (158.9±28.0 nM, mean ± SEM), 0 mV (440.8±130.5 nM), and +20 mV (734.8±232.9 nM) were not equal (one-way ANOVA, p = 0.046), and peak [Ca2+]i was significantly higher in the +20 mV group than in the −20 mV group (Tukey's HSD test, α = 0.05) (Figure 5A, left). However, we found no significant differences in τc between any of the three voltage steps (2.24±0.52 s, 1.75±0.19 s, 1.80±0.21 s for the −20, 0, and +20 mV groups respectively, one-way ANOVA, p = 0.543) (Figure 5A, right). We then set τc = 2.24 s, and chose kc based on previously reported measurements of Δ[Ca2+]i in SCN neurons evoked by a series of brief depolarizing pulses (Figure 6E of Irwin and Allen [27]) and during AP firing (Figure 4C of Irwin and Allen [27]). Figure 5C shows that our model produces similar changes in intracellular calcium during a train of spontaneous APs as seen in the experimental data (Figure 5B).
Next, we compared the intracellular calcium levels predicted by the model during quiescence (−65 mV RMP), extended AP firing, and DLAMOs. The model predicts that DLAMOs induce a much greater increase in steady-state intracellular calcium (ΔCac>290 nM) than AP firing at 6 Hz (ΔCac<55 nM) and 12 Hz (ΔCac<105 nM) (Figure 5D). Although intracellular calcium levels in hyperexcited clock neurons have yet to be reported, this prediction is consistent with the findings of two related studies. Firstly, Irwin and Allen [27] show that [Ca2+]i increases exponentially with increased membrane potential (ranging from −80 to −40 mV) in SCN neurons. Their data suggest that voltage steps to depolarized membrane potentials increase intracellular calcium more than firing action potentials for a similar duration. Secondly, Pennartz et al. [12] measured inward calcium currents in SCN neurons during sustained (>300 ms) depolarizing voltage steps. At −30 mV, the calcium current appears to be nearing a steady state of 40 pA or greater. This calcium current is larger than the calcium current predicted by our simulations during the peak of a DLAMO (33 pA). Taken together, these data provide strong, albeit indirect, evidence supporting our prediction of elevated intracellular calcium levels during DLAMOs.
Elevation of intracellular calcium levels may play an important role in the rhythmic gene expression that constitutes the molecular circadian clocks within SCN neurons. A major phase-shifting and entrainment pathway for these clocks involves CREB-dependent activation of Per1 and Per2 transcription [28]. Dolmetsch et al. [29] reported that for CREB activation in cortical neurons, [Ca2+]i levels in excess of 400 nM are required. In our simulations, such high calcium levels are achieved only during hyperexcited states.
To explore the relationship between hyperexcitation and the intracellular circadian clock, we integrated our model of SCN neuron excitability with a simple model of gene regulation based on the Goodwin oscillator [30]. In this model, a clock gene is transcribed into mRNA (M), the mRNA is translated into protein (P), the protein is phosphorylated (P*), and the phosphorylated protein binds to an enhancer (E-box) in the promoter region of the clock gene, inhibiting its transcription (see Figure 6A for a model schematic and Materials and Methods for the model equations). This negative feedback loop can lead to oscillations if one assumes positive cooperativity among a large number of molecules (n>8) in repression of the E-box [31]. Here, we set n = 4, and show through simulation that this model does not produce oscillations with the chosen parameters (Figure 6B). We then extended the model to incorporate SCN membrane excitability as diagrammed in Figure 6C. We assume there is another gene product (R) under the control of the same E-box as the clock gene, and that R downregulates the activity of potassium channels (specifically KCa and potassium leak currents) in the membrane. This is motivated by known circadian rhythms in the expression of these channels ([32]–[34], see Colwell [35] for review). These currents hyperpolarize the membrane potential (V), which closes voltage-gated calcium channels and reduces the inward calcium current. This affects intracellular calcium concentration (Cac), which regulates clock gene transcription through cAMP response elements (CREs). The extended model produces ∼24-hour oscillations in M (Figure 6D) using the same set of parameters (including n = 4) that did not produce oscillations in the basic gene regulation model (Figure 6B). The oscillations are enabled by the additional feedback present in the extended model: since the phosphorylated clock protein inhibits R, increasing P* leads to an up-regulation of potassium currents, membrane hyperpolarization, and less intracellular calcium, thus ultimately decreasing clock gene transcription. On the other hand, decreasing P* leads to down-regulation of potassium currents, membrane depolarization, and more intracellular calcium, thereby increasing clock gene transcription. These results suggest that electrical activity, and in particular hyperexcited states, are more than just an output signal of the intracellular clock and actually play a key role in rhythm generation.
We also note that the proposed mechanism of signaling from membrane to gene transcription within a single cell via depolarized states does not necessarily require AP firing. In simulations of our extended model, ∼24-hour oscillations in cytosolic calcium and gene expression persist in the presence of TTX (Figure S5). This result emphasizes the role of a cell's membrane potential, more so than its firing rate, as a regulator of its intracellular molecular clock.
Previous studies have shown higher levels of Per1 expression in the SCN [36], [37] during the time of day when RMPs are more depolarized [38]. In this study we provide evidence that the depolarized RMPs reported in Belle et al. [13] correspond with high levels of steady-state intracellular calcium concentration that may be needed for activation of the Per1 gene through CREB [39], [40]. This is in line with Irwin and Allen [27], who show an increase in calcium levels in response to depolarization of SCN neurons, as well as Dolmetsch et al. [29], who show that high calcium levels are needed for CREB signaling. Further experimental work is needed to validate our predictions, including measurement of intracellular calcium levels in spontaneously depolarized SCN neurons and the specific levels of calcium needed to activate CREB signaling in these neurons.
While the basic mechanism of circadian timekeeping in mammalian cells is a transcriptional-translational negative feedback loop, electrical activity has sometimes been considered part of the core timekeeping mechanism [41]–[43]. Our work generates the hypothesis that depolarized states in clock neurons are part of the intracellular timekeeping mechanism. Accordingly then, the molecular clock controls the transcription of potassium channels, which, when expressed, could take the electrical state of the neuron into and out of depolarization block or DLAMOs. Depolarized states trigger high calcium, which in turn triggers the transcription of Per1 and Per2.
This hypothesis is consistent with and extends previously published data. In cerebellar granule cell cultures, Per1 expression was found to be dependent on the depolarization state of the neuron, and prevention of Ca2+ influx (by pharmacological blocking of voltage-gated calcium channels) reduced mPer1 induction [44]. Remarkably, cerebellar granule neurons also exhibit a spontaneous depolarized state at RMPs of −28 to 34 mV [14], [15]. In Drosophila, electrical silencing of pacemaker neurons stops the molecular clock [45], [46]. In SCN neurons, inhibiting Ca2+ influx [17] or CREB signaling [47] interferes with circadian rhythm generation. Here, we propose that in SCN neurons, calcium-activated potassium currents are diminished during the day, leading to higher calcium levels and more expression of PER1/PER2 at the expected phase. In addition, CREB is known to positively regulate large-conductance KCa (BK) channel expression in Drosophila [48], providing another potential mechanism for feedback.
Mathematical modeling is an established tool for understanding the complex interaction of neuronal ion channels and calcium dynamics [49], [50]. Several models of SCN neurons and circadian clocks contain calcium as a key component [51]–[55]. Our model is the first to estimate calcium levels in hyperexcited SCN neurons and compare such estimates with experimental data on cytosolic calcium levels, as well as the increases in intracellular calcium from action potentials. We relied on measurements of the overall cytosolic calcium level, and found a simple exponential clearance was sufficient to reproduce our data. When further experimental data on calcium regulation (e.g. release from intracellular stores) in SCN neurons becomes available, these details should be incorporated into our mathematical models. In our model, we have focused on the effect of cytosolic calcium on KCa and potassium leak currents, based on previous reports of rhythmic regulation of these channels [32]–[34]. However, there is evidence that the circadian clock may also regulate A-type potassium and L-type calcium channels [12], [56], and the effects of this additional control could be considered in future models.
Several experimental studies have used TTX to assess the role of sodium-dependent APs on circadian rhythmicity. Infusion of TTX into the SCN of freely moving rats disrupts behavioral rhythms but not internal circadian timekeeping [57]. In vitro application of TTX that abolishes AP firing in rat and hamster SCN slices does not affect the rhythm in SCN metabolism [58]. TTX also eliminates firing in SCN cultures [59], but does not block the cytosolic calcium rhythm [60]. Finally, single-cell bioluminescence measurements in SCN slice cultures indicate that TTX-sensitive APs are required for both robust clock gene expression rhythms in individual neurons and synchronization of these rhythms across cells [61]. Our simulation results (Figure S5) suggest that TTX dampens clock gene expression rhythms measured grossly across a population of SCN neurons primarily through accumulation of desynchronization rather than damping of individual oscillators. These simulations also predict that TTXLAMOs would only occur at certain times of day, which may help explain why these types of oscillations have not been widely reported in the literature outside of their initial discovery [12], [24].
Mizrak et al. [62] report that hyperexciting Drosophila clock neurons creates a morning-like expression profile for many circadian genes, while hyperpolarizing them creates an evening-like transcriptome. This underscores the different nature of neuronal signaling in depolarized versus conventional states. We predict that depolarized states in SCN neurons trigger transcription activation and can enhance rhythmicity, but future work is needed to understand how these depolarized states affect the dynamics of intracellular timekeeping in detailed mammalian clock models [63] and their other implications for neuronal information processing.
We extended the computational model of a SCN neuron from Belle et al. [13] to include L-type calcium current (ICaL), calcium-activated potassium current (IKCa), and intracellular calcium dynamics. The current balance equation for the revised model is:where C = 5.7 pF, Iapp = 0 pA, gNa = 229 nS, gK = 3 nS, gCaL = 6 nS, gCaNonL = 20 nS, gKCa = 100 nS, gK-leak = 0.0333 nS, gNa-leak = 0.0576 nS, ENa = 45 mV, EK = −97 mV, and ECa = 54 mV unless specified otherwise.
The dynamics of the gating variables q = m, h, n, rL, rNonL, fNonL, and s are:The L-type calcium current model is based on measurements of nimodipine-sensitive current in SCN neurons [24]. Since inactivation of L-type current is primarily calcium-dependent, we modeled it as:where K1 = 3.93E-5 and K2 = 6.55E-4 mM.
The IKCa kinetics follow the form of the voltage-independent calcium-dependent potassium current given in Yamada et al. [25], with parameter values chosen based on measurements of total KCa current in SCN neurons during action potential clamp experiments [24].
Intracellular calcium dynamics are extremely complex and involve many different mechanisms, such as buffering, uptake into and release from intracellular stores, and extrusion through membrane pumps. However, because in SCN neurons many of the details of these mechanisms have not been measured experimentally, we chose to use a very simple model of calcium dynamics that could be fitted directly to experimental data from these neurons. Our model represents all calcium handling mechanisms with a single term for the removal of free calcium ions from the cytosol, as in Booth et al. [64]. In our model, calcium enters the cytosol through voltage-gated calcium channels only; we do not explicitly consider release from intracellular stores. Thus, the concentration of free intracellular calcium ions is determined by the following equation:and is tracked separately in two compartments: one representing a thin spherical shell near the membrane surface where the binding of intracellular calcium ions to KCa channels occurs (Cas), and the other representing the entire cytosol (Cac). The parameter k converts Ca2+ current (pA) to Ca2+ concentration (mM), and τ is the Ca2+ clearance time constant. We set ks = 1.65e-4 mM/fC, corresponding to an SCN cell with a radius of about 5 µm [65] and a shell depth of 0.1 µm. This depth is a common choice for models where [Ca2+]i is relevant for KCa channel activation [25], [66]. Unless specified otherwise, we set kc = 8.59e-9 mM/fC, τc = 1.75e3 ms, and τs = 0.1 ms to match measurements of the total calcium current entering an SCN neuron during AP clamp from Jackson et al. [24], and our own measurements of [Ca2+]i in SCN neurons (see Figure 5). The constant term b sets the basal level of calcium in the absence of spiking, we chose bc = 3.1e-8 mM/ms (and bs = 5.425e-4 mM/ms) so that the steady-state values of Cac and Cas are approximately 54 nM in the absence of calcium entry (ICaL = ICaNonL = 0). See Figure S6 for a visualization of the evolution of Cac and Cas during AP firing.
Our basic model of gene regulation (Figure 6A) assumes linear degradation of mRNA (M), protein (P), and phosphorylated protein (P*) and has the form:The parameter a scales time, and was set to 5.6E-8 ms−1.
The extended gene regulation model (Figure 6C) incorporates membrane excitability by making the maximal conductances of calcium-activated potassium (gKCa) and potassium leak (gK-leak) dependent on the activity of the E-box:
In both the basic gene regulation model and the extended version, we assume that the CRE and the E-box interact multiplicatively [67]. In the basic model, the CRE activity is set to a constant value, CRE = 77.3, whereas in the extended model it tracks calcium concentration:
All differential equations are expressed in millisecond time units and all simulations were performed using the ode15s and ode23tb routines in Matlab® 2008 (Mathworks, Natick, MA). The variables M, P, and P* were assigned initial conditions of 0.1, and all other state variables were initialized to zero. Bifurcation diagrams were computed using XPPAUT [68].
We carried out targeted whole-cell electrophysiology in SCN neurons from fourteen male and female mice (∼2–3 months old) heterozygous for Per1::d2EGFP transgene (Per1::eGFP-expressing mice: a gift from D. McMahon, Vanderbilt University, TN, USA) bred and supplied by the Biological Services Facility of the University of Manchester. In these animals, a destabilized form of enhanced green fluorescent protein (eGFP) reports the activity of the mPer1 promoter [69]. Animal housing, mid-coronal SCN brain slice preparation, current-clamp recordings, and Per1::eGFP neuron visualization were performed as described in Belle et al. [13]. Drugs were bath applied in artificial cerebro-spinal fluid (aCSF) delivered to the slice by gravity feed. Stock solutions for nimodipine (Tocris, Bristol, UK) were prepared by dissolving in Dimethyl sulfoxide (DMSO): final working concentration of DMSO did not exceed 0.01%. Tetrodotoxin (TTX: Tocris) was dissolved in aCSF. All experimental procedures were carried out according to the provisions of the UK Animal (Scientific Procedures) Act 1986.
Simultaneous electrophysiological recordings and calcium imaging were performed using three male C57BL/6 mice (heterozygous for Per1::Venus expression, a gift from K. Obrietan, Ohio State University, OH, USA) that were housed for at least 1 week on a 12 : 12 h light : dark cycle. During the light phase, 7–12 week-old mice were anesthetized with isofluorane (Novaplus, UK), their brains removed and coronal hypothalamic slices (200–225 µm) containing the SCN were cut with a vibrating blade microtome (Leica-Microsystems VT1000S; Wetzlar, Germany). The tissue was surrounded by ice-cold artificial cerebrospinal fluid (ACSF) containing (in mM): NaCl, 126; KCl, 2.5; NaH2PO4, 1.2; MgCl2, 4; CaCl2, 0.5; glucose, 11; NaHCO3, 26; and saturated with 5% CO2 and 95% O2. The slices were maintained in a recording chamber (36°C) with a continuous laminar flow (1–2 mL/min) of an aCSF solution consisting of (in mM): NaCl, 132.5; KCl, 2.5; NaH2PO4, 1.2; MgCl2, 1.2; CaCl2, 2.4; glucose, 11; NaHCO3, 22; and bubbled with 5% CO2 and 95% O2. Whole-cell patch-clamp recordings of SCN neurons were performed during the night phase 1–8 h after slice preparation. Microelectrodes 5–10 MΩ were filled with an internal solution containing (in mM, Sigma-Aldrich, St. Louis, MO, USA): K-gluconate, 130; KCl, 5; HEPES, 10; MgATP 4; TrisGTP 0.4; ,di(tris) phosphocreatine 10; adjusted to pH 7.3 with KOH at (280–300 mOsm) to which was added bis-fura-2 hexapotassium salt (50 µm) (Molecular Probes-Life Technologies, Grand Island, NY, USA). After making a 1–5 GΩ membrane seal, the neuron was brought into whole-cell mode and rapidly filled with bis-fura-2. A HEKA EPC9 amplifier (HEKA, Lambrecht, Germany) and PatchMaster (HEKA) software were used to hold the neuron at a membrane potential of −60 mV and step (200 ms) the membrane potential to −20, 0, or 20 mV at 30–60 sec intervals to allow for recovery of the ensuing Ca2+ transient. While in voltage-clamp, 4 of the 10 neurons tested had TTX (0.5 µM) present but demonstrated no difference in the voltage step-induced change in [Ca2+]i (steps −20, 0 and 20 mV respectively, p = 0.50, p = 0.68, p = 0.56; unpaired two-tail t-test). Recordings were performed within 10 min of entering whole-cell mode. Calcium measurements were obtained and converted to estimated [Ca2+]i values as previously reported [27], [70]. The Institutional Animal Care and Use Committee of OHSU approved, in advance, all procedures involving animals.
|
10.1371/journal.pcbi.1005328 | Psychotic Experiences and Overhasty Inferences Are Related to Maladaptive Learning | Theoretical accounts suggest that an alteration in the brain’s learning mechanisms might lead to overhasty inferences, resulting in psychotic symptoms. Here, we sought to elucidate the suggested link between maladaptive learning and psychosis. Ninety-eight healthy individuals with varying degrees of delusional ideation and hallucinatory experiences performed a probabilistic reasoning task that allowed us to quantify overhasty inferences. Replicating previous results, we found a relationship between psychotic experiences and overhasty inferences during probabilistic reasoning. Computational modelling revealed that the behavioral data was best explained by a novel computational learning model that formalizes the adaptiveness of learning by a non-linear distortion of prediction error processing, where an increased non-linearity implies a growing resilience against learning from surprising and thus unreliable information (large prediction errors). Most importantly, a decreased adaptiveness of learning predicted delusional ideation and hallucinatory experiences. Our current findings provide a formal description of the computational mechanisms underlying overhasty inferences, thereby empirically substantiating theories that link psychosis to maladaptive learning.
| Predictive coding theories represent a unifying account of psychosis, stating that the central psychosis-related alteration affects the interplay between prior predictions and incoming information. Since every incoming information is imprecise and potentially allows for different interpretations, prior expectations achieve the enforcement of interpretations with a higher prior probability. Disturbances in this basic framework might let unlikely interpretations come into effect, resulting in proneness for delusions and hallucinations. Here, we contribute to these theories by devising a novel computational model for behavior in a reasoning task that quantifies the participants' readiness to draw inferences from very surprising information. We thereby demonstrate that precisely this increased learning from surprising and thus potentially spurious information, as opposed to non-specific alterations in the general learning speed, predispose healthy individuals for delusions and hallucinations. The present results hence speak for the hypothesis that hallucinations and delusions arise when noisy information is considered as precise and is thus not suppressed by opposing prior beliefs. In this sense, our findings also tie with recent neurophysiological models of psychosis that posit aberrations in modulatory neurotransmitters such as dopamine (or its interactions with GABAergic interneurons) as a correlate of perturbed computations of information precision in the cortex.
| Psychotic symptoms are a core symptom of devastating psychiatric disorders such as schizophrenia. They comprise many different kinds of experiences, among others beliefs that are unfounded in the external reality (delusions), and percepts in the absence of a causative stimulus (hallucinations). Accordingly, it poses a key challenge to theoretically and empirically establish models that can capture the multifariousness of psychotic experiences by a few (or even one) core alterations.
Influential theories [1–3] explain psychotic symptoms in the framework of predictive coding [4–6]. According to predictive coding, one central challenge for the brain is to draw inferences about the state of the external world from incoming information of relatively poor quality. It is stated that the brain deals with this challenge by recurring to predictive beliefs about the world. Such predictive beliefs are proposed to shape incoming information via top-down signals, thereby enabling stable and unitary inferences from imprecise and ambiguous information and constituting a protection against an over-interpretation of sporadically occurring irrelevant information. Importantly, predictive beliefs are assumed to be continuously updated by prediction errors. Such prediction errors are thought to drive learning via bottom-up signals, and to arise when predictive beliefs do not precisely match incoming information. Hence, ongoing learning in response to surprising information is thought to ensure the flexible adaptation of belief-dependent inferences.
Along these lines, psychotic symptoms can be framed as maladaptive learning that occurs if irrelevant information is considered as surprising and relevant due to altered prediction error signaling [1,7,8]. As a result, no stable and valid predictive beliefs would be built up and the brain would become susceptible to overhasty and erroneous inferences yielding delusions and hallucinations. In line with the idea that overhasty and erroneous perceptual inferences from irrelevant noise information are implicated in hallucinations and hallucination-proneness, hallucinatory experiences have been repeatedly associated with a greater tendency to perceive illusory contents in auditory noise [9,10]. Moreover, delusional ideation has been consistently linked to “jumping to conclusions” (JTC, see [11–13] for detailed meta-analyses), a cognitive reasoning bias that leads to a rash acceptance of hypotheses based on little evidence. However, it is a matter of ongoing debate, which particular cognitive alteration predisposes delusional and delusion-prone individuals for an overhasty acceptance of possible hypotheses [14–17]. With regard to the predictive coding account of psychosis outlined above, we suggest that JTC might reflect a pivotal alteration underlying psychotic symptoms, namely maladaptive learning from irrelevant information, leading to overhasty inferences.
To empirically test the claim that maladaptive learning contributes to psychotic symptoms, one will necessarily have to tackle the question of what constitutes adaptive learning, or, in other words, how non-psychotic individuals can generate and adapt beliefs sufficiently quickly in response to relevant information, and, nevertheless, resist inadequate belief revision due to irrelevant noise. Common computational learning models (e.g., [18]) formalize learning in terms of prediction errors and learning rates. Here, the current belief is obtained as a function of the prediction error that denotes the difference between the expectation (i.e., the belief before the actual observation) and the actual observation. The magnitude of this prediction error multiplied with a subject-specific learning rate determine the degree to which the belief is updated (i.e., the learning). An alternative formulation of evidence accumulation (and state estimation) calls on Bayesian filtering schemes as metaphors for neuronal computations. These schemes accumulate evidence for hidden states of the world in proportion to their estimated precision or reliability. The most celebrated Bayesian filter is called the Kalman filter, where the Kalman gain corresponds to the relative precision (inverse variance) of sensory evidence in relation to prior beliefs. Biologically plausible implementations of Kalman filtering include predictive coding, where Bayesian belief updating (i.e., evidence accumulation) is mediated by precision weighted prediction errors. In short, Rescorla Wagner models, Bayesian filtering and predictive coding are all equivalent formulations of evidence accumulation (see [19]). They all speak to the importance of precision as learning rates in modulating the impact of prediction errors on belief updating, which we will refer to as adaptive learning.
Thus, these common computational learning models capture adaptive learning, as opposed to maladaptive learning from irrelevant information that lead to overhasty and erroneous inferences, by small learning rates. Hence, resilience against irrelevant information would be formalized by smaller learning rates and thus comes at the expense of a generally decreased speed of learning (see Fig 1). Here, we propose a novel computational learning model that is able to capture the resilience against irrelevant information without substantially impairing the general speed of learning. The central and very simple idea of our model is that prediction errors are processed in a non-linear fashion. Concretely, we introduce a saturating non-linear function of prediction error that attenuates the effect of very large prediction errors on belief updating, relative to smaller prediction errors. Effectively, this means that very surprising or large prediction errors are treated as imprecise information; very much in the same way that we discard outliers in statistical analyses of data. In the technical literature this is known as Winsorizing and represents one of the simplest and most fundamental modifications of linear predictive coding. Formally, this compressive non-linearity can be considered a hyperprior that certain prediction errors are generated by a class of outliers that can be construed as "irrelevant". In other words, the non-linearity enables the accumulation of evidence in a way that is resistant to the effect of spurious (i.e., very surprising) events. Importantly, learning from small prediction errors is preserved, leading to adaptive inferences in response to moderately surprising and hence relevant information. Thus, our model captures the resilience against irrelevant information, and hence overhasty and erroneous inferences, by the non-linearity of prediction error processing (see Fig 1). Conversely, we would predict that a weaker resilience against irrelevant information that leads to overhasty and erroneous inferences in psychotic and psychosis-prone individuals is paralleled by a more linear processing of prediction errors.
In this work, we sought to devise a formal approach to assess and quantify the maladaptive learning mechanisms underlying overhasty and erroneous inferences related to psychotic symptoms. To this end, we devised an adapted probabilistic reasoning task that allowed us to continuously track participants’ belief trajectories. We then used this task to quantify overhasty inferences in a sample of healthy individuals with varying degrees of delusional ideation and hallucinatory experiences, based on the view that clinically relevant psychotic symptoms represent an extreme of a trait continuously distributed in the general population [20,21]. In order to investigate the computational mechanisms underlying psychosis-related biases in learning and inference, we fitted the behavioral data with our novel learning model that quantifies the adaptiveness of learning by a non-linear prediction error processing. We hypothesized that psychosis-related experiences would inversely relate to the resilience against irrelevant information quantified by the non-linearity of prediction error processing.
Ninety-eight healthy individuals from the general population were recruited for study participation through advertising. The study was approved by the Ethical Committee of the Charité, Universitätsmedizin Berlin. Participants who received treatment due to psychiatric diseases were excluded. After complete description of the study to the participants, written informed consent was obtained in accordance with the Declaration of Helsinki of 1975 before participation.
The participants' tendency towards delusional ideation was quantified using the Peters Delusion Inventory (PDI, [22]). The 40 items of this self-rating questionnaire cover a wide range of delusional convictions, including beliefs in the paranormal, grandiosity ideas or suspicious thoughts. For every endorsed belief, the questionnaire asks for dimensional ratings on the degree of belief-related distress, preoccupation and conviction. The total score obtained by adding up these three dimensional ratings was used for analyses.
Additionally, proneness to hallucinatory experiences was assessed with the Cardiff anomalous perception scale (CAPS, [23]). This 32-item self-rating scale assesses anomalous perceptual experiences in different sensory domains like proprioception, time perception, somatosensation and visual and auditory perception. The intensity of every anomalous perception is quantified from one to five on subscales for intrusiveness, frequency and distress. Again, the total score was calculated by adding up all subscore ratings and used for analyses.
An adapted version of the “beads task” [24] was used to assess psychosis-related alterations in probabilistic reasoning, especially overhasty inferences such as the JTC bias. In the beads task, beads are continuously drawn from one of two different urns that contain different numerical proportions of different kinds of beads. The participants have to infer from which urn beads are currently being drawn based on their knowledge about the numerical proportions of different kinds of beads in the two urns and the number of already drawn beads of each kind. The task thus implies a continuous update of the belief about the correct urn with every new draw, which can be either consistent with the current belief about the correct urn (relevant information) or inconsistent with it (irrelevant information).
In our version of this task, the participants were shown pictures of two different lakes (a “mountain lake” and a “flatland lake”) and told that these lakes are home to a different proportion of carps and trouts with the mountain lake containing 70% carps and 30% trouts and the flatland lake 30% carps and 70% trouts. For reasons of simplicity, we will refer to the mountain lake as the "carp lake" and to the flatland lake as the "trout lake" in the following.
The task was structured in 30 rounds with a varying number of draws. On each round, fishes were sequentially angled from one of the two lakes and the participants were instructed to evaluate from which of the lakes the fishes were more likely angled in this round using the number of so far angled carps and trouts and their knowledge about the numerical proportion of fishes in the two lakes (thus with every angled carp making the carp lake more probable and every angled trout making the trout lake more probable). Moreover, participants were told that both lakes contained so many carps and trouts that the numerical proportions did not change due to the fishing.
Each round started with only one angled fish and, accordingly, with a rather imprecise information about which of the two lakes being correct in this round. To gain further information, participants were allowed to make new draws until they felt confident enough to make a final decision about the correct lake in this round (Fig 2). With every new draw, one new fish was angled and the number of so far angled trouts and carps was updated and presented. After each draw, the participants indicated their new belief about from which lake the fishes were probably angled in this round. For this purpose, they entered their guess and its certainty using the mouse on a visual scale (ranging from absolute certainty of the carp lake being correct at the very left to absolute certainty of the trout lake being correct at the very right with positions close to the center indicating uncertainty). In this way, we obtained a continuous assessment on the participants' current belief for each draw. After having placed their guess, the participants were asked if they wanted to commit themselves to the given response on the correct lake (by pressing either the up or the down arrow key). If they did not commit to their response, a new fish was angled (new draw). If they committed to their response, a final decision on the lake was made and a new round started with once again only one angled fish and accumulating evidence with every further draw.
To induce prediction errors even in rounds with few draws, we added a prior information about the lakes' probabilities in the form of a high- or low-pitched tone that was played shortly before every newly angled fish. In one round, always the same tone pitch was played. If the fishes were angled from the carp lake in the round, the high-pitched tone was played more frequently (80%) and if the fishes were angled from the trout lake, the low-pitched tone was played more frequently (80%). Thus, the tone pitch constituted a probabilistic initial information about the lake probabilities for each round. The associations between tone pitch and lake probabilities were learned in a preceding learning run of 15 rounds and did not change throughout the experiment. By these means, we could assess prediction errors already in the first draw (one angled fish) and increase the variance of prediction error values occurring throughout the course of the experiment.
To quantify the tendency towards overhasty inferences in each participant, we calculated the mean number of draws a participant needed on each round before committing to a final decision. This measure ("draws to decision") is an accepted measure for the JTC bias found to be associated with psychotic symptoms [14].
To replicate prior findings that participants with growing psychosis proneness tend to exert jumping-to-conclusions (see introduction), we tested associations between the participants’ draws to decision and the tendency towards delusional convictions (PDI scores) as well as the proneness to hallucinatory experiences (CAPS scores) in two different ways. Firstly, as suggested by [25], we performed a binary analysis with our sample separated into two groups (with and without JTC). There were only six of 94 participants showing JTC according to the commonly applied threshold of two draws to decision, probably due to the differing set-up of our adapted version of the lake task (introduction of prior knowledge associated with the tone, usage of continuous response bar). Thus, we used a slightly higher threshold and considered participants in the lowest quartile of draws to decision (i.e., with an average of 3.2 or less draws to decision) as exhibiting JTC and compared their PDI and CAPS scores with the remaining (non-JTC) sample. Secondly, we investigated continuous relationships by correlating PDI and CAPS scores with the mean number of draws to decision.
Since the distribution of PDI and CAPS scores differed significantly from a normal distribution (Z = 1.374, p = 0.046 for PDI scores and Z = 1.941, p = 0.001 for CAPS scores, one-sample Kolmogorov-Smirnov tests), we used non-parametric Mann-Whitney tests for the first (categorical) and Spearman rank correlations for the second (correlational) analysis.
To our knowledge, there are no previous studies reporting associations between hallucinations and JTC, giving our analysis on relationships between CAPS scores and JTC a rather exploratory character. Nevertheless, because we tested associations between JTC and both PDI and CAPS scores, we report among uncorrected p values also p values with adjustment for multiple testing (tests for PDI and CAPS scores, e.g., two tests with correlated outcomes). To this end, p values were adjusted according to the approach proposed by [26] and outlined in [27] for multiple comparisons with correlated outcomes.
By fitting the behavioral data with computational learning models, we aimed at quantifying the resilience against irrelevant information and thereby assessing the adaptiveness of learning, which we expected to be inversely related to psychotic symptoms.
Two computational models were designed to track the participants' trajectories of belief in the probabilistic reasoning task. Firstly, we applied a conventional linear prediction-error-based learning model (e.g., [28]). Secondly, we developed a novel model that enabled the quantification of the participants’ resilience against irrelevant information through a non-linear relationship between prediction error and learning, which we expected to provide a more precise description of adaptive learning in probabilistic reasoning.
In both models, the participants’ beliefs about the correct lake were captured on a trial-by-trial basis as a continuous value between 0 (certainty that the “carp lake” is correct) and 1 (certainty that the “trout lake” is correct). Thus, the high-pitched tone as well as newly angled carps brought the belief nearer to the 0 and the low-pitched tone as well as newly angled trouts nearer to the 1. Eq 1 shows accordingly, that the neutral belief of 0.5 was initially shifted towards 1 in case of the "trout-lake"-associated low-pitched tone and towards 0 in case of the "carp-lake"-associated high-pitched tone and that the magnitude of the tone-dependent belief shift depended on the subject-specific parameter θ. Since the neutral belief of 0.5 could be shifted by maximally 0.5 by the tone, we used a uniform distribution between 0 and 0.5 as a prior distribution for the estimation of θ values based upon choice behavior.
Initialtone-dependentbelief(±:+iftroutisangled,-ifcarpisangled).b1=0.5+θ
Eq1
Whereas this initial tone-dependent belief was calculated in the same way in both models, the effect of newly angled fishes differed between the conventional linear and our novel non-linear model. In the linear model, the prediction error determined the learning linearly. Eq 2 shows that the belief update here depended on the non-modified prediction error bi-1 –oi (difference between the former belief bi-1 and the current observation oi) that was multiplied with a subject-specific constant learning rate α that captures the general rapidity of belief generation regardless of the typicality of the new information. Since the learning rate is naturally bounded between 0 and 1, we used a uniform distribution between 0 and 1 as a prior distribution for estimation of α values based upon choice behavior.
In the non-linear model on the other hand, the learning depended on the prediction error with a varying degree of non-linearity expressed by the non-linearity parameter ζ. Please note that high values of ζ imply a marked non-linearity / flattening of the relationship between prediction error and learning, whereas this relationship is linear for ζ = 0. Thus, high values of ζ imply a strong resilience against irrelevant information, since high prediction errors have a reduced impact on learning in this case: Hence, this modulation can be thought of as a dynamic learning rate that adaptively decreases if information is unreliable and potentially irrelevant. As in common behavioral learning models, the resulting non-linear learning term was multiplied with a subject-specific constant learning rate α that captures the general rapidity of belief generation regardless of the typicality of the new information. Eq 3 shows how the current belief bi is updated depending on the learning rate α and the prediction error bi-1 –oi, whose impact on the learning decreases with increasing values of ζ. Compared to other possible implementations of a non-linear prediction error, the definition outlined above has the advantage of yielding one simple parameter that determines the degree of non-linearity and is zero for an entirely linear relationship between prediction error and learning. Furthermore and importantly, it cannot generate overshooting beliefs below zero or above one without having to assume an additional softmax transformation (see proof in the Supplementary Material). Fig 1 shows exemplary relationships between prediction error and learning, with linear relationships (ζ = 0) in blue (α = 0.6) and yellow (α = 0.2) and a non-linear relationship in red with ζ = 4 and α = 1. Since such a non-linear prediction error processing has to our knowledge not been implemented so far, we used a uniform distribution between 0 and 5 (thus allowing for a wide range of non-linearity) as a prior distribution for estimation of ζ values based upon choice behavior.
Both models were applied to explain the trajectory of the participants' beliefs about the correct lake throughout the course of the experiment. For this purpose, the trial-by-trial belief indicated on the continuous response bar was scaled between 0 and 1, yielding the trajectory of belief vector g. Subsequently, each participants' trajectory of belief g was fitted with both models using the VBA Toolbox for Matlab [29]. This approach uses Variational Bayesian methods to estimate the parameter values of our two models for which the trajectory of the belief b predicted by the model optimally traces the real belief g indicated by the participants (Fig 3). Furthermore, the (lower bound on the) model's evidence (marginal likelihood), i.e., the likelihood that the real trajectory of belief g could have been generated by the respective model, was computed and used for model comparison (see below).
Summing up, the following parameters were estimated to optimally model the participants' behavior:
θ: Tone-dependent initial (i.e., prior) belief
α: General learning rate
ζ: Non-linear prediction error processing (resilience against irrelevant information, ζ = 0 in case of the linear model)
To test if the non-linear model that allowed for a non-linear relationship between prediction error and learning explained the participants’ behavior better than the conventional linear model, we performed a formal Bayesian model comparison between the two models. Therefore, both models were used to fit the participants' continuous belief trajectories and the resulting model evidences were compared using the approach outlined in [30] and implemented in Statistical Parametric Mapping 12 (SPM 12). In this approach, the ability of a model to accurately predict the participants behavior is balanced against its complexity, where growing model complexity is punished. In our case, this means that if the non-linear model proves to be superior in the model comparison, the growing complexity which results from the inclusion of the additional non-linearity parameter is overcompensated by the gain in accuracy afforded by it. In addition to formal model comparison, we calculated the explained variance R2 of each participants’ belief trajectory by the model in order to obtain a clear measure of how well the models were able to capture the participants’ behavior. Please note that in contrast to Bayesian model comparison as described above, this assessment of model fit does not take into account the model complexity and is therefore not an appropriate measure for formal model comparison.
The introduction of the tone as prior information about the lakes' probabilities allowed us to assess prediction errors even in participants with few draws to decision (see above). However, it has to be ensured, that the tone meanings has been learned by the participants during the learning run. Moreover, it has to be ensured, that differences in learning the tone meaning associated with varying psychosis-proneness constitute no alternative explanation for the relationships between psychosis-proneness and altered information processing assessed in this study.
For the former purpose, we performed a “proof of concept” model comparison between the full model with the tone parameter estimated as a free parameter and a model in which the tone parameter was fixed to 0 (i.e., that assumes that the tone has not been used by the participants). A superiority of the full model over the model with fixed tone parameter would prove that the tone was indeed used by the participants. Formal model comparison was carried out as described in the “model comparison” paragraph above.
For the latter purpose (excluding psychosis-related differences in learning the tone), we correlated PDI and CAPS scores with the value of the tone parameter θ. A lack of such a relationship would demonstrate that there is no evidence that learning and usage of the tone depended on the participants' psychosis proneness.
Contrary to most computational learning models, our model included a non-linear relationship between prediction error and learning that captures a reduced impact of high prediction errors resulting in adaptive reduced learning from irrelevant information, and, thus, in a resilience against overhasty and erroneous inferences. The degree of the adaptiveness of learning is quantified by the non-linearity parameter ζ, with higher values of ζ indicating a stronger adaptiveness of learning. Because there are to our knowledge no prior studies that implemented this kind of non-linear prediction error processing, we computed the subject-specific values of ζ without prior assumptions on its distribution, i.e., with a uniform prior distribution that made every value between 0 and 5 equally likely. To generally assess the form in which the degree of resilience against irrelevant information was distributed in our sample of participants, we tested the hypotheses that the estimated values of ζ were uniformly distributed (like the naive prior distribution) or normally distributed (like many psychological and biological variables). To this end, one-sample Kolmogorov-Smirnov tests were applied.
To test if a low resilience against irrelevant information was related to overhasty inferences (i.e., to jumping-to-conclusions), we correlated the values of ζ with the participants' mean number of draws to decision in the probabilistic reasoning task. This analysis primarily served as a proof-of-concept since the parameter values of an appropriate behavioral model for the probabilistic reasoning task should be able to explain a large part of the interindividual variance in jumping-to-conclusions.
To additionally demonstrate that JTC can be more accurately explained by a reduced resilience against irrelevant information compared to a generally increased learning speed, we subsequently calculated correlations between the number of draws to decision and the participants’ learning rates (values of α in the linear model).
To test our hypothesis that psychosis-related experiences would inversely relate to the adaptiveness of learning, we calculated correlations between the participants’ values of ζ and the proneness for delusional convictions (PDI scores) as well as hallucinatory experiences (CAPS scores). Since the distribution of PDI and CAPS scores differed significantly from a normal distribution (as reported above), we again used non-parametric Spearman rank correlations for this purpose.
To additionally demonstrate that this relationship is specific for a reduced resilience against irrelevant information and does not only reflect a generally increased learning speed, we repeated the analysis including the participants’ learning rates (values of α in the linear model) as a covariate in a Spearman partial correlation between PDI and CAPS scores and ζ values.
Finally, we repeated these analyses correcting for multiple comparisons with correlated outcomes (according to [26], see above).
Four participants were excluded from analyses because they showed excessively high error rates at the final decision of the probabilistic reasoning task (more than two standard deviations above sample mean, corresponding to more than 26.2% wrong decisions), suggesting that they did not perform properly in the task. The characteristics of the remaining sample of 94 participants are summarized in Table 1.
A large body of evidence has linked psychosis and psychosis-proneness to a reduced number of draws to decisions in the beads task (JTC, [11,31]). To test whether we could replicate this relationship in our current sample, we related the mean number of draws to decision to PDI scores (proneness for delusional convictions) and CAPS scores (proneness for anomalous sensory experiences). Indeed, our categorical assessment revealed a significantly increased PDI in the JTC group (n = 23, median of PDI scores = 79) as compared with the no-JTC group (n = 71, median of PDI scores = 45), Mann-Whitney test with U = 585, p = 0.042, two-sided. No significant difference was found for CAPS scores (median CAPS score in JTC group = 30, median score in no-JTC group = 15, Mann-Whitney test with U = 681,5, p = 0.233).
Correlational analyses reproduced the effect for PDI scores on a trend level (fewer draws to decision with rising PDI scores, rho = -0.177, p = 0.089, two-sided Spearman rank correlation), whereas CAPS scores again showed no significant relationship (rho = -0.154, p = 0.139, two-sided Spearman rank correlation).
When adjusted for multiple comparisons with correlated outcomes, the relationship between JTC and PDI scores was still present, but failed to reach statistical significance (p adjusted = 0.051 in the categorical analysis and p adjusted = 0.108 for the correlational analysis).
To test if our model that allowed for a non-linear relationship between prediction error and learning explained the participants’ behavior better than a standard linear model, we performed a Bayesian Model Comparison between the two models. Here, the non-linear model proved to be superior to the standard linear model in explaining participants’ behavior with a protected exceedance probability of 100%. The mean value of explained variance (R2) of the participants trajectory of belief was 0.696 for the non-linear and 0.661 for the linear model. Since this measure of accuracy does not take into account the differing model complexity, it is not suitable for directly comparing the quality of the models. It however shows that especially the non-linear model appropriately tracked the course of the participants’ belief.
With two proof-of-concept analyses, we aimed at ensuring, that the tone meaning was learned by the participants during the learning run, but that psychosis-proneness was not associated with differences in learning the tone meaning.
To ensure the significance of the tone for the participants' belief updating, we performed a formal model comparison between the full model, in which the tone parameter was estimated as an individual free parameter and a model without tone parameter (i.e., with θ fixed to 0, assuming that the tone was not used by the participants). Here, the model, in which θ was freely estimated, proved to be superior with a protected exceedance probability of 99.5%. Thus, taking into account the usage of the tone significantly improved the tracking of the participants' belief trajectory.
No significant or trend-wise relationships were found between the value of the tone parameter and delusion-proneness (PDI scores, rho = 0.030, p = 0.771, two-sided Spearman correlation) or hallucination-proneness (CAPS scores, rho = 0.038, p = 0.714), not providing any evidence that individual differences in the learning of the tone might provide an alternative explanation for the found relationships between psychosis proneness and the lowered resilience against irrelevant information.
The non-linearity parameter ζ quantifies the degree to which the impact of the prediction error on learning is attenuated if new information is very surprising. Accordingly, low values indicate maladaptive learning with a low resilience against irrelevant information. In common learning models, a linear relationship, i.e., a parameter value of ζ = 0, is assumed. We estimated the values of ζ that optimally explained our participants' behavior with an uninformative prior distribution uniformly distributed between 0 (linear relationship) and 5 (strongly non-linear relationship). It turned out that none of our participants showed a linear processing of the prediction error. Instead, more than 95% of the participants showed values of ζ between 2.5 and 4 with a marked peak around 3.5 (Fig 4). Interestingly, the distribution of the estimated values of ζ differed significantly from the prior uniform distribution (Z = 4.663, p < 0.001, Kolmogorov-Smirnov test), but not from a normal distribution (Z = 1.051, p = 0.219, Kolmogorov-Smirnov test). These results suggest that in our probabilistic reasoning task, learning depended on the prediction error in a non-linear fashion and all participants showed resilience against irrelevant information, although the degree of this resilience varied across participants.
To test if the participants’ resilience against atypical information was associated with jumping-to-conclusions behavior, we correlated parameter values of ζ with the participants mean number of draws to decision in the probabilistic reasoning task. This analysis yielded a strong positive correlation (r = 0.705, p < 0.001, Pearson Correlation), indicating that participants with a lower resilience against irrelevant information took decisions based on less evidence (JTC).
As predicted, the learning rate α of the linear model also showed a (negative) correlation with draws to decision, albeit weaker than ζ in the non-linear model (r = -0.460, p < 0.001, Pearson Correlation).
According to predictive coding models of psychosis, maladaptive learning with a reduced resilience against irrelevant information would result in overhasty and erroneous inferences, and should therefore be related to an increased proneness towards delusional ideation and hallucinatory experiences.
Notably and in line with this hypothesis, estimated parameter values of ζ were negatively correlated with PDI scores (rho = -0.235, p = 0.022, two-sided Spearman rank correlation, Fig 5A) and trend-wise with CAPS scores (rho = -0.198, p = 0.056, two-sided Spearman rank correlation, Fig 5B), indicating that individuals with a low resilience against irrelevant information showed an increased proneness for delusional ideation and (as a tendency) hallucinatory experiences. To exclude significant correlations due to four outliers with ζ values two standard deviations below mean (i.e., below 2.794, outliers are marked with white squares in Fig 5), we repeated these analyses excluding the outliers (thus with n = 90). This resulted in rather stronger effects (rho = -0.261, p = 0.014 for the correlation between PDI scores and ζ, rho = -0.212, p = 0.044 for the correlation between CAPS scores and ζ).
These correlations remained unchanged when controlling for the unspecific learning rate obtained from the linear model: A Spearman partial correlation including the linear learning rate as a covariate revealed similar results to those reported above (r = -0.234, p = 0.024 for correlation between PDI scores and ζ, r = -0.196, p = 0.062 for correlation between CAPS scores and ζ). This shows that effects between psychosis proneness and belief updating specifically affect the treatment of irrelevant information and cannot be explained by the nonspecific linear learning rate.
When p-values were adjusted for multiple comparisons (two tests with correlated outcomes for CAPS and PDI scores), the relationship between PDI scores and ζ values remained significant (adjusted p of two-sided Spearman rank correlation = 0.027) and the relationship between CAPS scores and ζ values a statistical trend (adjusted p of two-sided Spearman rank correlation = 0.068).
In the present study, we tested the claim put forward by predictive coding models [1–3] that psychotic experiences may be linked to maladaptive learning, i.e., an aberrant encoding of precision, that results from a reduced resilience against irrelevant information and leads to overhasty and erroneous inferences. In line with our hypothesis, we found that delusional ideation and hallucinatory experiences of healthy individuals were predicted by a low resilience against irrelevant information in a probabilistic reasoning task.
In order to quantify the resilience against irrelevant information, we applied a novel computational learning model that allowed for a non-linear relationship between prediction error and learning. Compared to a linear relationship, individuals with a non-linear processing of the prediction error are relatively resilient against an overestimation of excessively surprising information, since the relationship between prediction error and learning is flattened for high prediction errors. On the other hand, they are still capable of rapidly building predictive beliefs about the world, since learning of moderately surprising information (low prediction errors) is not relevantly impaired. By this approach, we were thus able to disentangle inter-individual differences in the general speed of learning (that is captured in the learning rate) from specific differences in the impact of former beliefs on learning (that is captured in the resilience against irrelevant information). Our results suggest that specifically the latter is related to an increased proneness for delusional and hallucinatory experiences. Considering that every incoming signal is noisy and naturally contains both relevant and irrelevant information, it seems plausible that an attenuation of specifically the excessively surprising and hence irrelevant information constitutes an effective protection against overhasty and erroneous inferences, while a weakening of this attenuation in turn predisposes for delusional beliefs and hallucinatory percepts. This understanding is additionally supported by our finding that the resilience against irrelevant information indeed was the parameter with the strongest association with hasty inferences (jumping-to-conclusions) and that jumping-to-conclusions was, consistent with prior studies [11,31], in turn related to the proneness for delusional convictions. Obviously, the adaptiveness of a non-linear prediction error processing depends on the particular task: In volatile tasks with frequent changes of the underlying probabilities, large prediction errors might in fact provide vital information, namely that the context of learning has changed. Our task however included no volatility in that sense, because in one round, there were no changes in the task probabilities (the lake, from which fishes were being angled remained the same). Thus, the degree of non-linearity indeed provides a measure for adaptive learning in the adapted beads task.
The idea that a core alteration in psychosis lies in the weighting of new information with regard to prior beliefs has a longstanding history in cognitive schizophrenia research evolving from the suggestion that “the basic disturbance in schizophrenia is 'a weakening of the influences of stored memories of regularities of previous input on current perception‴ [32] via hypotheses of an aberrant attribution of salience to stimuli [8,33] to predictive coding frameworks that embed these hypotheses into a broader framework of Bayesian information processing in the brain [2,3]. In line with this, schizophrenia and/or psychotic symptoms have been consistently linked to the aberrant attribution of salience to stimuli ([34,35] for reviews). Similarly, schizophrenia and psychotic symptoms have been associated with a decreased influence of prior beliefs in perceptual inference ([36], see [37] for a review on visual illusions), although recent work suggests a complex interplay between prior beliefs and perceptual inference in psychosis-related conditions [38,39]. By the use of a feasible and interpretable model, our current study provides a formal description for the interaction between prior beliefs and new information, thereby elucidating the computational mechanisms underlying maladaptive learning and inference in psychosis.
Intriguingly, we found that our participants showed without exception a resilience against irrelevant information that cannot be captured in models that assume a linearly processed prediction error. Especially considering the significant clustering of parameter values in a region with a marked non-linearity in our experiment, this finding suggests that learning in some tasks might not be driven linearly by prediction errors. From a more general perspective, a growing non-linearity between prediction error and learning implies that the impact that a certain new information has on the belief (i.e., the learning) becomes increasingly independent from the current belief itself: Whilst under the assumption of a linearly processed prediction error, one and the same information (e.g., a certain fish in our task) has a massively different impact on the learning depending on whether it is surprising or not, a strong non-linearity implies that every fish is treated more or less equally, regardless of the current belief. Similar concepts have previously been proposed in terms of precision-weighted prediction errors, where the learning of strongly surprising information is attenuated if a marked and precise opposing belief has already been built, e.g., if the precision of the belief is high and the precision of the new information low [19]. Compared to these frameworks, our approach has the advantage of simplicity and that the degree of resilience against irrelevant information is captured in one single and easily interpretable parameter: It is noteworthy, that a reduced non-linearity of the relationship between prediction error and learning that could be proven to be associated with psychosis-proneness in this study effectively and straightforwardly models what has been theoretically proposed as a core alteration behind psychotic experiences, namely “a reduction in the precision of prior beliefs, relative to sensory evidence” [1]. Nevertheless, whilst providing a substantial model fit in rather simple tasks like the one carried out in this study, it is questionable if our non-hierarchical model can sufficiently account for more complex environments (e.g., environments with changing volatility). Based on the continuity view of psychosis, we studied psychotic experiences in a sample of non-clinical participants. Mounting evidence suggests that clinical and non-clinical psychotic experiences reflect different expressions of a continuously distributed trait, as they share a common factor structure [40], similar risk factors and demographics [21] as well as a co-clustering in relatives [41,42]. It could moreover be prospectively demonstrated that an increased, but non-clinical proneness for psychotic experiences massively increases the risk of developing a "full" clinical psychosis in the future [43–45], further indicating that non-clinical and clinical psychotic experiences can be explained in terms of similar underlying mechanisms. Importantly, studying psychotic experiences in non-clinical participants does preclude potential confounds associated with clinical diseases and their pharmacological treatment. Hence, although future work is needed to confirm whether our current findings generalize to patients suffering from psychotic disease, the link between maladaptive learning and psychotic experiences established here might generally shed light on the computational mechanisms underlying both non-clinical psychotic experiences and psychosis.
One of the studies limitations is that we only yielded a modest and non-significant association between conventional JTC measures (draws to decision) and psychosis proneness. This is however consistent with previous reports on the relationship between JTC and psychosis proneness in healthy individuals that yielded small effects and mixed results [46–49] and suggest that conventional JTC measures such as draws to decision might not provide a sufficiently fine-grained measure for individual psychosis-related differences in learning and reasoning in healthy individuals.
Taken together, our current findings suggest that a less non-linear processing of prediction error gives rise to overhasty and erroneous inferences, thereby leading to delusional ideas and hallucinatory experiences. Our current work thus empirically substantiates theories that link maladaptive learning to psychotic experiences both in health and disease.
|
10.1371/journal.ppat.1003120 | Discovery of a Siderophore Export System Essential for Virulence of Mycobacterium tuberculosis | Iron is an essential nutrient for most bacterial pathogens, but is restricted by the host immune system. Mycobacterium tuberculosis (Mtb) utilizes two classes of small molecules, mycobactins and carboxymycobactins, to capture iron from the human host. Here, we show that an Mtb mutant lacking the mmpS4 and mmpS5 genes did not grow under low iron conditions. A cytoplasmic iron reporter indicated that the double mutant experienced iron starvation even under high-iron conditions. Loss of mmpS4 and mmpS5 did not change uptake of carboxymycobactin by Mtb. Thin layer chromatography showed that the ΔmmpS4/S5 mutant was strongly impaired in biosynthesis and secretion of siderophores. Pull-down experiments with purified proteins demonstrated that MmpS4 binds to a periplasmic loop of the associated transporter protein MmpL4. This interaction was corroborated by genetic experiments. While MmpS5 interacted only with MmpL5, MmpS4 interacted with both MmpL4 and MmpL5. These results identified MmpS4/MmpL4 and MmpS5/MmpL5 as siderophore export systems in Mtb and revealed that the MmpL proteins transport small molecules other than lipids. MmpS4 and MmpS5 resemble periplasmic adapter proteins of tripartite efflux pumps of Gram-negative bacteria, however, they are not only required for export but also for efficient siderophore synthesis. Membrane association of MbtG suggests a link between siderophore synthesis and transport. The structure of the soluble domain of MmpS4 (residues 52–140) was solved by NMR and indicates that mycobacterial MmpS proteins constitute a novel class of transport accessory proteins. The bacterial burden of the mmpS4/S5 deletion mutant in mouse lungs was lower by 10,000-fold and none of the infected mice died within 180 days compared to wild-type Mtb. This is the strongest attenuation observed so far for Mtb mutants lacking genes involved in iron utilization. In conclusion, this study identified the first components of novel siderophore export systems which are essential for virulence of Mtb.
| In the late 19th century the French physician Armand Trousseau recognized that treating anemic tuberculosis patients with iron salts exacerbated the disease. In 1911 Twort postulated that mycobacteria produce an essential growth factor which was identified in 1953 as mycobactin. The hydrophobic mycobactin and its more water-soluble cousin carboxymycobactin are small molecules made by Mycobacterium tuberculosis to scavenge iron from its human host. While the biosynthesis of these siderophores has been decoded, it was unknown how M. tuberculosis secretes these molecules. In this study, we identified two similar transport systems, MmpS4/MmpL4 and MmpS5/MmpL5, which are required for biosynthesis and export of siderophores by M. tuberculosis. The lack of these transport systems drastically decreased the number of M. tuberculosis cells in the lungs and spleens of infected mice. Lung examination and histological assessment in mice infected with the mmpS4/S5 deletion strain showed almost no signs of infection. Further, none of the mice infected with this strain died within 180 days in contrast to wild-type M. tuberculosis. In this study, we identified the first components of a novel siderophore export system in M. tuberculosis and showed the importance of siderophore export for virulence of M. tuberculosis.
| Iron is an essential micronutrient for most forms of life on earth because of its vital role as a redox cofactor of proteins required for critical cellular processes. Pathogenic bacteria have evolved an array of intricate mechanisms to scavenge limited iron from the host [1]. Mycobacterium tuberculosis (Mtb), one of the most successful human bacterial pathogens, is no exception. Mtb meets its iron demands by stripping host iron stores employing two hydroxyphenyloxazoline siderophores, mycobactin (MBT) and carboxymycobactin (cMBT). To counteract these bacterial iron acquisition processes, the alveolar macrophage in which Mtb thrives, keeps phagosomal iron levels extremely low by the natural resistance-associated macrophage protein Nramp1 in particular after activation by interferon-γ [2], [3]. MBT and cMBT increase the biologically available iron within the phagosomal compartment almost by 20-fold indicating that the Mtb siderophores can overcome these host defense mechanisms [4]. Furthermore, studies using siderophore biosynthesis and uptake mutants underpin the importance of siderophore-mediated iron acquisition to the virulence of Mtb [5], [6], [7].
In Mtb, siderophore biosynthesis is induced under low-iron conditions. When sufficient iron is available the regulator IdeR represses expression of MBT biosynthesis genes mbtA-N. The inner membrane transporter IrtAB and the Esx-3 secretion machinery are required for utilization and uptake of siderophores [6], [7], [8]. In M. smegmatis, export of the siderophore exochelin was shown to be mediated by ABC-like exporter ExiT [9]. Given mycobacteria's unique outer membrane [10], it is likely that a siderophore secretion system of Mtb requires both inner and outer membrane components [11], similarly to the EntS-TolC system of E. coli [12], [13].
In this study, we examined two iron-regulated genes encoding predicted outer membrane proteins MmpS4 and MmpS5. We show that either MmpS4 or MmpS5 is required for growth of Mtb under low iron conditions. While single mmpS4 or mmpS5 deletion mutants do not exhibit a low iron growth phenotype, they have diminished virulence compared to the wild-type strain. Deletion of both mmpS4 and mmpS5 drastically decreases synthesis and secretion of siderophores in Mtb and greatly reduces its virulence in mice. Subcellular fractionation reveals that MmpS4 and MmpS5 are membrane associated. This study identifies MmpS4 and MmpS5 as the first components of a novel siderophore export system that is crucial for survival of Mtb in its host.
To investigate whether MmpS4 and MmpS5 are important for growth under iron-deplete conditions, mutants with in-frame deletions of the mmpS4 and mmpS5 genes were constructed using homologous recombination in both virulent Mtb H37Rv and avirulent Mtb mc26230 (ΔRD1, ΔpanCD; Table S1). Since no low-iron growth defect was observed with the single ΔmmpS4 and ΔmmpS5 mutants (Figs. 1, S1–2), and expression of both mmpS4 and mmpS5 is induced under iron-limited conditions [14] we suspected that they might have redundant functions. Therefore, we constructed a double mmpS4/mmpS5 mutant using the single ΔmmpS5 mutant as the parent strain in both virulent and avirulent Mtb. The mutant strains were unmarked by site-specific recombination and confirmed by Southern blot analysis (Fig. S3). Western blot experiments demonstrated the absence of MmpS4 and MmpS5 in the double mutant and in the respective single mutants, while wild-type levels of both proteins were observed in the complemented strains (Figs. 1A, S4). No differences in growth of the single ΔmmpS4 and ΔmmpS5 strains were observed on self-made low iron glycerol-alanine salts (GAS) agar plates (Fig. 1B). By contrast, the ΔmmpS4/S5 mutant did not grow on GAS agar plates (Fig. 1B). Growth of the double mutant was partially rescued when GAS agar plates were supplemented with 5 µM hemoglobin that was previously shown to function as an iron source for Mtb (Fig. 1C) [15]. However, in liquid medium, the addition of hemin completely rescued the growth of the ΔmmpS4/S5 mutant (Fig. S1) verifying that the growth defect of this strain is indeed iron dependent. Complementation of the ΔmmpS4/S5 mutant with mmpS4 and mmpS5 restored growth on low iron plates to wt levels (Fig. 1B). Interestingly, blocking siderophore biosynthesis by insertion of a resistance cassette into mbtD in the ΔmmpS4/S5 double deletion mutant (ΔmmpS4/S5/ΔmbtD::hyg strain) also restored growth on hemoglobin plates to the level of the wt strain (Fig. 1C). These results indicate that siderophore biosynthesis impairs the growth of this mutant despite the availability of an alternative iron source.
To provide further evidence for the iron growth defect of the ΔmmpS4/S5 double mutant, growth experiments were conducted in various low iron conditions that included self-made low iron 7H9 medium, or the addition of 2,2′-dipyridyl (DIP) or desferrioxamine (DFO) as ferrous and ferric specific chelators, respectively, to standard 7H9 medium (Figs. S1–2, S5–6). Under each low iron growth condition, ΔmmpS4/S5 failed to grow. Unlike on solid media, in iron-replete liquid media, ΔmmpS4/S5 had only a slightly delayed growth phenotype and eventually reached optical densities equal to the wt strain. It is concluded that deletion of mmpS4 and mmpS5 confers a low-iron growth defect phenotype in Mtb.
The inability of Mtb ΔmmpS4/S5 to grow under iron-limiting conditions may be due to defects in siderophore biosynthesis, iron sensing, uptake or secretion of siderophores. Recently, a biosynthetic pathway has been proposed based on the substrate specificities of enzymes encoded by the mbt gene cluster [16] which accounts for all enzymatic activities required for MBT biosynthesis. Therefore, it is unlikely that MmpS4 and MmpS5 play a direct role in biosynthesis of MBT and cMBT. To test whether the ability of the ΔmmpS4/S5 mutant is impaired in sensing low iron conditions, we utilized a gfp-based iron-regulated reporter construct [17]. Under low iron conditions, transcription from IdeR-regulated promoters was induced in wt Mtb containing the reporter construct as indicated by a strongly increased GFP fluorescence (Fig. S7). However, when wt Mtb was grown under high iron conditions, only background fluorescence was observed confirming that Mtb senses iron availability (Fig. 2A). To examine the iron sensing capability of the ΔmmpS4/S5 mutant we exploited the observation that removal of the antibiotic resistance cassette from the MBT biosynthesis mutant ML1600 (ΔmbtD::hyg) [18] resulted in the strain ML1610 (ΔmbtD::loxP) (Table S1) with only a partial low-iron growth defect in vitro (Fig. S8). This result suggests that replacing mbtD with the hyg cassette inhibits expression of downstream genes thereby completely eliminating siderophore production. IdeR-regulated promoters are induced under high-iron conditions in Mtb ΔmbtD::loxP, but the addition of exogenous cMBT to this mutant repressed these promoters, demonstrating that this mutant is capable of sensing iron availability and is suitable as a control strain (Fig. 2A). Likewise, IdeR-regulated promoters were induced in the ΔmmpS4/S5 mutant under high iron conditions, but were repressed after addition of cMBT (Fig. 2A), demonstrating that the ΔmmpS4/S5 mutant is capable of sensing iron availability.
To test whether MmpS4 and MmpS5 are involved in siderophore uptake we monitored the accumulation of 55Fe-loaded cMBT. Iron-loaded cMBT has been shown to donate its iron to MBT in the cell envelope of Mtb in addition to being taken up via the inner membrane ABC-transporter IrtAB [8], [19]. To rule out the possibility that differences in MBT levels affected the measured iron uptake rates, we examined 55Fe-cMBT uptake at 37°C in the siderophore biosynthesis mutant ΔmbtD::hyg and the triple mutant ΔmmpS4/S5/ΔmbtD::hyg (Fig. 2B). Despite the absence of MBTs/cMTBs no differences were observed in the amount of iron accumulated by these strains. Another control experiment showed that only background 55Fe levels were associated with cells at 4°C, indicating that the cell-associated 55Fe observed at 37°C was indeed transported inside the cell and not adsorbed at the cell surface (not shown). Taken together, these results demonstrate that MmpS4 and MmpS5 are not involved in uptake of cMBT.
The low iron growth defect of the ΔmmpS4/S5 mutant is not caused by an iron sensing defect or by lack of cMBT uptake. An alternative explanation might be a defect in secretion of cMBT. To this end, cMBTs in wt Mtb and in the ΔmmpS4/S5 mutant were radioactively labeled by feeding the bacteria the biosynthetic precursor 7-[14C]-salicylic acid. Cell-associated and secreted siderophores were extracted using chloroform from cell pellets and from the culture filtrate and analyzed by thin-layer chromatography (TLC). As controls, purified and deferrated MBTs/cMBTs from M. bovis BCG were loaded with 55Fe and used to visualize siderophore spots. TLC analysis demonstrated that purified siderophores from M. bovis BCG had the same Rf values as siderophores from Mtb validating their use as controls (Fig. 3). The extracts of cell pellets and of culture supernatants showed that the single deletion mutants Mtb ΔmmpS4 and ΔmmpS5 synthesized and secreted siderophores as wt Mtb (Fig. 3). By contrast, the double deletion mutant ΔmmpS4/S5 produced much less cell-associated and secreted siderophores compared to wt Mtb, but was still capable of synthesizing siderophores in contrast to the ΔmbtD::hyg mutant (Fig. 3). It should be noted that MBT was detected in the culture supernatants of Mtb in addition to cMBT. This is most likely caused by partitioning of cell surface-associated MBT with the medium in the presence of detergents. Taken together, these results suggest that MmpS4 and MmpS5 are part of siderophore export system of Mtb.
MmpS4 in M. smegmatis was previously shown to be involved in biosynthesis and export of glycopeptidolipids (GPLs) [20] which Mtb does not synthesize. To examine whether the deletion of mmpS4 and mmpS5 caused an altered lipid profile, we performed a complete lipid analysis by TLC (Figs. S9, S10). All major lipids of Mtb were identified in wt Mtb and the ΔmmpS4/S5 mutant (Fig. S9, S10) indicating that MmpS4 and MmpS5 are not involved in lipid biosynthesis. However, a lipid which was shown to be produced by Mtb under iron limiting conditions [21] was not identified in the ΔmmpS4/S5 mutant (Figs. S10A–C). Bacon et al. [20] showed by 1H-NMR that this lipid consists of a long alkyl chain with a cis double bond and an ester unit. It is unclear whether the absence of this lipid is a direct consequence of the lack of MmpS4/S5, or might be caused indirectly by the slow growth of the double mutant under iron-limiting conditions.
Our data suggests that MmpS4 and MmpS5 are involved in siderophore export, but it is unclear how MmpS4 and MmpS5 contribute to MBT transport. Proteomic analysis of subcellular fractions of Mtb yielded contradictory results regarding the localization of MmpS4 and MmpS5 [22], [23]. In order to determine the subcellular localization of MmpS4 and MmpS5, the culture filtrate containing secreted proteins was prepared. Membrane and cytoplasmic fractions were obtained by ultracentrifugation of cell lysates of Mtb. Both MmpS4 and MmpS5 were present in the membrane, but not in the cytoplasmic or secreted fractions (Fig. 4A). All fractions were well separated as indicated by the control proteins, the membrane-associated OmpATb (Rv0899), the cytoplasmic regulator IdeR and the secreted Ag85 protein (Fig. 4A). Thus, MmpS4 and MmpS5 are the first examples of membrane-associated proteins that are required for export of siderophores in Mtb.
The strongly reduced MBT/cMBT level is a striking phenotype considering the intact biosynthesis capacity of the Mtb ΔmmpS4/mmpS5 mutant. Based on the previous observation that MmpS4 connects glycopeptidolipid biosynthesis enzymes with the MmpL4 transporter in M. smegmatis [20] we hypothesized that MmpS4 might provide a link between MBT/cMBT biosynthesis and export in Mtb. However, in vivo crosslinking experiments with formaldehyde in the avirulent Mtb strain mc26230 (Table S1) expressing a chromosomal copy of a gene encoding hexahistidine- and HA-tagged MbtG did not show direct binding of MbtG to MmpS4. Next, we examined the subcellular localization of MbtG, the lysine monooxygenase which activates MBT/cMBT by hydroxylating dideoxymycobactins as the predicted last step in MBT biosynthesis [24]. In order to catalyze this reaction MbtG has to be in the cytoplasm because it requires access to the cytoplasmic co-factors NADPH and FAD+. Subcellular fractionation experiments in wt Mtb mc26230 revealed that MbtG is membrane-associated although no transmembrane helices and no signal peptide are apparent (Fig. 4B). This result indicates that MbtG might fractionate with membranes due to interactions with another protein and provides the first hint how MBT/cMBT biosynthesis and export might be coupled in Mtb.
The mmpS genes are located in operons with mmpL genes [25]. In order to genetically determine if MmpS4 and MmpS5 interact with their cognate MmpL proteins, the triple mutants ΔmmpS4/L4/S5 and ΔmmpS4/S5/L5 were constructed from the ΔmmpS5 and ΔmmpS4 strains, respectively, by deleting the respective mmpSL operon (Fig. S11). Similar to the double deletion mutant ΔmmpS4/S5, these triple mutants failed to grow in iron-deplete media (Fig. 5A). To this end, each triple mutant was complemented with either an empty integrative vector or integrative vectors containing either mmpS4 or mmpS5. The mmpL5 containing strain (ΔmmpS4/L4/S5) complemented with either mmpS4 or mmpS5 grew in low iron medium (Middlebrook 7H9 supplemented with 50 µM DIP) (Fig. 5A). However, the mmpL4 containing strain (ΔmmpS4/S5/L5) was only complemented with mmpS4 but not with mmpS5. These results indicate that MmpL4 only interacts with its cognate MmpS4 protein, while MmpL5 is capable of interacting with MmpS4 and MmpS5 to mediate siderophore export by Mtb.
To confirm and further define the interaction between MmpS4 and MmpL4, an in vitro pull-down assay was employed. According to the topology predictions MmpS4 possesses an N-terminal transmembrane (TM) helix and a C-terminal soluble domain, while MmpL4 contains eleven TM helices and two long loops—L1 between TM1 and TM2, and L2 between TM6 and TM7 (Fig. S12). We tested the interaction between the purified soluble domains of MmpS4 (residues 52–140) and the predicted loops L1 (58–199) and L2 (416–763) of MmpL4. The soluble domain of MmpS4 formed a complex with loop L1 (Fig. 5B), but not with loop L2 (data not shown) of MmpL4. The in vitro interaction of the soluble domain of MmpS4 with loop L1 of MmpL4 also shows that both peptides form independently folding domains.
The mmpS4 gene encoding an N-terminally truncated MmpS4 protein lacking the predicted transmembrane helix was expressed in E. coli and the water-soluble domain of MmpS4 (residues 52–140) was purified by chromatography. The structure of MmpS452–140 was solved by NMR using 762 nuclear Overhauser effect (NOE) and 127 paramagnetic relaxation enhancement (PRE) distance restraints, and 122 dihedral angle restraints (Table S4). The 20 lowest energy structures were selected out of 200 accepted structures. The statistics about the quality and precision of these structures is summarized in Table S4. The backbone superimposition of the final 20 conformers and the representative structure are presented in Fig. 6A. The MmpS4 structure shows seven consecutive β-strands and an unstructured C-terminus (residues 131–140) (Fig. 6B) which might be due to the lack of resonance assignment in this region. The seven β-strands are arranged in two layers, with β4-β1-β6-β7 in one layer and β3-β2-β5 in the other layer.
To assess the role of MmpS4 and MmpS5 for virulence of Mtb, BALB/c mice were infected with low dose aerosols containing the Mtb H37Rv parent strain ML617, the ΔmmpS4/S5 mutant (ML618), and the double deletion mutant complemented with mmpS5 (ML619), mmpS4 (ML620), and mmpS4/S5 (ML624). The growth kinetics of the parent Mtb H37Rv strain in lungs showed the expected logarithmic increase during the acute phase followed by a plateau during the chronic phase of infection. Similar growth kinetics in spleens demonstrated that this strain is competent for dissemination. Loss of the single mmpS4 and mmpS5 genes also compromised the ability of Mtb to survive in the lungs as the number of viable bacteria decreased by 100-fold from the initial burden compared to wt Mtb. However, loss of these genes alone did not alter the ability of Mtb to disseminate to and proliferate in the spleen. The ΔmmpS4/S5 mutant failed to proliferate in lungs and spleen as reflected by a 24,000- and 1,800- fold, respectively, decreased bacterial burden compared to wt Mtb after 16 weeks of infection (Fig. 7). Loss of these genes resembles the “severe growth in vivo” (sgiv) phenotype [26] and, to our knowledge, is the strongest in vivo phenotype observed so far for genes involved in iron utilization by Mtb. The single mmpS4 or mmpS5 genes partially complemented the virulence defect of the double mutant. Full complementation of the double mutant by both genes confirmed that the mmpS4 and mmpS5 genes are essential for virulence of Mtb (Fig. 7). Gross mouse lung examination and histological assessment in mice infected with the ΔmmpS4/S5 double deletion strain showed almost no signs of infection (Figs. 8, S13–15). However, lungs of mice infected with either Mtb H37Rv wt or the fully complemented Δmmps4/S5 strain exhibited extensive lesions (Figs. 8, S13) and displayed significant lymphocytic infiltrates (Fig. S14). Lungs of mice infected with the mmpS4 or mmpS5 singly complemented strains showed lesions and lymphocytic infiltrates, but to a much lesser degree than lungs of mice infected with wt or the fully complemented strain.
In survival experiments loss of mmpS4 and mmpS5 severely attenuated virulence of Mtb as none of the mice infected with the ΔmmpS4/S5 double deletion mutant died within 180 days (Fig. 9). Similarly, loss of either mmpS4 or mmpS5 alone resulted in attenuation of virulence. The difference in mean survival time between the groups of mice infected with wt and the fully complemented strain was longer than expected and could partly be explained by a lower bacterial burden of the fully complemented strain in the lungs. In conclusion, the infection experiments revealed that mmpS4 and mmpS5 are essential for virulence of Mtb in mice.
In this study, we showed that the lack of MmpS4 and MmpS5 strongly reduced siderophore secretion and caused a growth defect of Mtb under low iron conditions. Pull-down experiments demonstrated that the MmpS4 protein forms a complex with the inner membrane transporter MmpL4 in vitro. This observation was corroborated by genetic complementation experiments demonstrating that MmpS4 and MmpS5 interact with their respective MmpL proteins to restore growth of Mtb under iron-limiting conditions. Considering that siderophore uptake is not altered in Mtb lacking mmpS4/S5 and that the MmpL proteins are inner membrane transporter proteins, it is concluded that the respective MmpS/MmpL complexes translocate siderophores across the inner membrane into the periplasmic space. Such a transport process is defined as export [27].
Proteins which enable siderophore export in Mtb have been unknown so far [11], largely because Mtb does not have any proteins resembling known siderophore export systems such as EntS of Escherichia coli [13] or PvdE of Pseudomonas aeruginosa [28]. Previously, MmpS5 and MmpL5 have been implicated in drug efflux due to weak similarity with RND efflux pumps of E. coli [29]. MmpL3, MmpL7 and MmpL8 were shown to export lipids such as trehalose monomycolate [30], [31], [32], phthiocerol dimycocerosate [33], [34], and sulfolipid 1 [35] leading to the hypothesis that the MmpL proteins are lipid transporters. Since carboxymycobactins and in particular mycobactins are quite hydrophobic molecules and have similar chemical properties as lipids, this finding is rather an expansion than a deviation from the rule. Taken together, we conclude that MmpS4/MmpL4 and MmpS5/L5 constitute novel bacterial siderophore export systems.
The lack of the MmpS4/5 proteins also reduced the amount of detectable carboxy/mycobactins suggesting a role in biosynthesis of these siderophores in Mtb. Recently, a biosynthetic pathway comprising all enzymatic activities required for MBT/cMBT biosynthesis has been proposed based on the substrate specificities of enzymes encoded by the mbt operons [16]. Modifying enzymes to generate nonmethylated or α-methylated MBT derivatives have not been identified yet, but they are not expected to alter the total MBT amount. Thus, it is concluded that the strongly reduced siderophore levels in the Mtb ΔmmpS4/S5 mutant likely result from an indirect effect of these proteins on biosynthesis. Indeed, such a mechanism has been proposed for the MmpS4 protein which is required for efficient synthesis and export of surface-exposed glycopeptidolipids (GPL) in M. smegmatis [20]. Co-localization of MmpS4 with FadD23 and MbtH indicated that the GPL biosynthesis enzymes form a multi-protein complex with the membrane proteins MmpS4 and MmpL4a/b in M. smegmatis. Since lack of MmpS4 resulted in enzyme diffusion in the cytoplasm, biosynthesis of GPLs was much less efficient in M. smegmatis [20]. This phenotype was complemented by the Mtb mmpS4 gene indicating that Mtb MmpS4 also enables formation of a biosynthetic multi-enzyme complex at the inner membrane of M. smegmatis. In this study we show, that MmpS4 is involved in siderophore export in Mtb. The fact that the siderophore biosynthesis enzyme MbtG is located at the inner membrane, as shown in this study, supports the hypothesis that a similar multi-enzyme complex for efficient siderophore synthesis and transport exists in Mtb.
In principle, block of transport caused by the mmpS4/mmpS5 deletions and degradation of siderophores as is observed for the ferric enterobactin esterase IroD, IroE, and Fes of E. coli and Salmonella [36] would also explain the low level of MBTs/cMBTs in the Mtb mmpS4/mmpS5 double mutant. However, there are no enzymes in Mtb with similarities to known siderophore esterases. In addition, degradation of imported siderophores to release iron in the cytoplasm is rare and has only been observed for trilactone siderophores such as enterobactin of E. coli and Salmonella [37]. The high energy cost of MBT/cMBT production [38], [39] also argues in favor of a synthesis tightly regulated by the requirement for iron and the capacity to export newly synthesized siderophores. Coupled synthesis and export would also prevent toxic accumulation of siderophores in Mtb as has been observed in other bacteria [40], [41], [42].
An interesting question is whether the MmpL4/MmpS4 or the MmpL5/MmpS5 systems are specific for MBTs or cMBTs. The single mutants clearly produce and secrete both siderophores indicating that the MmpS4/MmpS5 proteins are not specific for either substrate. This conclusion is supported by the observation that the Mtb mmpS4 gene complements the GPL synthesis and transport defect of the M. smegmatis mmpS4 mutant, although GPLs do not exist in Mtb [20]. It is more likely that the transporters themselves, namely MmpL4 and MmpL5, confer specificity for MBTs or cMBTs. This hypothesis is currently under investigation. Interestingly, we observed that both MmpS4 and MmpS5 interact with MmpL5, while MmpS5 cannot restore growth of an Mtb triple mutant ΔmmpS4/S5/L5 expressing only mmpL4 under iron-limiting conditions. Thus, MmpS4 seems to be more promiscuous in its interactions with MmpL proteins. In this regard, it should be noted that the genetic complementation experiments indicate that the MmpS5/MmpL5 pair is more efficient in restoring wt growth of Mtb under low iron conditions. In conclusion, it appears that Mtb ensures efficient siderophore export by employing at least two partially redundant transporters.
The NMR structure of MmpS452–140 revealed no similarity to any protein of known function, but was similar to an uncharacterized protein from Parabacteroides distasonis (PDB: 2LGE). The superimposition of the MmpS4 structure with that of this putative calcium-binding protein showed a root-mean-square deviation of 3.6 Å over the Cα atoms of 75 aligned residues (Fig. S15A) with similar secondary and tertiary structures (Fig. S15B). Secondary structure prediction [43] indicated an eighth β-strand including the residues 131–137 of MmpS4. This gave rise to the hypothesis that the C-terminus of MmpS4 might be unordered in its unbound state, but may form a more stable structure with two 4-stranded sheets when bound to MmpL4. Further experiments are required to provide evidence for this hypothesis. Importantly, the structure of MmpS4 shows no similarity to AcrA [44] or other periplasmic adapter proteins from drug efflux systems of Gram-negative bacteria [45] indicating that the mycobacterial MmpS proteins constitute a novel class of accessory proteins in complex transporter systems.
In this study, we identified a novel siderophore export system of Mtb which is composed of the transporters MmpL4 and MmpL5 and their associated MmpS proteins. Previously, it was proposed that the MmpS proteins function as periplasmic adapter proteins [29] which was based on the low sequence similarities between the transporters of tripartite efflux pumps of Gram-negative bacteria [46] with MmpL proteins [47]. The localization of MmpS4 in the periplasm of M. smegmatis [20] and our observation that the MmpS4/5 proteins interact with their respective MmpL transporters support their role as accessory transport proteins. In addition, we show that, in contrast to their Gram-negative counterparts, the MmpS4/MmpS5 proteins are not only required for export, but also for biosynthesis of cMBT/MBT. Therefore, we hypothesize that MmpS4 functions as a scaffolding protein to couple synthesis and export of MBT/cMBT in Mtb as has been proposed for GPLs in M. smegmatis [20]. The surprising result that the MBT/cMBT activating enzyme MbtG is membrane-associated despite the absence of any recognizable membrane anchor domain suggests that MbtG might interact directly or indirectly with membrane proteins such as MmpL4/L5 in Mtb. These findings are summarized in the model depicted in Fig. 10. Hitherto unknown are the hypothetical outer membrane proteins required for cMBT/MBT secretion to the extracellular medium and for uptake of cMBT. The role of the Esx-3 system in cMBT/MBT-mediated iron acquisition is also unknown [7].
An interesting observation was that growth of the Mtb ΔmmpS4/S5 mutant under low iron conditions was not fully restored by adding hemoglobin as an iron source (Fig. 1). This result is in contrast to the Mtb ΔmbtD::hyg mutant which is unable to synthesize mycobactins [18]. However, the ΔmmpS4/S5 mutant grew like wt Mtb with hemoglobin as the sole iron source when MBT/cMBT biosynthesis was additionally eliminated. These results indicate that low level synthesis of siderophores and their intracellular accumulation due to the lack of export inhibits growth of the ΔmmpS4/S5 mutant, e. g. by chelating iron or other cations from essential proteins of Mtb. The mechanism of this peculiar type of growth inhibition is currently under investigation.
Deletion of both mmpS4 and mmpS5 drastically reduced the virulence of Mtb in mice. Considering the strong growth defect of the ΔmmpS4/S5 mutant under low iron conditions in vitro and the known requirement of siderophore biosynthesis and utilization for growth of Mtb in vivo [5], [6], [7], it is likely that the attenuation of the ΔmmpS4/S5 mutant is due to its inability to take up sufficient iron in the absence of siderophores. However, it cannot be excluded that other functions of MmpS4 and MmpS5 contribute to the virulence defect of the ΔmmpS4/S5 mutant. In favor of this conclusion is the observation that expression of mmpS5 fully restores siderophore export and growth of the Mtb ΔmmpS4/S5 mutant under low iron conditions in vitro, but still has a significant virulence defect in mice. This result indicated that other functions of MmpS4, which are not present in MmpS5, may contribute to the virulence defect of the ΔmmpS4/S5 mutant. Interestingly, the Mtb mmpL4 mutant showed a 10-fold reduced bacterial burden in the lungs of mice [47]. This is consistent with our finding that the number of bacteria of an Mtb strain which lacks only mmpS4 was between 10- and 100-fold lower in the lungs of mice after the acute phase of infection. Similarly, the slight attenuation of an Mtb strain which lacks only mmpS5 is consistent with the in vivo growth defect of an Mtb mmpS5 mutant in transposon site hybridization (TraSH) studies [48]. The loss of virulence of Mtb ΔmmpS4/S5 mutant in mice is much more pronounced than that observed for the irtAB mutant which lacks an ABC transporter required for cMBT uptake [6]. This correlates with the different magnitude of their in vitro phenotypes: While the irtAB mutant showed only a minor growth defect under low iron conditions [6], loss of mmpS4 and mmpS5 completely abolished growth of Mtb under those conditions (Figs. 1, S1, S2).
In this study, we identified that interaction of the membrane proteins MmpS4 and MmpS5 with their cognate MmpL transporters is required for siderophore export in Mtb and propose a model for siderophore secretion. These novel siderophore transport systems are essential for virulence of Mtb in mice. Considering the almost universal requirement of bacterial pathogens for iron [49] it is tempting to speculate that these systems might be good drug targets. However, more work is required to determine whether these two partially redundant transporters can be poisoned by a single drug.
BALB/c mice were obtained from the Charles River Laboratories and were housed and cared for in a pathogen-free biosafety level 3 vivarium facility at Johns Hopkins University. Mice were provided food and water ad libitum as well as appropriate monitoring and clinical care. The protocols used in this study were reviewed and approved by the Johns Hopkins Institutional Animal Care and Use Committee and are described in protocol MO09M101. The Johns Hopkins Animal Care and Use Committee complies with Animal Welfare Act regulations and Public Health Service Policy. Johns Hopkins University also maintains accreditations with the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC) International.
The strains used in this study are listed in the supplement (Table S1). Media, growth conditions and construction of plasmids are described in detail in the supplement (Text S1).
The mmpS4 (rv0451c), mmpS5 (rv0677c), mmpS4/S5, mbtD (rv2831c), and mmpS4/S5/mbtD deletion mutants of Mtb H37Rv and Mtb mc26230 were constructed using a two-step selection strategy as described in SI. Complementation of ΔmmpS4/S5 double deletion mutant of Mtb H37Rv and Mtb mc26230 were performed using L5 and Ms6 phage integration systems as described in the supplement (Text S1).
Low-iron GAS plates were prepared by dissolving 150 mg Bacto Casitone, 2 g K2HPO4, 1 g citric acid, 0.5 g L-alanine, 0.6 g MgCl2•6H20, 0.3 g K2SO4, 1 g NH4Cl, and 8.3 ml 60% glycerol in 450 ml high grade Millipore water (Barnstead Nanopure Diamond; 18.2 MΩ-cm), the pH was adjusted to 6.6 with NaOH and 5 g Agar Noble (BD Biosciencese) was added. The volume was brought up to 500 ml in an acid washed glass bottle (6 M HCl), autoclaved, supplemented with pantothenate, hygromycin, and split into acid washed bottles to which 5 µM human hemoglobin was added when required. Pre-cultures were grown in 7H9 Middlebrook medium supplemented with 10% OADC, 0.2% casamino acids, 24 µg/ml pantothenate, 50 µg/ml hygromycin, 0.02% tyloxapol (7H9 MR) and 20 µM hemin. Once in mid-logarithmic phase (OD600 = 0.5–2.0) cells were filtered through a filter with 5 µm pores and washed once in low-iron GAS medium. Cells were diluted to an OD600 = 0.01 and 10-fold serial dilutions were prepared in low-iron GAS medium. 3 µl of each dilution was deposited on low-iron GAS plates or low-iron + hemoglobin plates using a multi-channel pipette. Plates were incubated for nine weeks at 37°C.
Strains were grown in 7H9 MR medium. Cultures were inoculated in biological triplicate, grown at 37°C and split at mid-logarithmic phase (OD600 = 0.2–0.3). Purified Fe-cMBT-BCG (1 µg/ml) was added to one set of triplicates, while other triplicates were left untreated. Optical densities were determined in 1 cm path length cuvettes by diluting cells to OD600 = 0.1–1 in the above media. Readings were taken every day until stationary phase was reached. Fluorescence intensities reported in Fig. 2A were two days after the addition of Fe-cMBT.
Green fluorescent protein (GFP) fluorescence intensities were determined using a Biotek Synergy HT plate reader with a 485 nm excitation and a 528−/+20 nm emission filter. Fluorescence intensities were normalized to the optical density of the same samples according to the following equation:
Fe-cMBT-BCG (93 µg) was deferrated as previously described by incubation in the presence of 50 mM EDTA pH = 4.0 at 37°C for 18 hours [8]. Precipitated EDTA was pelleted by centrifugation, supernatant was extracted twice with chloroform, washed twice with water and evaporated to dryness. Deferrated residue was suspended in a 1∶1 mixture of EtOH and 50 mM KH2PO4 buffer pH = 7.0. 55Iron- (396 µCi) was added to the mixture and incubated for 1 hour at room temperature (at which point, the solution developed a brown hue). One ml of water was added to the mixture and extracted twice with 2 volumes of chloroform. The chloroform extract was washed twice with water and evaporate to dryness. The material was resuspended in warm ethanol. This preparation yielded 16.2 µM 55Fe-cMBT-BCG with the radioactive concentration of 47.5 µCi/ml.
The strains ΔmbtD::hyg and ΔmmpS4/S5/ΔmbtD::hyg were grown in 7H9 MR medium, and 20 µM hemin to OD600 = 1.0. Cells were washed on ice with a low iron media consisting of 500 µM MgCl2•6H20, 7 µM CaCl2•2H2O, 1 µM NaMoO4•2H2O, 2 µM CoCl2•6H2O, 6 µM MnCl2•4H2O, 7 µM ZnSO4•7H2O, 1 µM CuSO4•5H2O, 15 mM (NH4)2SO4, 12 mM KH2PO4 pH = 6.8, 1% (w/v) glucose, which was supplemented with 10% OADC, and 0.2% casamino acids. Cells were resuspended in the same media to an OD600 of approximately 3.0 on ice. For uptake experiments, 2 ml of cell suspensions were equilibrated at 37°C for 15 min and shaken at approximately 400 rpm. 55Fe-labeled cMBT was added to the cells at a final concentration of 0.25 µM cMBT, 0.45 µCi 55Fe. 200 µl samples were removed at 1, 2, 4, 8, and 16 minutes and added to 400 µl of a killing buffer consisting of 100 mM LiCl, 50 mM EDTA in 4% formaldehyde in Spin-X filter microcentrifuge tubes. Cells were immediately centrifuged and washed twice in killing buffer. The radioactivity of the cells was quantified using liquid scintillation counting (Beckman Coulter LS6500). 55Fe counts were converted to total iron by the use of a standard curve and normalized to dry weight of cells by determining the dry mass of 4 ml of the washed cell suspensions. All experiments were done in triplicate.
Radiolabelling of siderophores was performed in a similar manner as previously described with modifications [5], [50]. Iron free self-made 7H9 media supplemented with 0.2% glucose and 0.01% Tyloxapol was deferrated using Chelex-100 to remove any trace contaminants of iron. Pre-cultures were grown under iron rich conditions to OD600 of 1–2. To deplete intracellular iron stores, iron free media was inoculated with pre-culture to an OD600 = 0.05, and allowed to grow to OD600 of 1–2. Only ML618 (ΔmmpS4/S5) and ML1424 (ΔmbtD::hyg) were not grown in iron free media because these strains do not grow under low iron conditions; however, IdeR derepression occurs even under iron-replete conditions in these strains (this study). Five ml of cell cultures were adjusted to OD600 = 0.2 using iron free media and incubated with 1 µCi/ml [7-14C]-salicylic acid (21.3 µM final concentration) for 11 days while shaking at 37°C. Cultures were centrifuged at 4,000× g for 7 min and supernatants and cell pellets were collected. Ferric chloride (20 mg/ml FeCl3•6H2O in ethanol) was added to the supernatants at a final concentration of 0.6 mM and allowed to incubate at room temperature for one hour. The supernatants were extracted twice with 5 ml CHCl3 and the organic fraction was retained. The cell pellets were resuspended in 2.5 ml of ethanol and incubated with shaking for 12 hours at 37°C. After centrifuging for 7 min at 4,000× g, the ethanol supernatant was retained and 2.5 ml of water and FeCl3 (to 2.2 mM) was added. This mixture was allowed to incubate at room temperature for one hour, after which it was extracted twice with 5 ml CHCl3 and the organic fraction retained. The cell and supernatant extracts were then evaporated using a Vacuufuge (Eppendorf) and resuspended in 500 µl CHCl3. After having normalized based on CPMs, extracts were then subjected to TLC on 10 cm×10 cm, 250-µm-thick silica gel 60 (Sigma) developed in ethanol/hexanes/water/ethyl acetate/acetic acid, 5∶25∶2.5∶35.5 [51]. Plates were allowed to dry and then exposed to a phosphoimager screen for 60 hours and analyzed with a Storm Phosphoimager (Molecular Dynamics). 55Fe-loaded MBT and cMBT, as well as radiolabelled salicylic acid substrate, were also subjected to TLC alongside the extracts. Rf values for MBT (0.42) and cMBT (0.16) were the same as those previously reported and the salicylic acid, MBT, and cMBT loading controls ran the same as their extracted radiolabelled counterparts.
Prior to virulence studies, all strains were demonstrated to have PDIMs and positive neutral red assessments. Mid-log phase cultures of wt Mtb (ML617), the mmpS4/S5 double deletion mutant (ML618), the mmpS5 singly complemented mutant (ML619), the mmps4 singly complemented mutant (ML620), and the doubly mmpS4/S5 complemented mutant (ML624) were diluted to OD600 ∼0.1 to implant ∼1,000 bacilli in the lungs of mice using a Middlebrook inhalation exposure system (Glas-Col). Twenty four 4- to 5-week old female BALB/c mice (Charles River) were infected with ML617, ML618, ML619, ML620, or ML624. Four mice from each group were weighed and sacrificed at days 1, 14, 28, 56, 84, and 112 post-infection to determine the number of bacilli in the lung and spleen. Mouse organs were aseptically removed, homogenized, and serially diluted. Appropriate dilutions were plated onto Middlebrook 7H11 agar plates to determine the colony forming units. For histological analysis, representative tissue samples from each group at days 1, 28, 56, 112 post-infection were fixed in 10% formaldehyde, embedded in paraffin, sectioned, and stained with hematoxylin and eosin using standard procedures. Thirteen 4- to 5-week old female BALB/c mice were infected with 7,500–10,000 bacilli using the five strains mentioned above using the same method already described. Time to death was followed and survival proportions of mice infected with high dose aerosol were calculated. The experiment was stopped after 180 days post-infection.
Detailed protocols for other experiments are provided in the supplement (Text S1).
|
10.1371/journal.pgen.1003159 | Polycomb Controls Gliogenesis by Regulating the Transient Expression of the Gcm/Glide Fate Determinant | The Gcm/Glide transcription factor is transiently expressed and required in the Drosophila nervous system. Threshold Gcm/Glide levels control the glial versus neuronal fate choice, and its perdurance triggers excessive gliogenesis, showing that its tight and dynamic regulation ensures the proper balance between neurons and glia. Here, we present a genetic screen for potential gcm/glide interactors and identify genes encoding chromatin factors of the Trithorax and of the Polycomb groups. These proteins maintain the heritable epigenetic state, among others, of HOX genes throughout development, but their regulatory role on transiently expressed genes remains elusive. Here we show that Polycomb negatively affects Gcm/Glide autoregulation, a positive feedback loop that allows timely accumulation of Gcm/Glide threshold levels. Such temporal fine-tuning of gene expression tightly controls gliogenesis. This work performed at the levels of individual cells reveals an undescribed mode of Polycomb action in the modulation of transiently expressed fate determinants and hence in the acquisition of specific cell identity in the nervous system.
| Epigenetic mechanisms are essential to define cell identity, and the Polycomb and the Trithorax Group proteins (PcG and TrxG, respectively) control the body plan by maintaining the epigenetic state of homeotic genes. PcG and TrxG act by triggering stable chromatin modifications that are “remembered” after cell division and keep gene expression in an OFF or ON state. Recent genome-wide analyses call for additional targets of PcG proteins, but the role of these chromatin factors in dynamic transcriptional states and/or in specific cell fates is difficult to apprehend, mostly because very sensitive readouts are required. This in vivo study performed at the single-cell level shows that PcG proteins affect the levels and the kinetics of the transiently expressed Drosophila glial determinant and transcription factor Gcm/Glide. Thus, PcG proteins also finely tune gene expression, and this is independent of memory mechanisms, suggesting that “transient” promoters may have a different affinity to PcG proteins compared to “stable” promoters. PcG proteins negatively affect Gcm/Glide autoregulation, thereby promoting neurogenesis at the expense of gliogenesis. Thus, PcG genes act in the fate choice between two types of differentiated cells, implying that distinct cell populations have specific requirements for general chromatin modifiers.
| One of the most challenging issues in biology is to elucidate the mechanisms underlying cell fate determination and maintenance. The Drosophila melanogaster Glial cell missing/Glial cell deficient transcription factor (Gcm/Glide, referred throughout the text to as Gcm) is transiently expressed and is key to decide between glial and neuronal fates in the multipotent neural precursors [1]–[6]. Threshold levels of Gcm are necessary and sufficient to induce gliogenesis and the tight regulation of its expression prevents defective/excessive gliogenesis [7]–[11]. These features make Gcm an ideal tool to study cell differentiation and plasticity.
Two major classes of proteins that modify the chromatin structure and its condensation state, the Polycomb group (PcG) and the Trithorax group (TrxG), are known as critical regulators of HOX transcription factors, which act as molecular switches that are maintained in a silent or in an active state [12]. PcG and TrxG proteins act in large multimeric complexes that bind specific DNA regions called Polycomb (or Trithorax) response elements (respectively PREs and TREs) [13]. PcG and TrxG complexes trigger posttranslational modification of histone tails that have opposite effects on gene activity, mainly methylation of H3K27 induced by PcG complexes (negative regulation) and methylation of H3K4, H3K36 as well as acetylation of H3K27 by TrxG complexes (positive regulation) ([12], [14] and references therein). PcG proteins enter two main conserved complexes called Polycomb Repressive Complex 1 and 2 (PRC1 and PRC2). The latter is formed by four core components including, in flies, Enhancer of zeste (E(z)), and catalyzes the reaction that leads to di- and tri-methylation of H3K27. This epigenetic mark is recognized by Polycomb (Pc), which belongs to the PRC1 complex.
Recent chromatin immunoprecipitation studies have shown that PcG and TrxG binding is also associated with dynamic transcriptional states modulating different processes including mitogenic pathways and progression from multipotency to differentiation ([12], [15]–[19] and references therein). Understanding the mode of action of PcG and TrxG proteins in dynamic processes, however, requires analyses at the level of identified cells and times. This is particularly important for developmental genes that are expressed transiently and in specific cell populations. The present in vivo study analyzes the role of Pc in fly gliogenesis.
To identify components and regulators of the Gcm pathway, we designed a screen for genetic modifiers of a dominant phenotype due to gcm ectopic expression and identified PcG and TrxG proteins. Importantly, mutations in PcG components and in TrxG members found in chromatin remodeling complexes enhance the gcm dominant phenotype, whereas mutations in TrxG proteins known to specifically counteract PcG function rescue it. This suggests that a balanced action of these chromatin modifiers regulate Gcm function. Moreover, we demonstrate that the gcm regulatory sequences carry a PRE and are bound by Pc. Finally, Pc inhibits the autoregulatory loop ensuring threshold Gcm levels [7] and hence gliogenesis.
To our knowledge, this is the first direct evidence that PcG proteins negatively modulate a transiently expressed fate determinant, thereby affecting a specific lineage in the nervous system.
The need of tight Gcm regulation prompted us to screen for interactors using a sensitized background. This approach allows the dissection of molecular cascades when the loss of a gene product is embryonic lethal. The Drosophila thorax (notum) carries a stereotyped number of sensory organs called macrochaete or bristles. gcmPyx/+ flies ectopically express gcm in the larval notum, which triggers the differentiation of supernumerary sensory organ precursors (SOPs) and bristles (Figure 1A–1C) [20]. gcmPyx/+ females show, in average, 18,5 bristles instead of the 11/heminotum typical of wild-type (wt) animals. Using large overlapping deficiencies (67–75% genome coverage, Deficiency kit, Bloomington), we performed a primary screen and identified 42 genomic regions that dominantly enhance or suppress the gcmPyx dominant phenotype when deleted (Figure 1D–1E, Figure S1Aa and S1B). These regions were selected for quantitative analyses (Figure S1Ab), which identified weakly and strongly modifying deficiencies. We further analyzed the latter ones (Figure S1Ab, S1B) and identified 28 interacting genomic regions. A secondary quantitative screen with smaller deficiencies (Figure S1Ac, Table 1) allowed us to identify those that act as strong modifiers, based on statistical analyses. Single gene loss of function mutations within those deficiencies were then analyzed and the interaction was confirmed for 18 of them (Figure S1Ad, Figure S2, Table 1). In sum, the Deficiency kit allowed us to identify large interacting regions and to progressively refine the analysis to single mutations.
To evaluate the specificity and the sensitivity of the screen, we asked whether the selected deficiencies eliminate genes expected to interact with gcm. The gcmPyx phenotype correlates with the ectopic formation of proneural territories and precursors of the central (CNS) and peripheral (PNS) nervous systems, neuroblasts (NB) and SOPs, respectively [21], [22]. Thus, mutations of NB/SOP specific genes should act as gcmPyx suppressors and indeed, the large and the small deficiencies covering three genes – escargot (esg), worniu (wor) and snail (sna) – expressed in most embryonic NBs act as gcmPyx suppressors (Table 1, Figure S1C). Testing single gene loss of functions confirmed that sna and esg mutations act as gcmPyx suppressors. Accordingly, esg overexpression triggers the opposite phenotype (Figure S1C). Finally, genes as pimples (pim) and crooked legs (crol), identified in a microarray as induced by Gcm [23], were also identified in our screen (Figure S1D).
The fact that known and predicted gcm interactors were identified validates our screen and shows that the dominant bristle phenotype is a reliable and very sensitive readout.
A genomic region identified in the screen covers the trxG gene brahma (brm), which encodes a transcriptional coactivator related to yeast SWI/SNF proteins and plays a role in ATP-dependent nucleosomal remodeling [24]. The large and the small deficiencies covering brm, Df(3L)brm11, Df(3L)th102 and, most importantly, a null brm allele, enhance the gcmPyx phenotype (Figure 1F). To extend our findings, we tested osa, an integral component of the Brahma complex [25]. osa loss of function also enhances the gcmPyx phenotype, whereas osa gain of function (GOF: hs-Gal4;UAS-osa flies) suppresses it (Figure 1G, 1I). Thus, osa acts as brm, moreover, double brm/osa heterozygous mutants show an even stronger phenotype.
Furthermore, a deficiency covering Enhancer of bithorax (E(bx)) and the E(bx) mutation itself enhance the gcmPyx phenotype (Table 1, Figure 1G, 1I). Interestingly, E(bx) (also called NURF301) encodes a transcription coactivator that belongs to the ISWI chromatin remodeler complex, another TrxG complex, and negatively regulates the JAK-STAT pathway [26], which is known to interact with gcm [27].
We then tested members of two TrxG complexes that specifically counteract Pc function. Trx is a SET-domain containing protein able to induce H3K4 methylation [28]. It has been purified as a subunit of the Drosophila COMPASS-like complex [29] and of the TAC1 complex that combines histone methylating and acetylating activities (reviewed in [30]). The trx null mutation acts as a suppressor of the gcmPyx phenotype (Figure 1G, 1I). Ash1 is a SET-domain protein reported to have histone methyltransferase activity [30]: its null mutation also suppresses the gcmPyx phenotype (Figure 1G, 1I). Finally, the Drosophila CREBS-binding protein (dCBP) encoded by nejire (nej) is responsible for H3K27 acetylation [31] and is associated with both TAC1 and ASH1 complexes. The nej null mutation suppresses the gcmPyx phenotype (Figure 1G, 1I). In conclusion, we found that mutations in TrxG proteins known to specifically counteract PcG function [12] act as suppressors of the gcmPyx phenotype, whereas TrxG members found in chromatin remodeling complexes that are involved in more general transcriptional regulation act as enhancers. We therefore tested members of the two PcG complexes, PRC1 (Pc) (three null alleles) and PRC2 ((esc), E(z)), as well as the PcG protein recruiter pipsqueak (psq). Mutations in all four genes enhance the gcmPyx phenotype (Figure 1H, 1I). Thus, PcG mutations act in the same way as mutations in the TrxG genes brm and osa, but have opposing effects compared to mutations in the TrxG genes Ash1, trx and nej. This suggests that a balanced action of these chromatin modifiers regulate gcm function.
In sum, the screen allowed the identification of several chromatin factors as gcm genetic interactors.
gcm was identified as a putative Pc target in genome-wide chromatin immunoprecipitation (ChIP) studies on Drosophila embryos and different cell lines [32]–[34], we therefore focused on this chromatin factor. As seen in Figure 2B, a Polycomb Response Element (PRE) is present around the transcription start sites (TSS) of gcm and gcm2, which are organized head to head in a 30 kb region [35]. PRC1 binding at the TSS is accompanied by the H3K27 methylation mark (H3K27me3), the profile of which is much broader, extending throughout the gcm-gcm2 5′ regulatory region. As expected, the profile of H3K4methylation complements that of H3K27me3 (Figure 2B). Pc binding was further validated and quantified by qChIP analysis on specific regions including the TSS region for each gene (gcm, gcm2), or an adjacent region (GlacAT) and a negative control (Rp49) (Figure 2A, 2B).
We then asked whether the upstream region of the gcm gene bound by PcG proteins is able to recruit PcG proteins in transgenic assays. For this, we examined PcG binding to a transgene containing the upstream region of the gcm locus on salivary gland chromosomes by Immuno-FISH experiments. Similar to the endogenous gcm locus, which associates with both Pc and Ph proteins (Figure 2E–2F′″), a transgene carrying a gcm construct including 9 kb from the promoter region (9 kb gcm) induces the recruitment of PcG proteins to an ectopic site (Figure 2B, 2G–2Hb′″). Interestingly, a transgene carrying a shorter construct (2 kb gcm) is not able to efficiently recruit PcG proteins (Figure 2B, Figure S3). Importantly, this shorter construct triggers very limited rescue when reintroduced in gcm mutant embryos, whereas the 9 kb gcm construct rescues the embryonic mutant phenotype almost completely [8], suggesting a correlation between Pc binding and transgene activity. Of note, the transgenes do not contain gcm2, excluding the requirement of a gene complex for Pc binding. Moreover, gcm2 plays a minor role in gliogenesis and its mutation is viable [35] allowing us to focus on gcm.
Finally, we tested the 9 kb construct for pairing sensitive silencing (PSS), as transgenes carrying PREs/TREs in Drosophila have been shown to share this property ([36], [37]). Transgenic flies carrying the mini-white gene typically have eye colors ranging from yellow to orange in a white mutant background. Normally, flies that are homozygous for such a transgene have a darker eye color than heterozygotes, as the genetic dose of mini-white is doubled. However, with transgenes carrying PRE/TREs, the eye color is similar in homozygotes and heterozygotes or even darker in the latter. This is what we also observed in our transgenic lines (Figure 2C–2D).
Altogether, these data indicate that the gcm promoter region contains a PRE and suggest that PcG proteins directly regulate gcm expression.
We next scored for Pc gcm interaction in a physiological asset, i.e., in loss of function conditions for both genes. The gcm-Gal4 line, an insertion in the gcm locus, is a hypomorphic semiviable allele in homozygous conditions and can be used to follow gcm activation and glial cells using a UAS-green fluorescent protein (GFP) line [38]–[40]. We analyzed the expression of GFP as well as that of an independent glial marker (Repo) and a neuronal marker (Elav) in homozygous gcm-Gal4,UAS-GFP (referred to as gcm-Gal4) animals and in homozygous gcm-Gal4 animals that are also heterozygous for Pc. As a control, we used heterozygous gcm-Gal4 animals.
The Drosophila wing contains two major nerves, L1 and L3, covered by glia that depend on gcm [41] (Figure 3A, 3B). Because of their simple organization, we focused on the L3 glia, which arise from three SOPs called L3-3, L3-1 and L3-v. Each SOP produces a sensory neuron and a glial precursor (GP) that proliferates and produces four to eight glia that are GFP+ (Figure 3C–3H).
gcm-Gal4 homozygous flies show the glia to neuron transformation observed in gcm null clones [41], albeit at much lower penetrance (Figure 3I–3M). To analyze the phenotype at single cell level, we followed glia from a specific lineage, the L3-v, at the time the GP is generated. At this stage, control L3-v lineages contain a GFP+ cell that expresses Repo and a neuron that expresses Elav (Figure 3S, 3T–3T″). In gcm-Gal4 homozygous animals, the GFP+ cell expresses Elav rather than Repo (9% penetrance) (Figure 3S, 3U–3U″). By 24 hr after puparium formation (APF), the number of GFP+ and Repo+ cells present in the control animals increases, whereas only one GFP+ cell is present in the transformed lineage, due to lack of proliferation, and this cell is a neuron (Figure 3C–3M). The penetrance of ectopic neurons does not decrease during development (16 and 18% by 20 and 24 hr APF, respectively, Figure 3S), indicating that low Gcm levels trigger a stable fate conversion; a similar phenotype was observed on L1 glia (Kumar and Giangrande, unpublished data).
Based on the genetic data, we then asked whether Pc downregulation rescues the phenotype of homozygous gcm-Gal4 wings. Indeed, no evidence of stable glia to neuron transformation was found in homozygous gcm-Gal4 wings that carry only one Pc functional allele (Figure 3N–3S). The phenotype was verified at early and at late stages of wing development, to exclude the possibility of unstable rescue. These data strongly suggest that Pc affects gcm expression in the gcm-Gal4 line.
In order to extend the above findings, we analyzed late gliogenesis upon lowering the dose of Pc. Differentiated gcm-Gal4 homozygous wings carry fewer glia than wt wings in which the three glial precursors have divided more than once in most of the cases (Figure S4A, 24 hr APF wings). Given the low penetrance of the fate transformation phenotype, this suggests an additional, later, effect on the glial cell number. To clarify the nature of the phenotype we counted the Repo+ cells just after the first division of the three L3 GPs in gcm-Gal4 homozygous wings that showed no fate transformation. We could confirm a decreased number of cells (Figure 4B, 4E, S4B, 20 hr APF wings), complementing the finding that sustained gcm expression induces glial overproliferation (embryo: [11]; wing: Kumar and Giangrande, unpublished data). Of note, the gcm-Gal4/+ wings already show a minor but consistent defect as there are cases in which the three GPs have not proliferated yet, which does not occur in wild type wings of the same stage (Figure 4A and 4E, Figure S4B). Moreover, heterozygous wings show a high variance in the number of Repo+ cells. Finally, homozygous gcm-Gal4 wings expressing a single Pc show a higher number of glia compared to those found in homozygous gcm-Gal4 wings (Figure 4D and 4E, Figure S4B), confirming that Pc negatively controls Gcm. This was confirmed by the significant P values obtained with different robust non-parametric tests comparing the homozygous wings with the homozygous wings that carry one dose of Pc (Mann Whitney test P = 0,0127; Wilcoxon test P = 0,0122). Moreover, one-way Anova comparison of the three genotypes (gcm-Gal4/+, gcm-Gal4 and gcm-Gal4; Pc/+) also produces a significant value (0,0028). These data indicate a partial rescue of the gcm-Gal4 proliferation phenotype by Pc, the moderate difference likely depending on the fact that only one dose of Pc is deleted.
To understand the role of Pc in gliogenesis, we also analyzed Pc mutant animals in an otherwise wt background and asked whether the mutation affects the number of glia (Figure 4F–4H, Figure S4C 24 hr APF wings) and the frequency of glial dividing cells (Figure 4I). Since removing Pc completely leads to pleiotropic defects, we used heterozygous Pc animals and counted the number of Repo+ cells on the L1 nerve, which shows massive gliogenesis, compared to the sparse glial cells present on the L3 nerve [41]. While the number of Repo+ cells increases very moderately in Pc/+ compared to wt wings (P = 0,03), a stronger, significant, increase is observed in E(z)/+ wings (P = 0,0007), which have a compromised PRC2, and an even stronger phenotype is observed in double heterozygous Pc/E(z) animals (P = 3,9×10−6), which display compromised PRC2 and PRC1 (Figure 4F–4H, Figure S4C). Finally, we labeled wings with Repo and phospho-histone H3 (PH3) as a mitotic marker. By 24 hr APF, the Repo/PH3+ cells are very rare in wt wings (1 Repo-PH3+ cell in 1/11 wings) (Figure 4I). E(z)/+ or Pc/E(z) double heterozygous animals, which show the most significant increase in glial cell number, show a significant increase in the number of wings with proliferating glia, whereas Pc/+ animals, in which the increase in glial cell number very small, do not. Thus, PcG proteins likely synergize and affect both glial differentiation and proliferation.
We next analyzed the role of Pc on the gcm expression profile. Positional cues first trigger initiation of transcription, then Gcm positively autoregulates [7] and, as the glial fate is established, gcm expression progressively decreases so that its transcripts are no longer present in mature glia [42]. We analyzed the initiation of gcm transcription in gcm-Gal4/+; Pc/+ wings. Previous analyses showed that the gcm RNA becomes detectable by 8–9 hr APF (Van de Bor and Giangrande, unpublished data). We therefore analyzed 7–8 hr APF wings and found that the GFP appears at the same time as in wt animals (data not shown). Since the binary Gal4 system may not faithfully reproduce the temporal pattern, we analyzed wings carrying one dose of Pc and the P-mediated insertional gcmrA87 allele expressing the LacZ reporter and confirmed that the β-Gal labeling starts as in wt animals (Figure S5). The finding that Pc does not affect initiation of gcm expression is in line with the wt number of GFP+ cells observed in homozygous gcm-Gal4 wings at early stages, even in cases in which glial cells convert into neurons.
We also performed in situ hybridization with a gcm-specific probe in Pc/+ wings. We took advantage of the supernumerary glia phenotype to see whether Pc helps repressing the maintenance of gcm expression. gcm transcripts are well visible on both wt and Pc/+ wings by 19 hr APF, a stage at which the glial precursors have differentiated (Figure 5A, 5D). By 24 hr APF, however, they are absent in wt, but still present in Pc/+ wings (Figure 5B, 5E), which correlates with the slight increase in glial number observed in Pc/+ animals. Interestingly, Pc/+ wings do not show gcm expression at ectopic positions, suggesting that the absence of Pc induces a failure in repressing gcm maintenance rather than a global loss of silencing in whole tissues.
We extended the data by analyzing other stages and tissues. In the brain, gcm is expressed in several cell populations: GPC and its glial progeny, lamina neurons, central brain neurons and medulla glia [39], [40], [43], [44]. We focused on gcm expression at the position of lamina glial precursors (GPCs), which produce numerous cells that migrate and form the glia of the lamina visual ganglion (Figure 5K) [39], [40], [43]. For the sake of simplicity, we analyzed the optic lobes at a stage at which gcm is detectable in the GPC area but just starts being expressed in the other regions. In wt animals, gcm expression fades away as glia differentiate and migrate (Figure 5G, 5K, 5L), whereas in Pc/+ animals gcm is expressed in an expanded area (Figure 5H, 5L). Moreover, gcm is overexpressed in brm/+ brains and this phenotype is suppressed in brm,trx/+ animals. This shows that brm acts similar to Pc on gcm expression, and both act antagonistically to trx, in line with the genetic data (Figure 5H–5J, 5L). All the phenotypes were quantified by comparing the intensity and the area of the gcm signal (see Text S1, Figure S4E). In the double mutant, the area of labeling resembles that observed in wt animals and the intensity of the signal is even lower than that observed in wt animals. Future analyses will determine whether the increase of gcm expression in the mutant backgrounds reflects longer perdurance in migrating glia, production of more glia or production of more glial progenitor cells in the larval lamina. In some preparations, labeling in other regions is also observed, depending on sample orientation. Even though we cannot formally exclude the possibility that this represents ectopic labeling, these regions correspond to the other positions at which gcm accumulates at slightly later stages in wild type animals, suggesting that in those regions as well Pc negatively controls gcm expression.
Finally, we analyzed gcm transcripts in Pc embryos. In wt animals, gcm is expressed at early stages of glial development and transcripts subsequently fade away, first in the ventral cord and then in the brain [42]. The most frequent phenotype of Pc mutant embryos is a persisting gcm expression in the brain, but we also found extreme cases of late gcm expression in the ventral cord (Figure 5C, 5F). The embryonic and the postembryonic brains contain too compact and numerous glia and the perdurance in the ventral cord is a rare event, likely due to the Pc maternal component.
Although these tissues/stages do not allow quantitative analyses of glial cells, the expression data and the wing phenotype strongly suggest that Pc represses gcm maintenance. Altogether, our observations highlight the importance of Pc in tightly regulating Gcm levels.
To assess whether Pc directly represses gcm, we used in vivo and in vitro assays. Gcm directly and positively autoregulates and alteration of this feedback loop severely affects its gliogenic potential, providing further evidence for the importance of Gcm maintenance at a precise developmental time [7], [9]. In vivo autoregulation can be documented in gain of function experiments by using the gcmrA87 allele. We asked whether Pc negatively controls Gcm autoregulation by comparing animals that simultaneously overexpress Gcm and Pc to control animals that only overexpress Gcm. Compared to controls, Pc and Gcm cooverexpressing embryos show a drastic reduction in the number of β-Gal+ cells as well in the intensity of β-Gal labeling (Figure 6A, 6C). Accordingly, co-overexpression reduces the number of ectopic glia as assessed by the Repo marker (Figure 6D, 6F, 6G, 6I). Moreover, and in line with these results, overexpressing Gcm in Pc loss of function embryos triggers a significant increase in the number of autoregulating cells compared to that observed in control animals (Figure 6A, 6B). Accordingly, these animals show an increased number of ectopic Repo+ cells (Figure 6D, 6E, 6G, 6H). These data were quantified upon counting the number of β-Gal+ and Repo+ cells (Figure 6J). Loss and gain of function of Pc do not, on their own, alter the expression of the Repo marker (Figure S6).
To evaluate whether the inhibitory effects of Pc in the Gcm pathway are direct, we used transactivation assays in which we transfected S2 cells with a Gcm expression vector and a reporter of its activity in presence or in absence of a Pc expression vector.
We first analyzed the repo promoter, a major direct Gcm target that contains several Glide Binding Sites (GBSs) [45] (Figure S7D). This promoter is inactive in S2 cells, but Gcm expression is sufficient to activate it. Upon cotransfection with Gcm and Pc expression vectors, however, the transactivation induced by Gcm decreases significantly (Figure S7C, S7D). We repeated the same type of assay using a second, transiently expressed, promoter depending on gcm. The gcm2 2 kb proximal promoter contains four GBSs and was previously shown to be activated by Gcm in transfection assays [35] (Figure 6K, 6L), more robustly than the gcm 2 kb promoter itself, which only contains one GBS. As for repo, the cotransfection with Gcm and Pc reduces the activation of the gcm2 promoter. Thus, Pc represses the expression of Gcm stably and transiently expressed targets.
In sum, the above data support the hypothesis that Pc represses gcm autoregulation and Gcm downstream targets, thereby inhibiting glial development.
Cell fate determination and maintenance require pathways that finely modulate gene expression and hence ensure the proper balance of cell types in metazoa. The pleiotropic and genome-wide effects of such pathways still hamper clear understanding of their impact and mode of action at single cell level. Our screen and genetic analyses in the Drosophila model unveil the role of the Polycomb chromatin modifier in the generation of glial cells upon fine modulation of the transiently expressed fate determinant gcm.
The genetic screen over a sensitized background proved to be an extremely sensitive tool, as it allowed us to identify several genes that in heterozygous conditions are able to modify the strong dominant gcmPyx phenotype. The screen also provided hints onto the function of the interactors, suppressors or enhancers of a given phenotype. For example, sna and esg act as gcmPyx suppressors, in line with the fact that gcmPyx triggers the expression of NB-specific genes [20]. Identifying an interactor provided an entry point to find members of the same pathway that were initially underscored because located in deficiencies with moderate phenotypes (perhaps due to the presence of genes with opposite effects) or in regions that were not covered by the deficiencies. In the first case is Pc, in the second are osa and Ash1 (Figure S2). The screen also identified members of other signaling pathways (Table 1, Figure S2). One of them depends on Notch (N), which controls gcm expression [20]. While the used Deficiency kit does not cover N itself, we identified Suppressor of Hairless (Su(H)), which regulates the transcription of the N targets, and Lethal (2) giant disc 1, which negatively regulates N receptor trafficking ([46] and references therein). We also tested and validated the genetic interaction with other members of the cascade, including N, its ligand Delta, one of its targets, Enhancer of split, and Groucho, a transcriptional repressor and a partner of Su(H). Future studies will dissect the role of this and of the other pathways on the Gcm cascade.
Several TrxG proteins act as genetic modifiers of the gcmPyx phenotype. TrxG proteins were initially identified as positive regulators of HOX genes and considered as PcG counteractors. In recent years, however, it has become evident that they have a much wider role in gene regulation and it is unclear whether they mainly antagonize PcG functions or whether they globally control gene expression [12]. Interestingly, the three TrxG proteins that behave as positive regulators, Trx, Ash1 and dCBP, are found in TAC and ASH1 complexes that contain a histone acetylation activity. The dCBP histone acetyltransferase present in these complexes acetylates H3K27, a modification that is associated with PcG target genes when they are active [34]. This modification is incompatible with Pc dependent H3K27me3, as these modifications occur on the same amino acid. Thus, Trx- and Ash1-associated dCBP might be a key player in counteracting PcG-dependent silencing of the gcm gene [31]. Future studies will address the role of dCBP onto the Gcm cascade.
osa and brm act as negative regulators of gcm. TrxG proteins can form different complexes that have distinct properties and in some instances repress gene expression. For example, Trx and Brm, which belong to different molecular complexes [30], act positively on the HOX genes and influence a homeotic transformation phenotype in the same way [47], however, Brm-containing complexes mediate transcriptional repression of genes other than the HOX genes [48]. The emerging view is that the SWI/SNF TrxG proteins act as transcriptional activators or repressors depending on the temporal and spatial context [49]. Further studies will determine whether the TrxG proteins acting as negative regulators of gcm directly repress its expression or induce a gcm repressor.
PcG proteins repress homeotic genes to ensure the maintenance of transcriptional states and provide a cellular memory that is transmitted upon cell division, in contrast, their mode of action in the control of more dynamic processes remains elusive. We show in vivo that members of the PcG negatively regulate the gcm pathway during glial fate establishment and proliferation. At least in the first step, a process based on cell memory can be excluded, as Pc acts prior to the division of the GP, the cell in which gcm starts being expressed [41].
The qChIP assay as well as the expression, the S2 cell transfection and the autoregulation data strongly suggest that Pc directly represses gcm transcription maintenance. In addition, the phenotypes observed upon changing the relative gene dosage indicate that Pc and gcm need to be present in appropriate amounts. The importance of an adequate balance between positive (Gcm) and negative (Pc) factors in the establishment of the glial fate is also provided by a rare phenotype observed in a gcm-Gal4; Pc/+ background (1/17 wings) in which the GFP+ cell expresses Repo and Elav, indicating an intermediate glial/neuronal state (Figure 3V–3V″). Thus, Pc acts by finely tuning a transiently expressed fate determinant.
We speculate that the role and the mode of action of chromatin factors depend on the target. HOX promoters, which require to stay in an ON or OFF state, may involve strong binding/high accumulation of chromatin regulators and several studies have already shown that HOX activators drastically reduce K27me3 and also PcG protein binding (Figure 7A) [34], [50], [51]. More dynamically expressed genes may involve less strong binding, a configuration that allows modulation of gene expression. From a mechanistic point of view, as the activator of the transiently expressed genes disappears, PcG proteins may gradually bind and turn these genes OFF (Figure 7B) although we cannot formally exclude that PcG proteins may simply provide a constant repressive background as a threshold for activation (Figure 7C).
In line with these hypotheses, HOX and Gcm display different behaviors. A fragment of 219bp from Fab7, the classical PRE described on a HOX promoter, is sufficient to recruit PcG proteins on salivary glands [52], whereas a 2 kb gcm carrying the PRE seems very inefficient. In addition, the intensity of Pc, Ph and ‘recruiters’ peaks onto the gcm promoter is very low, definitely weaker compared to those found on the classical HOX PRE (Figure S8). Finally, the heterozygous Pc/+ mutation only temporarily prolongs gcm expression (Figure S5I), whereas it produces a long lasting HOX-dependent phenotype [53], [54].
Understanding the precise molecular events will require the development of new tools and the in vivo analysis of chromatin organization at the level of specific cell types or in single cells. Our data nevertheless clearly show that Gcm and Pc compete with each other: PcG proteins bind gcm genes as well as repo (Figure 2, Figure S7, Figure S8) [33] and counteract Gcm activity. We therefore speculate that Gcm displaces Pc from its target promoters, including itself, which would explain how a general chromatin regulator impinges onto a cell-specific transcriptional program. In mammals as well it has been suggested that cell fate transcription factors play a role in PcG recruitment and displacement and some of them were shown to be PcG targets ([55] and reference therein). Finally, 63 genes are common Pc and Gcm targets, as revealed by analyzing the Pc binding sites in embryos and in cell lines (from [33] and [34]) and the genes positively regulated by Gcm identified by microarray (from [23]). Clearly, genome-wide screens for direct Gcm targets will be necessary to support the hypothesis of Pc displacement. These studies will also assess whether the impact of the PRCs on glial proliferation is direct or mediated by Gcm.
The rescue of the gcm-dependent phenotype upon Pc downregulation indicates a role for this chromatin factor in glial repression. Interestingly, upregulating or downregulating Pc does not per se produce the opposite fate transformation (Figure S4D), whereas it does modify the number of glia, showing that distinct protein levels are required in different processes. In vertebrates, the PRC2 is also involved in the production of glial cells, which differentiate after a wave of neurogenesis. However, different results were obtained depending on the experimental asset. Livesey and collaborators ([56]) deleted Ezh2 constitutively, thereby altering the balance between self-renewal and differentiation, and found precocious astrocyte differentiation. In contrast, Gotoh and collaborators [57] used a conditional Ezh2 knockout and documented a decrease in astrocyte differentiation. In the first case, the authors speculated that the altered timing of neurogenesis and accelerated onset of gliogenesis are secondary to the primary function of PRC2 in cortical progenitor cells. In the second report, it was shown that Ezh2 represses Neurogenin1, which controls timing during corticogenesis and therefore the relative production of neurons and glia. While these studies indicate the importance of chromatin modifiers in the nervous system, they do not clarify the role of PRCs in gliogenesis. In our study, the combined use of sensitive tools demonstrates that the Pc chromatin factor directly inhibits gliogenesis and identifies gcm as a major target in the pathway. First, we used sensitized backgrounds rather than total knockouts, which makes it possible to score for subtle phenotypes. Second, we analyzed the mutants at the single cell resolution and therefore scored for direct, cell autonomous, effects of the Pc mutation. Third, we analyzed a gene that plays an instructive role rather than simply being permissive for gliogenesis. Fourth, gcm carries a functional PRE and competes with Pc on its targets. Altogether, these findings reinforce the view that distinct chromatin states characterize specific cell fates, as also illustrated by the low levels of histone acetylation observed in both fly and vertebrate glia [6], [58].
Flies were grown on standard cornmeal/molasses medium at 25°C. The deficiency kit was obtained from the Bloomington Stock Center (Bloomington, IN), see Supplementary Material and Methods.
For the qualitative screen: for each cross (180 deficiencies), double heterozygous females carrying the gcmPyx allele and a deficiency were scored for the supernumerary bristle phenotype and compared to sibling females carrying the gcmPyx allele and the balancer from the deficiency stock. This allowed us to classify each deficiency as gcmPyx modifier or not modifier (Figure S1Aa). 75 deficiencies covering 42 genomic regions were selected for quantitative analyses (Figure S1Ab); for each genotype we counted the bristles from 10–80 heminota. The flow chart in Figure S1A shows the details of the screen. Average values +/− SEM were calculated and, for genotype comparisons, the statistical significance was estimated by t-test.
To overexpress esg or osa, respectively, w; EP(2)0684/CyO or w; P{w[+mC] = UAS-osa}s2/CyO females were crossed with w; gcmPyx/Sp; hs-Gal4/Sb males. A 30 minute heat-shock pulse on 2nd instar larvae was performed at 37°C.
qChIP was performed as in [33]. Primers are listed in Figure S3A.
these assays were performed as in [41] and [44]. For the antibody list as well as for the protocol of wing and embryo mounting and analysis by confocal microscopy, see Text S1. Repo and β-Gal positive cells from embryonic VC were subjected to quantification in 3D image using Imaris 7.2 software. Masks were generated as a region of interest for three thoracic segments along the z-stack, then volume image was visualized and the “crop 3D” function was applied to isolate the region of interest. Voxels (volume picture element) corresponding to cells were identified based on size and intensity. Then automatic voxel (cell) counting was performed in the region of interest. t-test was used to quantify the difference between genotypes. For immuno-FISH staining on polytene chromosomes [59], three consequent probes covering around 3 kb around gcm TSS were used, see Figure S3. Unless specified, all quantitative analyses used the t-test.
The gcm2 promoter construct is pBLCAT6-1.96 from [35]. The 4,3 kb of the repo promoter [45] was cloned into the pRed H-Stinger vector (Berzsenyi and Giangrande, unpublished data). pPAC-gcm is described in [7]. UAS-gcm is described in [42]. pPAC-Pc and UAS-Pc were obtained by cloning the entire Pc cDNA in backbone vectors. pPAC-lacZ was a gift from T. Cook. Transient transfection of Drosophila S2 cells [60] was performed using Effectene (Qiagen) according to the manufacturer's instructions using 3 µg of total DNA. For CAT assay to evaluate the activation of the 2 kb gcm2 reporter construct (pBLCAT6-1.96), cells were harvested 48 hr after transfection and normalized for ß-Gal activity. CAT levels were determined using the CAT ELISA kit (Roche). For repoRFP, images of cells were acquired 48 hr after transfection, and the green (UAS-GFP)/red (repoRFP) cells, were quantified automatically using the ImageJ software.
|
10.1371/journal.ppat.1007284 | Host phosphatidic acid phosphatase lipin1 is rate limiting for functional hepatitis C virus replicase complex formation | Hepatitis C virus (HCV) infection constitutes a significant health burden worldwide, because it is a major etiologic agent of chronic liver disease, cirrhosis and hepatocellular carcinoma. HCV replication cycle is closely tied to lipid metabolism and infection by this virus causes profound changes in host lipid homeostasis. We focused our attention on a phosphatidate phosphate (PAP) enzyme family (the lipin family), which mediate the conversion of phosphatidate to diacylglycerol in the cytoplasm, playing a key role in triglyceride biosynthesis and in phospholipid homeostasis. Lipins may also translocate to the nucleus to act as transcriptional regulators of genes involved in lipid metabolism. The best-characterized member of this family is lipin1, which cooperates with lipin2 to maintain glycerophospholipid homeostasis in the liver. Lipin1-deficient cell lines were generated by RNAi to study the role of this protein in different steps of HCV replication cycle. Using surrogate models that recapitulate different aspects of HCV infection, we concluded that lipin1 is rate limiting for the generation of functional replicase complexes, in a step downstream primary translation that leads to early HCV RNA replication. Infection studies in lipin1-deficient cells overexpressing wild type or phosphatase-defective lipin1 proteins suggest that lipin1 phosphatase activity is required to support HCV infection. Finally, ultrastructural and biochemical analyses in replication-independent models suggest that lipin1 may facilitate the generation of the membranous compartment that contains functional HCV replicase complexes.
| Hepatitis C virus (HCV) infection is an important biomedical problem worldwide because it causes severe liver disease and cancer. Although immunological events are major players in HCV pathogenesis, interference with host cell metabolism contribute to HCV-associated pathologies. HCV utilizes resources of the cellular lipid metabolism to strongly modify subcellular compartments, using them as platforms for replication and infectious particle assembly. In particular, HCV induces the formation of a “membranous web” that hosts the viral machinery dedicated to the production of new copies of the viral genome. This lipid-rich structure provides an optimized platform for viral genome replication and hides new viral genomes from host´s antiviral surveillance. In this study, we have identified a cellular protein, lipin1, involved in the production of a subset of cellular lipids, as a rate-limiting factor for HCV infection. Our results indicate that the enzymatic activity of lipin1 is required to build the membranous compartment dedicated to viral genome replication. Lipin1 is probably contributing to the formation of the viral replication machinery by locally providing certain lipids required for an optimal membranous environment. Based on these results, interfering with lipin1 capacity to modify lipids may therefore constitute a potential strategy to limit HCV infection.
| Millions of humans are chronically infected by hepatitis C virus (HCV) worldwide [1]. Chronic HCV infection is a major biomedical problem as it causes liver inflammation and fibrosis, which can lead to severe liver disease, such as cirrhosis and hepatocellular carcinoma [2, 3]. There is no vaccine against HCV and, although blood-screening tests and other prophylactic measures have reduced the dissemination of this pathogen, a number of newly acquired infections still occur associated with risk behavior or with unknown origin [4, 5]. However, chronic HCV infection can be successfully eradicated from chronically infected individuals through specific direct-acting antiviral (DAA) combination therapies, virtually in all treated patients [6]. Since these specific treatments have only been in place recently, there are no sufficient clinical data on the long-term benefit of these treatments in relieving the severity of advanced liver disease [7, 8].
HCV is a Hepacivirus (Flaviviridae) with a positive sense, single-strand RNA genome that encodes a single open reading frame (ORF) flanked by untranslated regions (UTR), which are essential for viral polyprotein translation and viral genome replication. HCV ORF is co- and post-translationally processed by cellular and viral proteases to produce ten major proteins. These have been functionally classified in a replication module, that includes the minimal viral components of the RNA replicase (NS3, NS4A, NS4B, NS5A and NS5B) and an assembly module, which comprises the major structural components of enveloped HCV virions, the capsid protein (core) and the envelope glycoprotein complex formed by E1 and E2 heterodimers; as well as the polypeptides p7 and NS2, which are not structural components of virions but contribute to infectious particle assembly in a concerted action with the viral replicase [9].
HCV utilizes key aspects of cellular lipid metabolism for essentially every aspect of the virus replication cycle and strongly interferes with host cell lipid homeostasis [10–12]. In fact, chronic HCV patients display high rates of liver steatosis, severity of which inversely correlates with serum liver derived-lipoprotein [13]. Thus, although host immune response remains a major component in HCV pathogenesis, direct interference of HCV infection with hepatocyte lipid metabolism may contribute to overall disease progression [13].
One of the salient manifestations of HCV interference with lipid metabolism at the cellular level is the formation of a distinctive membranous web in HCV-replicating cells [14]. Within this structure, HCV RNA replication is thought to occur in double-membrane vesicles (DMVs) that emerge from the endoplasmic reticulum [15–17]. HCV replicase formation requires not only viral replicase components NS3 through NS5B [16], but also recruitment and subversion of different key cellular factors that cooperate to provide an optimal membrane microenvironment for the assembly of HCV replicase complexes [11]. In this sense, HCV replicase complexes are located in characteristic detergent-resistant membranes (DRMs) that co-fractionate with caveolin-2 [18–20] and the sigma-1 receptor (SIGMAR1) [21]. These endoplasmic reticulum (ER) microdomains containing the viral replicase are enriched in cholesterol, probably by the direct involvement of cellular proteins involved in non-vesicular cholesterol transport [11].
Assembly of progeny virions is thought to occur at the interphase between ER and lipid droplets (LD) [12] and HCV virion assembly and secretion processes are strongly dependent on host factors involved in biogenesis of triacylglycerol (TAG)-rich lipoproteins [22–25]. Thus, HCV infection requires de novo biosynthesis of phospholipids, like phosphatidylcholine (PC), in order to generate its membranous RNA replicase compartments [26] and TAG biosynthesis to ensure progeny virion production [12]. This, together with the fact that PC and TAG biosynthesis are altered by HCV infection [26–28], suggests an important role for glycerophospholipid metabolism in HCV infection.
Lipins are key players in glycerophospholipid metabolism, as they catalyze phosphate removal from phosphatidic acid (PA) to produce diacylglycerol (DAG), which is a precursor of TAG, but also a precursor for PC and phosphatidylethanolamine (PE) biosynthesis in mammalian cells [29]. Lipins may also translocate to the nucleus to directly interact with promoter DNA-binding transcription factors to stimulate or repress transcription of lipogenic genes [30–32]. Two conserved motifs are required for phosphatidic acid phosphatase (PAP) catalytic activity (DXDXT) and transcriptional regulation (LXXIL) respectively [31]. Substrate specificity is restricted to PA, a characteristic that, together with the lack of integral membrane domains and the dependence on Mg2+ for catalytic activity, differentiates lipins from other mammalian PAP [33].
Lipin1, the best-characterized member of the family, was discovered as a consequence of the genetic analysis of a mouse strain displaying fatty liver dystrophy phenotype (fld mice) [34, 35]. Interestingly, lipin1 deletion in the liver of fld mice, results in accumulation of lipin2 protein, a second member of the family predominantly expressed in the liver, which maintains normal cellular PAP activity and compensates some, but not all, aspects of glycerolipid metabolism [36]. Conversely, LPIN2 knock-out mice display increased lipin1 protein expression, as compared with wild type (wt) littermates, suggesting a similar homeostatic compensatory mechanism to maintain liver PAP activity [37]. Thus, lipin1 and lipin2 coordinate lipid homeostasis in the liver [37]. While individual LPIN1 or LPIN2 deletions are tolerated in mice, double knockout mice are embryonically lethal [37]. Despite this apparent functional redundancy, lipin1 is emerging as a key player in ethanol-induced steatosis [38] and a single nucleotide polymorphism in LPIN1 is associated with the severity of liver damage and fibrosis progression in pediatric human patients with histological non-alcoholic fatty liver disease (NAFLD) [39], suggesting that liver lipin1 dysfunction may contribute to steatosis-related liver pathogenesis.
In this study, we focused our attention on lipin1, as analysis of previously published transcriptomic profile datasets revealed that this gene is consistently regulated at the mRNA level during HCV infection. Silencing experiments indicate that HCV infection efficiency is strongly dependent on lipin1 expression. Using different cell culture models of infection, we identified a limiting role of lipin1 PAP activity in the generation of HCV replicase complexes. Defective replicase assembly leads to strong inhibition of HCV propagation in lipin1-deficient cells but not that of another (+) strand RNA virus, indicating a specific role for lipin1 in HCV infection.
Given the relevance of lipin1 for lipid metabolism and its steatogenic potential, we set out to independently verify data from published differential transcriptomic profile studies that suggested that HCV infection may alter LPIN1 mRNA abundance in cell culture [40, 41]. A specific and statistically significant LPIN1 mRNA induction (Fig 1A) was observed in Huh-7 cells after single cycle infection experiments [multiplicity of infection (MOI) 10] with a cell culture-adapted genotype 2a HCV (D183) variant [42] at the peak of the infection (48 and 72 hours post-infection) and as compared with mock-infected cells (Fig 1B). LPIN1 mRNA induction was prevented when infected cells were treated with 1μM sofosbuvir (Fig 1C), an HCV RNA polymerase inhibitor [43] that reduced viral RNA accumulation by more than two orders of magnitude (Fig 1D), indicating that active HCV replication is required to induce LPIN1 mRNA accumulation.
LPIN1 mRNA has been shown to be upregulated by mechanisms that involve induction of reactive oxygen species (ROS), as treatment of the cells with ROS-scavenger molecule N-acetylcysteine (NAC) is capable of preventing LPIN1 mRNA induction under glucose deprivation [44] or during H2O2 treatment [45]. Given that transcriptional activation of a subset of lipogenic genes during HCV infection is also prevented by addition of antioxidants [46], we sought to determine if LPIN1 mRNA induction by HCV infection is mediated by ROS production and therefore dampened by NAC treatment. LPIN1 mRNA induction was prevented in the presence of the antioxidant (Fig 1E), despite comparable HCV RNA accumulation in mock-treated and NAC-treated HCV-infected cells (Fig 1F), suggesting that virus replication-induced ROS production is required to induce LPIN1 mRNA accumulation. Western-blot analysis confirmed that the observed transcriptional change leads to a correlative protein accumulation (Fig 1G and 1H); reinforcing the notion that acute HCV infection alters lipin1 expression.
In order to determine if lipin1 subcellular localization was altered during HCV infection, we performed confocal microscopy studies in control and HCV-infected Huh-7 cells. Lipin1 staining was observed as cytoplasmic punctated structures both in control and HCV-infected cells. To study if lipin1 signal colocalized with viral antigens, we performed double staining with antibodies against lipin1 and double-stranded RNA (dsRNA) or replicase subunits NS3 and NS5A. None of the viral antigens strictly colocalized with lipin1 (Pearson´s<0.5) (S1A Fig). However, Mander´s coefficients indicate that the majority of lipin1 overlapped with a small fraction of NS3 and NS5A (S1C Fig). In contrast to lipin1, lipin2 signal did not overlap with that of viral proteins NS3 and NS5A (S1B Fig). Our results indicate that no major lipin1 rearrangements are observed after HCV infection and that only a minor fraction of NS proteins colocalize with lipin1.
To study if lipin1 plays any role in HCV infection, lipin1-deficient cells were generated by transducing human hepatoma (Huh-7) cells with lentiviral vectors expressing shRNAs targeting LPIN1 mRNA or a control vector expressing an irrelevant shRNA. Lipin1 expression silencing was verified by Western-Blot, typically 7 days post-transduction (Fig 2A). A partial (50%; shLPIN1-1) and a more profound (>95%; shLPIN1-2) reduction in lipin1 accumulation was observed after transduction with specific shRNAs as compared with the control (Fig 2B). As expected, lipin1 and lipin2 expression is inversely correlated in lipin1 shRNA-expressing cells (Fig 2B). The viability of lipin-deficient cells, as determined by MTT assay [47] was comparable to that of control cells (Fig 2B). These results illustrate that lipin1 shRNA-expressing cells respond homeostatically to functionally compensate partial (shLPIN1-1) and more pronounced (shLPIN1-2) loss of lipin1 (Fig 2B).
Control and lipin1-deficient cells were infected with a genotype 2a D183 at MOI 0.01 to study viral spread by determining extracellular infectivity titers at different times post infection. Fig 2C shows limited propagation of this virus in lipin1-deficient cells, with a statistically significant reduction of up to two (shLPIN1-1) and three orders of magnitude (shLPIN1-2) in viral titer between the control and lipin1-deficient cell lines at day 5 post-infection. These results indicate that lipin1 is required for efficient HCV propagation and that homeostatic lipin2 accumulation (Fig 2B) is not sufficient to support efficient HCV infection.
Since important differences in the interaction of different HCV genotypes with cellular lipid metabolism have previously been described [48], we set out to determine if the observations made with a JFH1-derived virus (genotype 2a) were extensive to other HCV genotypes. We performed low multiplicity infections (MOI 0.05) with genotype 1a TNcc virus strain in lipin1-deficient Huh-7.5 cell lines, which are susceptible to infection by this recombinant virus [49]. First, we verified that Huh-7.5.1 also display a significant reduction in genotype 2a infection efficiency when lipin1 is silenced (S2B Fig). In order to determine TNcc infection efficiency, intracellular HCV RNA accumulation was determined 72 hours post-inoculation in control and lipin1-deficient Huh-7.5 cells. Viral RNA detected under these conditions reflects the ability of genotype 1a to infect and replicate viral RNA in the different cell lines, as treatment of control cells with an HCV polymerase inhibitor 2´-C-methyladenosine (2mAde; 10 μM) [50] reduced viral RNA content by two orders of magnitude (Fig 2D). Lipin1 silencing consistently and significantly reduced TNcc replication by approximately 3-fold both in shLPIN1-1 and shLPIN1-2 cell lines (Fig 2D), indicating that lipin1 is also limiting for genotype 1a HCV infection.
To determine the specificity of these observations, identical lipin1-deficient and control Huh-7 cultures were inoculated at MOI 0.01 with a human alpha-coronavirus CoV-229E bearing a GFP reporter gene (hCoV-229E-GFP) [51]. Inoculation of control and lipin1-deficient cells with this virus resulted in comparable progeny virus production, as determined by infectivity titration in cell supernatants 48 hours post-infection (S3 Fig). These results suggest that lipin1 is not rate limiting for CoV-229E-GFP infection and that lipin1 expression is particularly limiting for HCV.
To determine which aspects of the HCV replication cycle are limited by lipin1 silencing, single cycle infection experiments were conducted by inoculating control and lipin1-deficient cell cultures at MOI 10 with genotype 2a D183 virus. Infection efficiency was measured by titration of progeny virus infectivity present in the supernatant of infected cells and intracellular HCV RNA accumulation at 48 and 72 hours post-infection. Infection of lipin1-deficient cells resulted in a significant reduction of progeny infectious virus production in shLPIN1-1 and shLPIN1-2 cells as compared with the titers observed in the supernatants of control cells (Fig 3A), reinforcing the notion that lipin1 silencing interferes with HCV infection. Reduced virus production is likely due to parallel reduction of intracellular HCV RNA levels observed in lipin1-deficient cells as compared with the control cell line (Fig 3B). This reduction was observed at all time points, except for that at 5 hours, indicating that the size of the inoculum and initial virus adsorption is comparable among the different cell lines (Fig 3B). Thus, lipin1 silencing suppresses HCV infection by interfering with a step of the HCV lifecycle preceding intracellular HCV RNA accumulation.
Next, we set out to determine if lipin1 silencing has any impact on persistent HCV infections to verify if lipin1 is also limiting for late aspects of the virus lifecycle. Persistently infected cells continuously replicate viral RNA, express viral antigens and secrete infectious virions. Thus, it is a valuable system to measure steady-state HCV RNA replication as well as infectious particle assembly and secretion. Persistently infected cultures were transduced with the lentiviral vectors described above to produce persistently infected, lipin1-deficient cells (S4 Fig). Analysis of extracellular infectivity titers revealed that infectivity titers in lipin1-deficient and control cells were comparable, with the exception of a marginal reduction in shLPIN1-2 expressing cells, indicating that lipin1 silencing does not strongly interfere with infectious virus production (Fig 3C). Intracellular HCV RNA levels in lipin1-deficient cells were comparable to that of the control cells (Fig 3D), indicating that lipin1 expression is not rate limiting for HCV RNA replication once infection has been established. Taken together, the results shown above indicate that lipin1 is only limiting at early steps of HCV infection leading to viral RNA accumulation.
As an independent verification of the hypothesis that lipin1 is limiting for early aspects of HCV infection, we used a single cycle surrogate infection model based on the production of HCV virions bearing a defective reporter genome encapsidated by trans-complementation (HCVtcp). HCVtcp are capable of producing abortive single-cycle infections, efficiency of which is proportional to the luciferase activity found in the target cells [52]. As expected from the results shown in Fig 3, infection of lipin1-deficient cells with HCVtcp resulted in a 75% reduction in the reporter luciferase activity in shLPIN1-1 cells and 90% in shLPIN1-2 determined at 48 hours post-infection, proportionally to the degree of silencing in these cells (Fig 4A and 4B). These results underscore the role that lipin1 plays at early steps of HCV infection and indicate that either viral entry or a step leading to efficient HCV RNA replication is impaired in these cells.
LPIN-1 mRNA is alternatively spliced in human liver to produce isoforms α and β, which may differ in catalytic activity, subcellular localization and gene expression regulation [53]. To determine the relative contribution of these isoforms to HCV infection, isoform-specific shRNAs were generated and used to transduce Huh-7 cells, as described above. LPIN1α-specific shRNA (shLPIN1-3) reduced total lipin1 protein by 70%, while LPIN1β-specific shRNA (shLPIN1-4) reduced total lipin1 expression by only 30% (Fig 4A). Infection of control and lipin1 isoform-deficient cells with HCVtcp revealed that lipin1α silencing resulted in a strong (85%) reduction in HCV infection while lipin1β silencing resulted in a milder (55%) but significant reduction in HCV infection efficiency (Fig 4B). These results indicate that both isoforms alpha and beta are limiting for HCV infection and suggest that the total amount of lipin1 present in the cell determines HCV infection efficiency. Taken together, the results obtained with four different shRNAs indicate that total lipin1 expression levels strongly correlate with HCVtcp infection efficiency (Fig 4C), underscoring a role for this host protein in early aspects of HCV infection.
Reduced HCV RNA accumulation in a single cycle infection (Fig 3) as well as reduced HCVtcp infection (Fig 4) may be due to a defect in entry of incoming virions. HCV E1E2-pseudotyped retroviral vectors bearing a luciferase gene (HCVpp) were used to measure viral entry because they constitute a sound model to study viral adsorption, receptor-mediated internalization and E1E2-mediated fusion in endosomes [54]. To assess the specificity of these observations, parallel cultures were inoculated with VSV-G pseudotyped retroviral vectors (VSVpp). Control and lipin1-deficient cell lines were infected with HCVpp (genotype 2a; JFH-1 strain) and VSVpp. As a positive control of inhibition of HCV entry, we used hydroxyzine (5 μM), which efficiently blocks HCV infection by interfering with viral entry [55]. As expected, hydroxyzine selectively inhibited HCVpp infection, as shown by reduced luciferase levels 48 hours post-inoculation only in HCVpp-infected cells (Fig 5A). Interestingly, lipin1-deficient cells (shLPIN1-1 and shLPIN1-2) were fully susceptible to HCVpp and VSVpp infection, as comparable luciferase activity levels were found in all cell lines 48 hours post-inoculation (Fig 5A). These results indicate that lipin1 is not rate limiting for receptor binding, particle internalization or E1E2-mediated endosomal fusion, which are steps recapitulated in this model [54]. Based on the data presented thus far, we hypothesized that lipin1 is limiting for a step in the HCV lifecycle downstream of HCV entry, leading to HCV RNA accumulation. To verify the hypothesis that lipin1 silencing causes strong reduction in initial HCV RNA accumulation by interfering with a step downstream of viral entry, we bypassed this step of the viral replication cycle by transfecting a subgenomic HCV RNA replicon bearing a reporter luciferase gene into control and lipin1-deficient cells. First, we evaluated HCV-IRES driven primary translation of incoming genomes by transfection of a replication-deficient mutant replicon that bears an inactivation mutation in the catalytic site of NS5B RNA polymerase [56]. Luciferase activity measured at 5 hours post-transfection was not reduced in any of the cell lines, indicating that transfection efficiency and HCV IRES-dependent primary translation was not significantly affected by lipin1 silencing (Fig 5B). A significant increase in Renilla luciferase activity was observed in shLPIN1-1 cells both when transfecting a replicon (Fig 5B) or a plasmid expressing Renilla luciferase under a minimal RNA polymerase II promoter (S5A Fig), suggesting that the increase in luciferase activity observed in these cells is not related with HCV. HCV RNA replication was evaluated by measuring accumulation of a reporter luciferase gene at 5 and 48 hours post-transfection of a replication competent HCV subgenomic replicon. Under these experimental conditions luciferase accumulation at 5 hours also represents HCV IRES-driven primary translation of the input RNA, while reporter luciferase activity at 48 hours depends on effective HCV RNA replication. In contrast to primary translation, HCV RNA replication inferred by luciferase activity at 48 hours was strongly reduced in lipin1-deficient cells (90% in shLPIN1-1 and 99% in shLPIN1-2 cells) (Fig 5C), indicating that initiation of HCV replication is dependent on normal lipin1 expression. This reduction was not due to a non-specific defect in luciferase expression, as co-transfection of a plasmid expressing Renilla luciferase lead to comparable luciferase accumulation in all cell lines, discarding the possibility that death of transfected cells or other spurious effects are responsible for the reduced luciferase activity accumulation (S5A Fig). Similar experiments were conducted in ATG4B-deficient cells, as this host factor was shown to be limiting for primary HCV translation [57]. Our studies confirmed that, while partial ATG4B silencing (S5B Fig) significantly interfered with primary translation (S5C Fig) and correlatively with HCV RNA replication efficiency, as expected [57], lipin1-silencing only affected HCV RNA replication (S5D Fig). Overall, these data indicate that HCV RNA replication is not initiated efficiently in lipin1-deficient cells and that blockade occurs at a step downstream translation of incoming HCV genomes.
In order to determine if any of the known functions ascribed to lipin1 is required for HCV infection, we tested the ability to restore HCV infection susceptibility of silencing-resistant wild-type (wt) or mutant lipin1 versions bearing a mutation in the catalytic site responsible for its phosphatase activity (DXDXT) and a mutant in the LXXIL motif, which is inactive both for transcriptional activation as well as for phosphatase activity [31]. Control and lipin1-deficient cells were transfected with wt lipin1 beta cDNA as well as with DXDXT and LXXIL mutants. Comparable overexpression levels of the wt and mutant proteins was obtained in each cell line, although overexpressed lipin1 levels were consistently higher in lipin1-deficient cells (Fig 6A). Cells were subsequently inoculated with HCV D183 at MOI 10 and relative infection efficiency was calculated by determining extracellular infectivity titers 48 hours post-infection. Overexpression of wt lipin1 did not significantly alter susceptibility to HCV infection in control cells, although we observed a small but consistent reduction in extracellular infectivity titers when overexpressing wt and mutant lipin1 constructs (S6A Fig). In contrast to control cells, wt lipin1 overexpression in lipin1-deficient cells consistently increased infectivity titers as compared to mock-transfected cells (S6B Fig) and in clear contrast with the reduction observed in control cells (S6A Fig) or when overexpressing mutant lipin1 in lipin1-deficient cells (S6B Fig). In order to take into account this divergent behavior in control and lipin1-deficient cells, we calculated the relative infection efficiency as the ratio of the infectivity titers found in lipin1-deficient cells and control cells (S6C Fig). Ratios were subsequently normalized to that found in mock-transfected cells in order to average the data from different experiments (S6 Fig). Using relative infection efficiency as readout of this set of experiments, we could clearly observe a statistically significant increase (2-fold) in the relative infection susceptibility in cells overexpressing wt lipin1 cDNA as compared with mock-transfected cells or cells expressing similar or higher levels of the mutants (Fig 6A), suggesting that wt lipin1 modestly, though significantly, rescues HCV infection while phosphatase (DXDXT) and transcriptional coactivation (LXXIL) mutants do not (Fig 6B). These results suggest that lipin1 transcriptional co-activation capacity is not sufficient to support HCV infection while lipin1 phosphatase activity is essential. However, given that LXXIL mutant is deficient both in transcriptional co-activation and PAP activity [31], we cannot determine if transcriptional co-activation by lipin1 is also required to support HCV infection.
The data described above suggest a role for lipin1 phosphatase activity in a step of HCV replication cycle between translation of the viral genome and formation of functional replicase complexes. Data regarding primary translation where inferred from a surrogate model of translation based on a reporter luciferase gene (Fig 5B). To address if indeed viral polyprotein is properly processed and inserted into detergent-resistant microdomains to form the characteristic membranous ultrastructures bearing the viral replicase, we used a replication-independent surrogate model of polyprotein expression. This system is based on a vector encoding the portion of the viral polyprotein corresponding to the replicase (NS3-NS5B) under the transcriptional control of the T7 polymerase and the translational control of encephalomyocarditis virus (EMCV) IRES (pTM-NS3/5B) [58]. Replication-independent polyprotein overexpression systems enable assessment of polyprotein processing as well as studying the formation of virus-derived membranous structures [16, 59].
Control and lipin1-deficient cells were infected with a recombinant vaccinia virus expressing T7 RNA polymerase (VacT7) and subsequently transfected with the plasmid pTM-NS3/5B to enable viral replicase expression. Sixteen hours post-transfection, cells were processed for Western-Blot using anti-NS3. Accumulation of NS3 is comparable in control and both lipin1-deficient cells, underscoring the notion that lipin1 is not limiting for polyprotein translation and processing (Fig 7A). Similarly, NS3 and NS5A expression and subcellular distribution was similar in all cell lines (Fig 7B). These results suggest that there are no major differences in accumulation of viral proteins in lipin1-deficient cells and that a step downstream is affected in these cells.
Transmission electron microscopy (TEM) of ultrathin cell sections of cells expressing HCV replicase components shows the expected accumulation of a mixture of characteristic double-membrane vesicles (DMV) as well as multiple membrane vesicles (MMV) (Fig 7D, 7E and 7F) that were not found in mock-transfected cells (Fig 7C), as reported in previous studies using similar systems [16]. Individual vesicle diameter displays heterogeneous size distribution in which the predominant population is distributed between 125–150 nm (median 150 nm; average ± SD: 167 ± 81 nm, n = 279) with larger vesicles being less predominant (Fig 7G and S7 Fig). This size distribution is compatible with that observed similar replication-deficient systems and during HCV infection. Treatment of these cells with 100 nM daclatasvir (DCTV), resulted in a strong reduction in the number of vesicles per section area (S7B Fig), without significantly altering the size distribution of the remaining vesicles (Fig 7G), as reported by Berger et al. [59]. Similarly, the diameter distribution of the vesicles found in lipin1-deficient cells was comparable to that in control cells (Fig 7G). However, HCV-induced structures were significantly less abundant 16 hours post-transfection in lipin1-deficient cells than in controls cells (S7C Fig). This reduced abundance is illustrated by a significant reduction in the fraction of cells displaying vesicular structures in lipin1-deficient cell cultures (Fig 7H) despite comparable transfection efficiency and viral protein expression levels, indicating that lipin1 may be required in a critical step leading to formation of the HCV-induced vesicular compartment.
To validate the TEM results independently, we set out to establish a biochemical assay to evaluate replication-independent replicase complex formation. One of the characteristics of the HCV replicase complexes is that they are located in detergent-resistant membranes (DRM) [19, 20]. In this sense, NS proteins are associated with replicase complexes that co-sediment with DRM markers such as caveolin-2 or sigma-1 receptor (SIGMAR1) in low-density fractions in isopycnic gradient ultracentrifugation experiments [19, 21]. Control and lipin1-deficient cells were infected with VacT7 and subsequently transfected with limiting doses of the plasmid pTM-NS3/5B [16]. Parallel samples were treated with DCTV (100 nM). Sixteen hours post-transfection, cell lysates were generated and subjected to equilibrium ultracentrifugation in 10–40% sucrose gradients. Gradient fractions were collected and subjected to Western-Blot analysis to determine the impact of lipin1 silencing on NS3, SIGMAR1 and actin sedimentation profiles. Fig 8A shows how, as previously shown for replicon and JFH1-infected cells, NS3 can be detected in DRM fractions (fractions 3–5), as determined by the presence of SIGMAR1 in those fractions (fractions 3–5; S8A Fig), although in this experimental system, unlike during viral infection [21], most NS3 co-sediments together with solubilized proteins, as shown for actin (fractions 9–12; S8A Fig). DRM-associated NS3 is reduced in DCTV-treated and lipin1-deficient cells, while total NS3 expression remains unchanged (Fig 8A). Four independent experiments were performed in which relative-DRM associated NS3, normalized to that found in solubilized fractions, was calculated for each experimental condition (Fig 8B). Lipin1-deficient cells display a consistent and statistically significant reduction in the DRM-associated, but not total NS3 abundance (Fig 8A and 8B; shLPIN1-1 and shLPIN1-2), similar to that observed in control cells in the presence of DCTV (Figs 8A and 8B; shControl+DCTV), consistent with the TEM data (Figs 7 and S7) and the notion that NS3 in DRM fractions may reflect the abundance of replicase complexes formed in these cells. This reduction is not due to an overall reduction in cellular DRM abundance in lipin1-deficient cells, as lipin1-deficient cells display similar SIGMAR1 distribution pattern as the control cells (S8 Fig). Overall, TEM data and DRM floatation assays strongly suggest that lipin1 is rate limiting for the generation of replicase complexes from fully processed polyprotein subunits.
Hepatitis C virus replication cycle is tightly linked to host cell lipid metabolism and interference with cellular lipid homeostasis contributes to viral pathogenesis [3]. One of the most evident consequences of this interference is the high prevalence of liver steatosis among chronically infected patients [13, 60]. This clinical manifestation of the infection has been linked to, among others, chronic ER stress, mitochondrial dysfunction and metabolite depletion induced by HCV infection, which result in the activation of persistent homeostatic adaptation of the cellular lipid metabolism to permit cell survival, at the cost of pathogenic metabolic alterations [11, 61–64].
Among the different regulatory networks that have been shown to be stimulated during HCV infection, PPARα [61], PGC-1α [65], HIF-1 [66] and SREBP [46, 67] have also been shown to regulate transcription of LPIN1 mRNA [31, 45, 68, 69]. Thus, it is likely that stimulation of one or several of these regulatory networks by HCV infection results in the LPIN1 mRNA transcriptional activation observed in this (Fig 1A) and other studies [40, 41]. Importantly, prevention of LPIN1 mRNA accumulation with NAC (Fig 1E) did not significantly interfere with HCV RNA replication (Fig 1F), suggesting that enhanced LPIN1 mRNA accumulation is not required for efficient HCV infection. We favor the hypothesis that ROS induced by HCV protein accumulation actively participates in LPIN1 induction, as treatment with the antioxidant NAC prevented LPIN1 mRNA accumulation, similar to what has been shown for other SREBP-regulated genes during HCV infection [46]. Accumulation of LPIN1 mRNA during HCV infection results in concomitant protein accumulation (Fig 1G and 1H). However, post-translational mechanisms such as phosphorylation, acetylation or sumoylation regulate lipin1 protein stability, membrane association as well as subcellular localization thus influencing the activity of lipin1 as PA-phosphatase and as transcriptional coactivator [70]. Hence, it is difficult to predict the implications of lipin1 protein accumulation during HCV infection. Thus, future studies on the interference of HCV infection with cellular lipin1 functions will be required to determine its role in HCV-related pathogenesis, particularly in its contribution to steatosis.
The data presented in this study provide evidence that lipin1 is rate limiting for HCV infection at an early step of the infection leading to formation of membranous HCV replicase complexes, downstream of viral polyprotein expression and processing. We provide evidence for reduced accumulation of viral RNA during single cycle infection experiments (Fig 2D), that is reminiscent of a faulty initiation of viral replication, as suggested by reduced replication of a transfected subgenomic replicon in lipin1-deficient cells (Fig 5D). Data obtained in replication-independent polyprotein expression models suggest that generation of the membranous compartment that contains functional replicase complexes is severely limited in lipin1-deficient cells, as suggested by a significant reduction of the fraction of cells where these structures could be visualized by TEM (Figs 7 and S7). This hypothesis is further supported by a significant reduction in DRM-associated NS proteins in lipin1-deficient cells (Fig 8), which may reflect limitations in the association of viral replicase subunits with cholesterol and sphingolipid-rich membranes in lipin1-deficient cells [19, 20, 71].
Four different shRNAs targeting LPIN1 mRNA decrease susceptibility to HCV infection proportionally to their ability to reduce total lipin1 protein accumulation (Fig 4). These results, together with the cDNA rescue experiments (Fig 6), strongly reduce the possibility of observing RNAi-associated off-target phenomena. Interestingly, homeostatic accumulation of lipin2 protein in lipin1-deficient cells (Fig 2A and 2B) is not sufficient to compensate for lipin1 loss to support efficient HCV infection. The notion that, despite being capable of mutually compensating basic liver functions [36, 37], lipin1 and lipin2 play non-redundant functions in the liver has previously been proposed [36, 72, 73]. Lipin1 is tightly regulated at many different levels and its activity accommodates PAP activity in response to different physiological situations such as fasting and insulin signaling [70]. Compelling evidence indicates that, while lipin1 and lipin2 cooperate to maintain liver lipid homeostasis, the two proteins differ in many aspects. For instance, lipin1 is transcriptionally induced by PGC-1α and it is also an inducible amplifier of this transcriptional network [31], whereas lipin2 is not [36]. Lipin1 is sumoylated and sumoylation regulates its nuclear localization and function, whereas lipin2 sumoylation could not be demonstrated, despite the presence of a canonical sumoylation motif in its primary sequence [74]. Lipin1 enzymatic activity is blocked by mammalian target of rapamycin (mTOR)-dependent phosphorylation in response to different metabolic stimuli [30, 75], whereas lipin2 is constitutively active even when phosphorylated [76]. Thus, lipin2 is considered more as a constitutive phosphatidic acid phosphatase with lower specific activity than lipin1 [76]. In addition to these differential regulatory networks, it has been shown that in vitro PAP activity of purified lipin1 and lipin2 is differentially influenced by the composition of the substrates (liposomes and lipid-detergent micelles) as well as the pH at which the assay is performed [76]. This differential lipid substrate recognition may be reminiscent of the different preferential association with membranes of different subcellular compartments (S1 Fig)[73, 76]. These differences suggest that, while lipin1 and lipin2 may share some common features, they are not functionally interchangeable, particularly not in the case of HCV infection [70].
Our data support the notion that lipin1 silencing has a strong impact on HCV infection without affecting basic cellular functions (Fig 2B) or significantly interfering with infection by an unrelated virus (S3 Fig). Despite great efforts and different overexpression systems, functional rescue of lipin1 functions by wt lipin1 cDNA overexpression in lipin1-deficient cells only lead to a small but consistent rescue of virus infection efficiency, which was only observed when overexpressing wt lipin1 (Figs 6 and S6). Given the multiple transcriptional, post-transcriptional and post-translational regulation levels existing for lipin1 expression, it is conceivable that only a fraction of the overexpressed lipin1 is fully competent to sustain HCV infection. Moreover, high overexpression levels could only be achieved in lipin1-deficient cells (Fig 6A), underscoring the notion that intracellular lipin1 levels are tightly regulated by the host. Nevertheless, overexpression of a mutant lipin1 lacking phosphatase activity (DXDXT) or a mutant inactive as transcriptional coactivator (LXXIL) were not capable of enhancing HCV infection in lipin1-deficient cells as compared with the wt lipin1 (Fig 6B). These data reveal that lipin1 phosphatase activity is required for lipin1 to support HCV infection and suggest that, while transcriptional co-activation by lipin1 may be important, this function is not sufficient to support HCV replication.
Lipins are important enzymes in the main pathway for de novo phospholipid biosynthesis by providing DAG derived from the glycerol-3-phosphate pathway to produce PC and PE through the Kennedy pathway [70]. Production of membranous replicase compartments likely requires de novo synthesis of PC and/or PE, which are major components of biological membranes that depend on DAG biosynthesis [29]. In fact, local PC biosynthesis is required for efficient replication of (+) RNA viruses and certain PC species accumulate in HCV-infected cells [26, 28]. Although increased lipin2 accumulation may be sufficient to compensate lipin1 silencing at the whole-cell level [37], it is possible that acute, local demand of de novo synthesized phospholipids is required at defined suborganellar compartments during early steps of HCV infection and that lipin1-deficiency shortens or alters the availability of different membrane components, demand that may not be satisfied by lipin2, given the differential regulation[76] and subcellular localization of these two proteins (S2 Fig).
Remarkably, deletion of the yeast lipin homologs pah1/smp2 (S. cerevisae) or ned1 (S. pombe) gene, results in deregulated proliferation of the ER and nuclear envelope membranes [77, 78], with concomitant enhancement in (+) RNA virus replication [79, 80]. The membranous alterations and elevation of total phospholipid content observed in pah1-deficient yeast have been ascribed to transcriptional activation of pah1-independent alternative phospholipid biosynthetic programs due to PA accumulation [77, 80]. Our data in mammalian cells are more compatible with a shortage of phospholipid production, which may be at the basis of the reduced abundance viral membranous structure (Figs 7 and 8). This opposite outcomes of infection may derive from the fact that yeast use mainly PA (lipin1 PAP substrate) as a precursor for PC biosynthesis, while mammals mainly use DAG (lipin1 PAP product) as precursor for PC biosynthesis through the Kennedy pathway [77, 80–82]. Moreover, in contrast to yeast and lower eukaryotes, which express only one lipin gene, three different lipin genes coordinate glycerolipid homeostasis in mammals [72]. Thus, interfering with expression of one of the members of the family may not be sufficient to observe the same effects observed when deleting pah1, as transcriptional and posttranslational homeostatic compensations are in place in mammals, particularly between lipin1 and lipin2 in liver tissue [36, 37]. In this sense, deletion of either lipin1 or lipin2 in mouse models results in a relatively balanced liver phospholipid content while simultaneous deletion of both lipins is embryonically lethal [37]. Accordingly, only minor alterations of ER membranes [83] and no significant alterations in total PC levels [36] have been reported in lipin1-deficient mouse liver. Given the fact that hCoV-229E is fully capable of replicating in these cells (S3 Fig), it is unlikely that a general disruption of de novo phospholipid biosynthesis occurs in lipin1-deficient cells, particularly since hCoV-229E infection also induces profound ER membrane rearrangements required for replication [84], some of which are structurally similar to DMVs observed during HCV infection [85]. Thus, we favor the hypothesis that a subcellular pool of glycerophospholipids is managed by lipin1 in Huh-7 cells and that lipin1 silencing perturbs local levels of PA and DAG, limiting local availability of precursors of structural components of virus-induced membranes.
Alternatively, lipin1 deficiency may alter local amounts of important signaling molecules, in particular, that of its substrate (PA) or its product (DAG). Deregulation of the local PA and DAG pools may cause important alterations for the host cell, as both metabolites are potent chemical messengers that regulate different aspects of cellular homeostasis [86–88]. Regarding PA conversion into DAG by lipins, it has been shown that pah1 (yeast lipin1 homolog) phosphatase activity is critical for transforming local pools of PA into DAG at the ER membrane to facilitate membrane fusion events mediated by SNARE complexes [89, 90]. Mammalian lipin1 phosphatase activity is also critical for transforming local pools of PA that accumulate at the surface of mitochondria to promote mitochondrial fission [91] or at the surface of endolysosomes to facilitate autophagy [92]. Thus, lipin1 and probably other members of the lipin family modulate different aspects of intracellular membrane signaling. Given that the function of host factors known to be involved in functional HCV replicase biogenesis, like VAPA, VAPB and OSBP [11] are indirectly regulated by local PA/DAG pools [93], it is tempting to propose that lipin1 silencing interferes with the function of one or several of these, or other yet uncharacterized cellular factors. In contrast to what has been reported for other host factors required for HCV replicase complex formation [58], we did not find evidence of lipin1 protein relocalization during HCV infection (S1 Fig). Thus, determining the precise mechanism by which lipin1 regulates HCV replicase formation is challenging, as association of lipin1 with different cell membranes is transient and highly regulated by posttranslational modifications [70]. Moreover, some of the known lipin1 cellular functions may be compensated by other lipins, particularly lipin2 in the liver. Nevertheless, our data clearly indicate that lipin1 participates at early stages of HCV replication and that the aforementioned homeostatic compensations by other lipins in regards to cellular metabolism may constitute an advantage when considering lipin1 as a host target for anti-HCV therapy.
HCV antiviral compounds 2´-c-methyladenosine (2mAde), sofosbuvir and daclatasvir were obtained from Boc Sciences (NY, USA), Selleckchem (Texas, USA) and Medchem Express (New Jersey, USA) respectively and dissolved in DMSO to obtain 10mM stock solutions. N-acetylcysteine (NAC) and puromycin were obtained from Sigma-Aldrich (Missouri, USA), dissolved in water to a final concentration of 0.5 M and 50 mg/ml respectively. Hydroxyzine pamoate was purchased from Sigma-Aldrich (Missouri, USA) and dissolved in DMSO to a final 10 mM concentration.
Human hepatoma Huh-7 and derived subclones Huh-7.5, Huh-7.5.1 (clone 2) have been described [94–96] and were kindly provided by Dr. Chisari (TSRI-La Jolla, CA). HEK-293T cells [97] were kindly provided by Dr. Ortin (CNB-Madrid, Spain). Cell cultures were maintained subconfluent in Dulbecco´s Modified Eagle´s Medium (Gibco) supplemented with 10 mM HEPES (Gibco), 100U/ml Penicillin/Streptomycin (Gibco), 100μM non-essential amino acids (Gibco) and 10% fetal bovine serum (Sigma-Aldrich).
Lentiviral vectors expressing control and LPIN1-specific shRNAs were used to inoculate Huh-7 cells. Twenty-four hours later, cells were subjected to selection with 2.5μg/ml of puromycin to assess the lowest lentivirus dose capable of conferring puromycin resistance to 100% of the cell population. Selected cell populations were subsequently cultured in the presence of puromycin until LPIN1 silencing was ascertained by Western-Blot using anti-lipin1 antibodies, typically at day 6–7 post lentiviral transduction, time at which all experiments were performed in the absence of puromycin. Before execution of all the experiments shown in this study, lipin1 expression was assessed by Western-Blot. Cell viability was determined by a thiazolyl blue tetrazolium blue (MTT) formazan formation assay [47].
Control and lipin1-deficient cell lines (5. 104 cells/well) were plated onto 12-well plates and were inoculated with D183 virus at a MOI 10 FFU/cell. Samples of the cells and supernatants were collected 24, 48 and 72 hours post-infection. For multiple cycle infection experiments (MOI 0.01), samples of the supernatants were collected at day 3, 5 and 7 post-inoculation. Cells were split 1:3 in the multiple cycle infection experiments at days 3 and 5 to maintain the cultures subconfluent. Extracellular infectivity titers were determined by endpoint dilution and infection foci counting as previously described [99]. Intracellular HCV RNA was determined by reverse transcription and quantitative PCR (RT-qPCR) as previously described [99].
Total protein samples were prepared in Laemmli buffer and separated using polyacrylamide denaturing gel electrophoresis (SDS-PAGE). Proteins were subsequently transferred onto PVDF membranes and incubated with 5% milk (lipins) or 3% BSA in PBS-0.25% Tween20 for one hour at room temperature (RT). Primary antibodies against lipin1 (clone B-12; Santa Cruz), lipin2 (H-160; Santa Cruz), NS3 (clone 2E3; Biofront), beta-actin (ab8226; Abcam) and tubulin (clone AA2; Sigma-Aldrich) were diluted in PBS-0.25% Tween20 and incubated for 1 hour (four hours for lipins) at RT. Membranes were subsequently washed for 20 minutes with PBS-0.25% Tween20 three times. Horseradish peroxidase-conjugated secondary antibodies were incubated for 1 hour at room temperature in 5% milk-PBS-0.25% Tween20 and subsequently washed three times for 20 minutes at room temperature. Protein bands were detected using enhanced chemoluminescence reactions and exposure to photographic films. Specific bands were quantitated using the ImageJ Software [100] on non-saturated, scanned films.
Huh-7 cells were grown on glass coverslips and infected at high multiplicity (MOI 10) with D183 virus. Forty-eight hours post infection cells were fixed for 20 minutes at RT with a 4% formaldehyde solution in PBS, washed twice with PBS and incubated with an incubation buffer (3% BSA; 0.3% Triton X100 in PBS) for 1 hour. Antibodies were diluted in incubation buffer: rabbit anti-lipin1 antibody (1:50; Cell Signaling-Leiden, The Netherlands), rabbit anti-lipin2 antibody (1:100; H-160; Santa Cruz), mouse anti-dsRNA (1:200; J2 clone; Scicons), anti-NS3 (1:500; 2E3 clone; Biofront) or anti-NS5A(1:200; 7E2 clone; Biofront). Primary antibodies were incubated with the cells for 1 hour (4 hours for lipin2 experiments) time after which the cells were washed with PBS and subsequently incubated with a 1:500 dilution of a goat anti-mouse conjugated to Alexa 488 or Alexa 594 (Invitrogen-Carlsbad, CA). Nuclei were stained with DAPI (Life Technologies) during the secondary antibody incubation using the manufacturer´s recommendations. Cells were washed with PBS and mounted on glass slides with Prolong (Invitrogen-Carlsbad, CA).
Confocal microscopy was performed with a Leica TCS SP8 laser scanning system (Leica Microsystems). Images of 1024 × 1024 pixels at eight bit gray scale depth were acquired sequentially every 0.13–0.3 μm through a 63x/1.40 N.A. immersion oil lens, employing LAS AF v 2.6.0 software (Leica Microsystems). Colocalization indexes were calculated using Jacop plugin for Image J [101] from a minimum of 10 regions of interest (ROI). Images were processed using ImageJ, were medians of 1 pixel were obtained for the different channels, only for illustration, not for analysis. Color levels, brightness and contrast were manipulated for illustration using technical and biological controls as reference.
Total RNA extraction was performed using the GTC extraction method [102]. Purified RNA (10–500 ng) was subjected to RT-qPCR using random hexamers and a Reverse Transcription Kit (Applied Biosystems). Quantitative PCR was performed using 2X Reaction Buffer from (Applied Biosystems) and specific oligonucleotides as previously described [99, 103]. Standard curves were prepared by serial dilution of a known copy number of the corresponding amplicon cloned in a plasmid vector.
Control and lipin1-deficient Huh-7.5 cells were inoculated with TNcc virus (MOI 0.05). Due to the relatively low propagation levels of the TNcc virus in this experimental setup, parallel cultures were infected and treated with 2´-C-methyladenosine (2mAde; 10μM) to determine the levels of non-replicative, background HCV RNA. Cells were incubated for 72 hours at 37°C, time after which samples of the cells were collected to determine intracellular HCV RNA levels by RT-qPCR.
To establish persistently infected cell cultures Huh-7 cells were inoculated at MOI 0.01 with JFH-1 HCV strain as previously described [24]. Cell cultures were maintained subconfluent for two weeks, time after which infection rates reach nearly 100% of the cells, as assessed by immunofluorescence microscopy. At this point cells were split and transduced with the corresponding lentiviral vectors in order to generate lipin1-deficient cell cultures as well as control cell lines. Once silencing had been verified by Western-blot, typically at day 6–8 post-transduction, cells were split and samples of the cells and supernatants were collected 24 hours later to determine intracellular HCV RNA levels by RT-qPCR and extracellular infectivity titers by end-point dilution and immunofluorescence microscopy.
Infectious, spread-deficient HCV particles produced by trans-complementation (HCVtcp) have previously been described [52]. Briefly, Huh-7.5.1 clone 2 cells expressing core-E1 and E2-NS2 regions from JFH-1 by lentiviral transduction, were electroporated with a JFH-1 subgenomic dicistronic replicon bearing a firefly luciferase gene with reagents kindly provided by Dr. Ralf Bartenschlager (U. of Heidelberg). Supernatants containing HCVtcp were collected 36, 48 and 72 hours post-electroporation, pooled and assayed for viral infectivity. HCVtcp infection efficiency was determined by inoculating naïve Huh-7 cells with the electroporation supernatants and measuring luciferase activity 48 hours post-infection using a commercially available kit (Promega).
Retroviral particle production pseudotyped with different viral envelopes has previously been described [54, 55] with the materials kindly provided by Dr. F. L. Cosset (INSERM, Lyon). Control and lipin-deficient cell lines were inoculated with HCVpp and VSVpp and incubated for 48 hours, time at which total cell lysates were assayed for luciferase activity using a commercially available kit (Promega). A selective HCV entry inhibitor, hydroxyzine pamoate (HDX) from Sigma-Aldrich (Missouri, USA), was used as positive control of inhibition [55].
A plasmid containing the sequence corresponding to a subgenomic JFH-1 replicon bearing a firefly luciferase reporter gene was kindly provided by Dr. Ralf Bartenschlager (U. of Heidelberg) [52]. After digestion with the restriction enzyme MluI, the linearized plasmid was transcribed in vitro using a commercial kit (Megascript T7; Ambion-Paisley, UK). The resulting products were digested with DNAse and precipitated with LiCl. Pelleted RNA was washed with 75% and 100% ethanol, and resuspended in nuclease-free water. In vitro transcribed RNA was transfected into the different cell lines together with a plasmid expressing Renilla luciferase under a minimal promoter (pRL-null; Clontech-California, USA) using Lipofectamine 2000 and the manufacturer´s recommendations (Life Technologies- California, USA). Firefly and Renilla luciferase activities were measured in the sample using a commercial kit (Dual Luciferase Assay System; Promega-Wisconsin, USA) at different times post-transfection.
Lipin1-deficient cells were generated by lentiviral transduction of shLPIN1-2 shRNA. At day 3 post-transduction, control and lipin1–deficient cell populations (5 X 104 cells/M24 well) were transfected in suspension using lipofectamine 2000 with plasmids (800 ng/M12 well) expressing wt lipin1beta isoform shLPIN1_2-resistant cDNA or DXDXT or LXXIL motif mutants [31]. Transfected cell cultures were incubated for 48 hours and subsequently inoculated at MOI 10 with D183 virus. Infection efficiency was determined by measuring extracellular infectivity titers 48 hours post-infection. Parallel cultures were used to determine relative wt and mutant protein expression efficiency by Western-Blot. Infectivity titers were measured as described above. The relative impact of cDNA expression was estimated by determining the ratio between the infectivity found in lipin1-deficient cells and the control cells transfected with the same plasmid. In order to average experiments with different raw infection efficiency, all the experiments were referenced to the ratio in the mock-transfected cells.
Control and lipin1-deficient Huh-7 cells were inoculated with CoV-229E (MOI 0.01) for 2 hours at 37°C. Cells were washed twice with warm PBS and replenished with DMEM-10%FCS. Extracellular infectivity titers were determined 48 hours post-infection by end-point dilution and fluorescence microscopy in Huh-7 cells.
Huh-7 cells were inoculated at MOI 10 with a recombinant vaccinia virus expressing the T7 phage RNA polymerase (VacT7) [104]. Two hours later, cells were transfected with the plasmid pTM-NS3/5B [16, 58] (kindly provided by Dr. Lohmann; U. of Heidelberg) and Lipofectamine 2000 (ThermoFisher-Massachussets, USA) following the manufacturer´s recommendations in terms of total DNA per well (typically 4μg per 35mm dishes with 7.5 X 105 cells/well) and 50% of the recommended lipofectamine:DNA ratio. Transfected cells were cultured in the presence of the DNA replication inhibitor cytosine β-D-arabinofuranoside (AraC; Sigma-Aldrich) for 16 hours to prevent VacT7 replication [105]. When indicated, media was also supplemented with 100nM daclatasvir (DCTV). Total cell extracts were used to determine viral protein accumulation by Western-blot using anti-NS3 antibody (clone 2E3; Biofront) and β-actin (Abcam; ab8226) as loading control.
For ultrastructural electron microscopy studies, control and lipin1-deficient cells expressing NS3-5B polyprotein (see above) were cultured on glass coverslips and fixed in situ after polyprotein expression with a mixture of 2% paraformaldehyde (TAAB) and 2.5% glutaraldehyde (TAAB) (1h at room temperature), post-fixed with 1% osmium tetroxide in PBS (45 min), treated with 1% aqueous uranyl acetate (45 min), dehydrated with increasing quantities of ethanol and embedded in epoxy resin 812 (TAAB). Ultrathin, 70-nm-thick sections were cut in parallel to the monolayer, transferred to formvar-coated EM buttonhole grids and stained with aqueous uranyl acetate (10 min) and lead citrate (3 min). Sections were visualized on a Jeol JEM 1200 EXII electron microscope (operating at 100 kV).
Quantitation of HCV-induced structures was performed as follows. To quantitate the differences in total vesicle abundance, TEM sections were visually inspected under the microscope for the presence/absence of vesicular structures. The number of positive cells and total number of cells were inserted in a 2 X 2 contingency table to determine the statistical significance of the differences between control and lipin1-deficient cells using two-tailed Fisher´s Exact Test or two-tailed Chi Square Test. In addition, we determined the frequency of HCV-induced vesicles in DCTV-treated control cells by dividing the number of structures per inspected area and calculating the average and SD of the frequencies found in the different images. The diameters of individual vesicles were determined manually using size-calibrated images and Image J software.
Lipin1-deficient and control cells (7.5 X 105 cells) were infected with VacT7 virus (MOI 10) and transfected with limiting doses of pTM-NS3/5B plasmid (typically 800 ng/ well), as higher plasmid doses may difficult observing the reported differences. Cells were lysed by adding 250 μl of TNE (50mM Tris-HCl pH 7.5, NaCl 150 mM and EDTA 2 mM) buffer containing 0.5% Triton X-114 and protease inhibitors (Complete; Roche- Basel, SW). Lysates were incubated for 30 minutes on ice before clearing them by 10-minute centrifugation at 12,000 r.p.m. Clear supernatants were mixed 1:1 with 60% sucrose TNE solution. This mixture was applied on top of a 40% sucrose-TNE cushion and was overlaid with 20% and 10% sucrose-TNE until completing the discontinuous gradient. Gradients were centrifuged for 16 hours at 120,000 X g. Fourteen fractions were collected from the top and analyzed by SDS-PAGE and Western-Blot using antibodies against NS3 (clone 2E3; Biofront), SIGMAR1 (S-18; Santa-Cruz), caveolin-2 (Epitomics; 3643–1) and beta actin as described previously [21]. NS3 signal was quantitated using ImageJ software and the fraction of DRM-associated NS3 was determined as the ratio of NS3 signal in fractions 3, 4 and 5 to the total NS3 signal in the gradient.
|
10.1371/journal.pbio.0050109 | Selective Translational Repression of Truncated Proteins from Frameshift Mutation-Derived mRNAs in Tumors | Frameshift and nonsense mutations are common in tumors with microsatellite instability, and mRNAs from these mutated genes have premature termination codons (PTCs). Abnormal mRNAs containing PTCs are normally degraded by the nonsense-mediated mRNA decay (NMD) system. However, PTCs located within 50–55 nucleotides of the last exon–exon junction are not recognized by NMD (NMD-irrelevant), and some PTC-containing mRNAs can escape from the NMD system (NMD-escape). We investigated protein expression from NMD-irrelevant and NMD-escape PTC-containing mRNAs by Western blotting and transfection assays. We demonstrated that transfection of NMD-irrelevant PTC-containing genomic DNA of MARCKS generates truncated protein. In contrast, NMD-escape PTC-containing versions of hMSH3 and TGFBR2 generate normal levels of mRNA, but do not generate detectable levels of protein. Transfection of NMD-escape mutant TGFBR2 genomic DNA failed to generate expression of truncated proteins, whereas transfection of wild-type TGFBR2 genomic DNA or mutant PTC-containing TGFBR2 cDNA generated expression of wild-type protein and truncated protein, respectively. Our findings suggest a novel mechanism of gene expression regulation for PTC-containing mRNAs in which the deleterious transcripts are regulated either by NMD or translational repression.
| A class of mutations found in many cancers introduces aberrant termination signals during the synthesis of mRNA. In mammalian cells, abnormal mRNAs containing premature termination codons (PTCs) are normally degraded by a process called nonsense-mediated mRNA decay (NMD), thus avoiding potentially deleterious effects from abnormal protein production. However, some PTC-containing mRNAs are known to escape from NMD. By screening protein expression from genes with serious mutations in colon cancers, we confirmed that PTC-containing mRNAs of some genes escape from NMD. However, their abnormal proteins were not found in the tumor cells. To study the means by which these proteins were regulated, we transfected separate cell lines with NMD-escape mutant genomic DNA, wild-type genomic DNA, and mutant cDNA. We found that truncated proteins are not generated from the NMD-escape mutant genomic DNA, whereas wild-type protein and truncated protein were generated normally. These results indicate that the translation of PTC-containing mutant mRNAs is repressed in the cytoplasm.
| A subset of colorectal carcinomas exhibit a molecular phenotype commonly referred to as high microsatellite instability (MSI-H) [1]. The microsatellite instability (MSI) pathway begins with the inactivation of one of a group of genes responsible for DNA nucleotide mismatch repair (MMR), which leads to extensive mutations in both repetitive and non-repetitive DNA sequences [2–4]. The mechanism of tumorigenesis in MSI-H tumors is thought to involve frameshift mutations of microsatellite repeats within the coding regions of affected genes, and the inactivation of these genes is believed to contribute directly to tumor development and progression [5,6]. The frameshift mutations observed in the affected genes are expected to generate previously undescribed amino acid sequences in the C-terminal part of the respective proteins (Figure S1). If abnormal mRNAs and proteins are generated from the frameshift-mutated genes, tumor-specific antigen may be generated. High peritumoral lymphocytic infiltration and a relatively good prognosis have been reported in MSI-H tumors [7,8].
One of the important consequences of frameshift mutations is the formation of premature termination codons (PTCs). In mammalian cells, mRNAs containing a PTC due to a nonsense mutation or a frameshift mutation are recognized and degraded by nonsense-mediated mRNA decay (NMD), thus eliminating the production of the potentially deleterious truncated proteins [9,10]. NMD of mRNAs carrying PTCs is mediated through the recognition of the PTC by its position relative to the 3′-most last exon–exon junction. As a general rule, mammalian transcripts that contain a PTC more than 50–55 nucleotides (nt) upstream of the last exon–exon junction will be subjected to NMD [11,12].
Although PTC formation in frameshift mutation-derived mRNAs and their subsequent degradation through NMD is widely accepted, PTCs located within 50–55 nt or downstream of the last exon–exon junction are not recognized by NMD (NMD-irrelevant), and some mRNAs with PTCs more than 50–55 nt upstream of their last exon–exon junction are not degraded by NMD (NMD-escape) [13,14]. In MSI-H tumors, several NMD-sensitive or NMD-escape PTC-containing mRNAs have been reported. A previous study compared the total mRNAs of affected genes from various cell lines [15]. However, this study did not differentiate the proportion of wild-type and mutant mRNAs, and did not confirm the mutant mRNAs through sequencing. This study also did not consider that the amount of mRNA from the affected genes might vary between cell lines. Moreover, the expression statuses and biological effects of the NMD-escape PTC-containing mRNAs are essentially unknown.
In order to clarify the protein expression status of affected genes with frameshift mutations and the role of NMD in these mutated genes, we selected MSI-H cancers as a model system because these cancers have accumulated genes with frameshift mutations, and the mRNAs expected from these mutated genes contain PTCs. We analyzed the expression of 20 mutant mRNAs from 12 genes and evaluated their regulation along with the regulation of associated proteins. We demonstrate that some PTC-containing mRNAs escaped from NMD, but did not generate truncated proteins, indicating that PTC-containing transcripts can be regulated either by NMD or translational repression.
To examine the effect of NMD on the affected genes with frameshift mutations in MSI-H tumors, we selected 12 genes from MSI-H tumors based on the reported frameshift mutation frequencies greater than 30% (ABCF1, ACVR2, hMSH3, hMSH6, hRad50, MARCKS, PRKWNK1, RFC3, SEC63, TAF1B, TCF-4, and TGFBR2). We used genome sequencing of these 12 genes to identify frameshift mutation status. In these 12 genes, we identified 20 frameshift mutations that fell into three categories: 12 mutations were single base pair (bp) deletions, six were 2-bp deletions, and two were single bp insertions in coding mononucleotide repeats (cMNR) (Table S1). All 20 frameshift mutations of the 12 genes resulted in mRNAs containing a PTC (Table 1).
We analyzed mRNA expression of the 12 genes by reverse transcriptase PCR (RT-PCR) in seven MMR-deficient (LS174T, HCT-8, SNU C2A, SNU C4, DLD-1, HCT116, and LOVO) and three MMR-proficient (NCI-H508, SW480, and HT29) colorectal cancer cell lines. Primers were designed to contain at least one exon–exon junction region and to amplify the coding repeat sequences (Table S2). Of the 20 frameshift mutations in the genomic DNA, mutation-derived transcripts were detected from ten alleles representing six genes (hMSH3, TAF1B, TGFBR2, ACVR2, MARCKS, and TCF-4), whereas ten alleles representing six genes (ABCF1, hMSH6, hRad50, PRKWNK1, RFC3, and SEC63) did not generate frameshift mutation-derived transcripts. No differences in expression of frameshift mutation-derived mRNA were observed between cell lines. Of the ten transcripts with frameshift mutations, five transcripts (representing three genes: hMSH3, TAF1B, and TGFBR2) had PTCs more than 50–55 nt upstream of the last exon–exon junction and were therefore expected to be degraded by NMD but instead escaped from NMD (NMD-escape). On the other hand, the five remaining transcripts (representing three genes: ACVR2, MARCKS, and TCF-4) had PTCs within 50–55 nt upstream of the last exon–exon junction and were therefore expected to be irrelevant to NMD (NMD-irrelevant). Accordingly, the 20 transcripts from 12 genes were classified as NMD-sensitive, NMD-escape, and NMD-irrelevant (Table 1).
In order to confirm the effect of NMD on the NMD-sensitive and NMD-escape PTC-containing mRNAs, we used RT-PCR and a ribonuclease protection assay (RPA) to analyze the expression of the target gene mRNAs after treatment with puromycin, a translation inhibitor. In the five NMD-escape alleles that generated detectable frameshift mutation-derived mRNAs, no expression differences were found after puromycin treatment. In the ten NMD-sensitive alleles that produced no detectable frameshift mutation-derived mRNAs, mutant transcripts were detected after puromycin treatment (Figure S2). We analyzed the amount of two degraded NMD-sensitive transcripts, hRad50 and hMSH6, by RPA and found a total loss of mutant transcripts, as evidenced by a 2-fold increase in hRad50 and hMSH6 products after puromycin treatment. In contrast, there was no loss of TGFBR2 mutant mRNA, an NMD-escape transcript, because the amount of product was unchanged after puromycin treatment (Figure 1).
Next, we evaluated the effect of down-regulating UPF1 or UPF2, which are key NMD factors, on the stability of the frameshift mutation-derived mRNAs, hRad50 and hMSH6, using specific small interfering RNA (siRNA). Upon the treatment of luciferase siRNA, expression of the mutation-derived hRad50 and hMSH6 mRNAs were not detected in the cell lines with hRad50 and hMSH6 mutations. In contrast, down-regulating UPF1 or UPF2 abundantly increased the frameshift mutation-derived mRNAs, as confirmed by RT-PCR, and sequence analysis. These findings indicate that frameshift mutation-derived mRNAs of hRad50 and hMSH6 are recognized and degraded by the NMD system (Figure S3).
In order to determine if the truncated protein products from PTC-containing mRNAs can be detected, we first performed Western blotting analyses using antibodies directed against the N-terminus of hRad50, hMSH6, and hMSH3. Truncated proteins were not detected for the NMD-sensitive (hRad50 and hMSH6) genes, whereas wild-type proteins were detected in the cell lines containing the wild-type allele. These results support the previous finding that NMD-sensitive PTC-containing mRNAs are degraded by the NMD system. In the NMD-escape hMSH3 gene, we detected full-length hMSH3 proteins in cell lines with no mutations or with monoallelic mutations in these genes; however, we could not detect truncated hMSH3 proteins in any of the cell lines carrying frameshift mutations (Figure 2). The hMSH3 antibody detected the truncated proteins from the cell lines transfected with mutant hMSH3 cDNA, indicating that the antibodies used in our experiments specifically react with the N-terminal region of hMSH3 protein (Figure S4). Furthermore, we could not detect the truncated proteins of hMSH3 genes after treatment with the proteasome inhibitors MG132 or E64 (unpublished data), which excludes the possibility of rapid degradation of mutated proteins.
We interpreted the failure to detect truncated protein from the NMD-escape PTC-containing mRNAs as follows: (1) truncated proteins were generated, but at an amount not sufficient for detection by Western blotting, (2) truncated proteins were generated, but then rapidly degraded, or (3) truncated proteins were not generated from the mutant mRNA.
To rule out an insufficient amount of endogenous truncated proteins, we constructed expression plasmids with NMD-escape PTC-containing genomic DNA or cDNA of TGFBR2: (1) wild-type cDNA of TGFBR2 (K: TGFBR2(WT)-cDNA), (2) PTC-containing mutant cDNA of TGFBR2 without downstream exons and introns (L: TGFBR2(−1)-cDNA), (3) wild-type TGFBR2 genomic DNA (M: TGFBR2(WT)-splicing), (4) mutant TGFBR2 genomic DNA with a 1-bp deletion (N: TGFBR2(−1)-splicing), and (5) mutant TGFBR2 genomic DNA with a 1-bp deletion and a PTC artificially located in the last exon (O: TGFBR2(−1)-irrelevant) (Figure 3A). Among three NMD-escape PTC-containing mutated genes that we found in MSI-H tumors, we selected TGFBR2. TAF1B and hMSH3 were excluded because of their large size and number of exons, which result in the failure or inefficient transfection of genomic DNA.
In all of the constructs described above, the nucleotide sequences encoding FLAG peptide was introduced immediately downstream of the initiation codon, which allows for detection of the encoded proteins by Western blotting. These vectors were designed to differentiate the effect of spliced wild-type mRNA, spliced mutant NMD-escape mRNA, and spliced mutant NMD-irrelevant mRNA in terms of truncated protein expression. We observed abundant expression of PTC-containing TGFBR2 mRNA in cell lines transfected with TGFBR2(−1)-cDNA, TGFBR2(−1)-splicing, and TGFBR2(−1)-irrelevant (Figure 3B). Cell lines transfected with TGFBR2(WT)-splicing, TGFBR2(−1)-splicing, and TGFBR2(−1)-irrelevant showed accurate splicing, and all normal and mutant mRNA products were confirmed by sequence analysis (unpublished data). A semi-quantitative RT-PCR analysis designed to detect exogenous TGFBR2 mRNA showed similar and abundant amounts of TGFBR2 mRNA expression in all of the cell lines transfected with the five different constructs (unpublished data). In this analysis of protein expression using the anti-FLAG antibody, we demonstrated the expression of wild-type TGFBR2 protein in cell lines transfected with TGFBR2(WT)-cDNA and TGFBR2(WT)-splicing. We also demonstrated the expression of truncated TGFBR2 protein in cell lines transfected with TGFBR2(−1)-cDNA and TGFBR2(−1)-irrelevant. Intriguingly, we could not detect any TGFBR2 protein in cell lines transfected with TGFBR2(−1)-splicing, indicating a selective translational repression of NMD-escape mutant mRNA (Figure 3C).
In order to confirm that translational repression is responsible for the failure to detect truncated protein from PTC-containing TGFBR2 mRNA, we examined the mRNA distribution of TGFBR2(WT)-splicing and TGFBR2(−1)-splicing by polysome analysis.
In the cell line with the TGFBR2(WT)-splicing vector, TGFBR2(WT)-splicing mRNA was found in the polysome-containing fractions similar to endogenous GAPDH mRNA (Figure 3D). However, in the cell line with the TGFBR2(−1)-splicing vector, a greater percentage of TGFBR2(−1)-splicing mRNA was found in the fractions that contained ribosomal subunits and monosomes, whereas endogenous GAPDH mRNA co-sedimented with polysomes (Figure 3E). Furthermore, upon the treatment of puromycin, a greater percentage of TGFBR2(WT)-splicing mRNA and endogenous GAPDH mRNA were shifted into ribosomal subunits and monosome-containing fractions (Figure 3F). In order to rule out the possibility that the weak polysome association of TGFBR2(−1)-splicing mRNA is due to its shorter open reading frame as compared to the TGFBR2(WT)-splicing mRNA, we repeated the same experiment using the TGFBR2(−1)-splicing vector with puromycin treatment. The results show that there is no significant difference in the cell line transfected with TGFBR2(−1)-splicing after puromycin treatment (Figure 3E and 3G). These results indicate that (1) the sedimentation of TGFBR2(WT)-splicing mRNA in heavy fractions was due to polysome association, and (2) the shift of TGFBR2(−1)-splicing mRNA into ribosomal subunits and monosome-containing fractions is due to translational repression, similar to TGFBR2(WT)-splicing mRNA and TGFBR2(−1)-splicing mRNA treated with puromycin (Figure 3E–3G). This novel mechanism, whereby PTC recognition itself triggers translational repression, is referred to as nonsense-mediated translational repression (NMTR).
We demonstrated the selective translational repression of the NMD-escape mutant TGFBR2(−1)-splicing mRNA after normal splicing, and this repression was not found in the NMD-irrelevant mutant TGFBR2 mRNA, which lacks a downstream sequence of the termination codon. Therefore, we examined other possible factors influencing the expression of the truncated protein by: (1) changing the 3′ UTR length (the length between the PTC and poly(A) tail) to check the possible effect of 3′ UTR length on translational repression [16], (2) treating with a proteasome inhibitor (MG132) in the cell lines transfected with TGFBR2(−1)-splicing and TGFBR2(−1)-irrelevant to rule out that the truncated proteins are generated but rapidly degraded by the proteasome, and (3) down-regulating key NMD factors, UPF1 and UPF2, to evaluate whether NMD factors are involved in the translational repression of NMD-escape PTC-containing spliced TGFBR2 mutant mRNA.
In order to change the 3′ UTR length, we constructed another TGFBR2(−1)-irrelevant with a full-length cDNA sequence spanning from the PTC to the 3′ end of TGFBR2 (P: TGFBR2(−1)-irrelevant-F) (Figure 4A). When the genomic DNAs of TGFBR2(−1)-splicing, TGFBR2(−1)-irrelevant, and TGFBR2(−1)-irrelevant-F were transfected, normal splicing and a large amount of mutant mRNAs were present in all three cell lines transfected with the different genomic DNAs (Figure 4C). However, no proteins were expressed in the cell lines transfected with TGFBR2(−1)-splicing, whereas a large amount of truncated proteins were expressed in the cell lines transfected with TGFBR2(−1)-irrelevant. In the cell lines transfected with TGFBR2(−1)-irrelevant-F, the amount of truncated proteins was reduced to about 35% of that of the cell lines transfected with TGFBR2(−1)-irrelevant. These findings indicate that the 3′ UTR length itself or specific cis-acting element(s) within the 3′ UTR seem to contribute to the translational inhibition of TGFBR2(−1) mRNA. However, more importantly, a splicing event downstream of the PTC may be involved in NMTR because truncated proteins are expressed in cells transfected with TGFBR2(−1)-irrelevant-F, but not in cells transfected with TGFBR2(−1)-splicing (Figure 4D).
We excluded the possibility that mutated proteins are generated, but then rapidly degraded by the proteasome, because cell lines transfected with TGFBR2(−1)-splicing and treated with MG132, a proteasome inhibitor, demonstrated no truncated proteins. In contrast, the cell lines transfected with TGFBR2(−1)-irrelevant-F and treated with MG132 demonstrated similar amounts of truncated proteins compared to cell lines only transfected with TGFBR2(−1)-irrelevant-F (Figure 4D).
Finally, we evaluated whether key NMD factors are involved in NMTR. We expected that the most significant difference between PTC-containing TGFBR2 mRNA and PTC-containing TGFBR2 mRNA, which lacks an intron downstream of the PTC, would be the presence of exon junction complexes (EJCs) behind the PTC. An EJC recruits the NMD factors, UPF1 and UPF2, which play a key role in mRNA quality control. We treated cells with UPF1 and UPF2 siRNA in order to elucidate whether these two NMD factors are involved in NMTR. The level of UPF1 was down-regulated to about 20% of normal, where normal is defined as the level in the presence of the nonspecific control, luciferase siRNA, whereas the level of UPF2 was down-regulated to about 10% of normal (Figure 4B). Treatment of any of the siRNAs failed to produce truncated proteins in the cell lines transfected with TGFBR2(−1)-splicing, indicating that at least these two NMD factors do not play an important role in the NMTR of NMD-escape mutant TGFBR2 mRNA (Figure 4D).
We then examined whether NMD-irrelevant PTC-containing mRNA can generate truncated protein. We selected one NMD-irrelevant mRNA, mutant MARCKS, and performed a transfection assay using wild-type MARCKS genomic DNA (MARCKS(WT)-splicing) and mutant MARCKS genomic DNA with a 2-bp deletion (MARCKS(−2)-splicing) (Figure 5A).
We found expression of wild-type and truncated protein in cell lines transfected with MARCKS(WT)-splicing and MARCKS(−2)-splicing, respectively, by Western blotting with an anti-FLAG antibody (Figure 5B). To verify that these protein products were identical to MARCKS, we performed Western blotting with the anti-MARCKS antibody and confirmed the expression of wild-type and truncated proteins (Figure 5C). We also found that the truncated MARCKS protein is subject to active proteasome-mediated degradation; the amount of truncated MARCKS protein increased with time when cells were treated with MG132, a proteasome inhibitor (Figure 5B and 5C).
In this study, we found that some PTC-containing mRNAs are not degraded by the NMD system, and their protein translations are repressed. We therefore suggest that PTC-containing mRNAs resulting from frameshift mutations can be classified into three groups: NMD-sensitive mRNAs, which are degraded by the NMD system; NMD-escape mRNAs, which are not degraded by the NMD system, but do experience repression of protein expression; and NMD-irrelevant mRNAs, which are not recognized by the NMD system, and generate truncated proteins. Our findings indicate that both NMD and NMTR, an additional surveillance mechanism for translational control, are involved in the recognition of PTC and suppression of truncated protein from PTC-containing genes that can be deleterious to cell function (Figure 6).
NMD is a quality control-based surveillance mechanism that protects cells from the potentially dominant negative effects of truncated mutant proteins. The primary role of the NMD pathway is to eliminate nonsense transcripts that result from faulty transcription, alternative splicing, or somatic mutation [17,18]. This pathway selectively degrades mRNAs that prematurely terminate translation due to a frameshift or nonsense mutation. Although NMD is a quality control-based surveillance mechanism, avoidance of NMD by PTC-containing mRNAs has been reported for the mutated genes of many diseases. Moreover, about one third of the alternative transcripts in cells are expected to contain PTCs due to splicing errors and regulated unproductive splicing and translation (RUST). Some of these PTC-containing mRNAs belong to the NMD-escape variety [14,15]. If translated, these NMD-escape mRNAs could produce truncated proteins that may critically interfere with cell viability. Among the PTC-containing mRNAs, some PTCs, which are called NMD-irrelevant mRNAs, are located within 50–55 nt or downstream of the last exon–exon junction and are not detected by NMD. Proteins generated from these types of PTC-containing mRNAs and their causal relationship to specific diseases have been well documented [19–21]. However, there are no reports describing translated proteins from mutation-derived NMD-escape mRNAs, although many NMD-escape PTC-containing mRNAs have been reported [22–24].
In this study, we demonstrated that NMD-escape TGFBR2 mRNA is subject to NMTR. Our transfection study of TGFBR2 constructs demonstrated that PTC-containing mRNAs from mutant TGFBR2 were abundant after transfection of mutant cDNA and mutant TGFBR2 genomic DNAs with a 1-bp deletion. However, truncated proteins were only detected in the cell lines transfected with mutant TGFBR2 cDNA, and no truncated proteins were detected in the cell lines transfected with mutant TGFBR2 genomic DNAs with a 1-bp deletion. In contrast, strong expression of TGFBR2 protein was observed in the cell lines transfected with wild-type TGFBR2 genomic DNA. Next, we confirmed using polysome analysis that the lack of truncated protein translated from PTC-containing TGFBR2 mRNA is due to translational repression, not instability of TGFBR2 mRNA.
The major expected differences between the two PTC-containing TGFBR2 mutant mRNAs and mRNA from the cDNA of TGFBR2 or TGFBR2 genomic DNA are the deposition of the EJC and the possible recruitment of NMD factors to the mRNA during translation termination. We also confirmed the NMTR by demonstrating that mutant mRNA and truncated proteins were efficiently expressed in the cell line transfected with mutant TGFBR2 genomic DNA containing a PTC in the last exon without downstream introns (Figures 3C and 4D). We therefore suspected that EJC proteins and/or NMD factors might play an important role in the NMTR. It is well known that NMD recognizes PTC and downstream splicing events that deposit an EJC at an exon–exon junction. The EJC is composed of proteins involved in splicing and the subsequent steps of mRNA transport and translation. EIF4A3, RNPS1, Y14, and MAGOH are involved in EJC formation, and the EJC–mRNA complex is then exported to the cytoplasm together with nuclear cap binding proteins CBP80/20 and nuclear poly(A) binding protein 2 (PABP2) [25–28]. The mRNA then recruits UPF2 and undergoes a so-called “pioneer” round of translation during mRNA export. NMD occurs when translation terminates more than 50–55 nt upstream of the last exon–exon junction. Transient SURF formation at the termination codon, which is composed of Smg1, UPF1, and translation termination factors eRF1–eRF3, is thought to interact with the downstream EJC so as to trigger phosphorylation of UPF1 and thereby elicit NMD [26–30]. In this study, we demonstrated that key NMD factors, UPF1 and UPF2, did not play an important role in the NMTR, which is evidenced by the fact that treatment of UPF1 and UPF2 siRNA did not produce truncated proteins in cell lines transfected with TGFBR2(−1)-splicing, even though both siRNAs drastically down-regulate endogenous UPF1 and UPF2. Moreover, down-regulating Y14 or EIF4A3, which are EJC components, using siRNA failed to restore translational repression of TGFBR2(−1)-splicing (unpublished data). Our results indicate that NMTR is at work on the NMD-escape PTC-containing TGFBR2 mRNA by some unknown surveillance mechanism. The involvement of another messenger ribonucleoprotein particle (mRNP) complex in this translational repression is essentially unknown. Future studies should be focused on the role of translational repression of the various RNA binding proteins in the PTC-containing mRNP complex.
Several recent reports have demonstrated the importance of termination codon context, especially 3′ UTR length, in PTC recognition [16]. Therefore, we tested two TGFBR2(−1)-irrelevant vectors with short and extended 3′ UTR lengths. If the PTC-containing, 3′ UTR-extended construct failed to produce truncated protein, then unlike NMD, splicing and EJCs may not be involved in the mechanism of NMTR. In this experiment, we demonstrated a 65% reduction of truncated protein in the cell lines transfected with TGFBR2(−1)-irrelevant vector with an extended 3′ UTR length. These findings indicate that 3′ UTR length is an important factor; however, other important factors are involved in NMTR. Because NMTR depends on 3′ UTR length or a putative cis-element residing in the 3′ UTR, it is, in part, reminiscent of EJC-independent NMD. A PTC within the penultimate exon of the β-globin or TPI gene elicits NMD depending on the position of PTC relative to the last EJC [12,31]. However, for a PTC within the penultimate exon that normally elicits NMD, deleting the last intron fails to eliminate NMD, indicating that the last exon has a so-called “failsafe” sequence that allows for PTC recognition and triggers NMD in the absence of a downstream EJC. Intriguingly, this element requires that splicing occur upstream of the PTC because the PTC-containing mRNA that is derived from an intronless TPI or β-globin gene is immune to NMD [12]. It remains to be clarified whether NMTR also requires a splicing event upstream of the PTC. Recently, another case for EJC-independent NMD has been reported in immunoglobulin-μ mRNA [16]. Similar to NMTR, EJC-independent NMD of this mRNA depends on the 3′ UTR length. However, the mode of PTC recognition looks quite different between NMTR and EJC-independent NMD of this mRNA, in the sense that NMTR does not require the NMD factors, UPF1 and UPF2, as shown in our study. Important questions of whether mRNAs targeted by EJC-independent NMD are subject to NMTR should be addressed.
Marked degradation of PTC-containing mRNAs and decreased protein synthesis from PTC-containing mRNAs by the NMD system have been reported in yeast [32]. Another important mRNA surveillance mechanism, nonsense-mediated altered splicing (NAS), has been reported in mammalian cells [33]. NAS induces alternative splicing in the PTC-containing mRNA, thus avoiding the production of toxic mutant proteins. Although the exact mechanism of NAS had not been reported, UPF1 plays an important role in NAS [34]. Together with our findings that UPF1 did not play a significant role in NMTR, all of these findings indicate that (1) NMD, NAS, and NMTR play important roles in the inhibition of deleterious mutant protein production, and (2) unlike NMD and NAS, UPF1 and UPF2 do not a play key role in NMTR, suggesting novel factor(s) or pathways exist in the NMTR.
In conclusion, we demonstrated three different molecular pathways of PTC-containing mRNAs. We propose a novel mechanism of gene expression regulation for PTC-containing mRNAs, in which the deleterious transcripts are regulated either by NMD or NMTR. Future studies of the NMD and NMTR control mechanism will enable us to better understand the reason for specific protein expression among the numerous mRNA isoforms, as well as the selective cellular control mechanism of protein expression.
Ten cell lines were obtained from either the American Type Culture Collection (ATCC; http://www.atcc.org) or the Korean Cell Line Bank (KCLB; http://cellbank.snu.ac.kr). Seven cell lines (LS174T, HCT-8, SNU C2A, SNU C4, DLD-1, HCT116, and LOVO) were MMR-deficient, and three (NCI-H508, SW480, and HT29) were MMR-proficient in terms of their MSI status, as determined by previous studies [35,36]. We confirmed the presence of MSI using BAT26 and BAT25 markers. Cells were grown in RPMI supplemented with 10% FBS (Life Technologies, Grand Island, New York, United States), penicillin, and streptomycin at 37 °C in 5% CO2.
Genomic DNA and cDNA preparation, analysis of MSI, and identification of target gene frameshift mutations were performed as described previously [37].
We used puromycin (Sigma, St. Louis, Missouri, United States) to inhibit the synthesis of proteins involved in NMD. Cells were grown to 80% confluence and then treated with 30-μg/ml puromycin. Six hours later, the cells were harvested, and total RNA was isolated using an RNeasy Mini kit (QIAGEN, Valencia, California, United States) according to the manufacturer's instructions.
Twenty-one–nucleotide RNAs were chemically synthesized using a Silencer siRNA Construction kit (Ambion, Austin, Texas, United States). Synthetic oligonucleotides were deprotected and gel-purified. HCT116 and SNU C2A cells growing in six-well dishes were transfected with 50 nM siRNA and oligofectamine (Invitrogen, Carlsbad, California, United States) according to the manufacturer's protocol. For RT-PCR analysis, total RNA was harvested 48 h after siRNA transfection. Targeted nucleotides, numbered relative to the start codon, were as follows: rent1/UPF1, 1,879–1,901 (5′-AAGATGCAGTTCCGCTCCATTTT-3′); rent2/UPF2, 1,423–1,445 (5′-AAGGCTTTTGTCCCAGCCATCTT-3′); and luciferase GL2, 153–173 (5′-AACACGTACGCGGAATACTTCGA-3′). The inhibition of UPF1 and UPF2 expression by siRNA targeting was evaluated by semi-quantitative RT-PCR or Western blotting.
To study expression of TGFBR2, we selected the secreted expression vector, pSecTag2B (Invitrogen). The vector was cut by Hind III, and the FLAG oligonucleotide was inserted into the Hind III site to allow for specific immunodetection, thereby creating the pSecTag-FLAG vector. The complete coding sequence of full-length TGFBR2 begins with the first ATG at codon 1 and encodes a 568–amino acid protein, with a stop codon at 569. Mutant TGFBR2 with a 1-bp deletion at ten adenosine repeats results in a premature stop at codon 162 (Figure S1). We constructed wild-type (constructs K and M) and truncated TGFBR2 expression vectors (constructs L, N, O, and P).
Wild-type TGFBR2 cDNA was obtained from a 293T cell line and mutant (1-bp deletion) TGFBR2 cDNA was obtained from a HCT116 cell line by RT-PCR and then cloned using a T&A cloning kit (RBC, Taipei, Taiwan). Construct K (TGFBR2(WT)-cDNA) and construct L (TGFBR2(−1)-cDNA) were generated by inserting the Hind III fragment from the TGFBR2 cDNA into the Hind III site of pSecTag-FLAG.
In order to analyze splicing and subsequent protein expression, we inserted wild-type and mutant TGFBR2 genomic DNA into the expression vectors. To generate constructs M (TGFBR2(WT)-splicing) and N (TGFBR2(−1)-splicing), exons 1 to 7, except exon 3, of TGFBR2 were obtained by PCR using 293T cell genomic DNA and cloned into the yT&A cloning vector (RBC). Exon 3 of TGFBR2 was obtained from LOVO and 293T DNA by PCR, because a 1-bp deletion in the cMNR exists in exon 3 of TGFBR2 in LOVO. All primers for cloning were designed to contain more than 100 bp of intron sequence on each side of the exon boundary to ensure accurate splicing of the construct. These exon fragments were ligated to each other using a restriction enzyme site in the multiple cloning site (MCS) of the yT&A vector. In the case of exon 1, deletion of the signal peptide was performed using Dpn1 for N-terminal FLAG tagging. The ligated genomic construct of TGFBR2 was inserted into the pSecTag-FLAG vector. To generate construct O (TGFBR2(−1)-irrelevant), exon 4 to exon 7 of TGFBR2(−1)-splicing was deleted by cutting with BstX1, and the remaining construct was self-ligated. To generate construct P (TGFBR2(−1)-irrelevant-F), cDNA from exon 4 to exon 7 of TGFBR2 was inserted into the pSecTag-FLAG vector using EcoR1 and Pst1 restriction enzyme sites, and then genomic DNA from exon 1 to exon 3 was inserted into the same vector using the BamH1 restriction enzyme site.
For the expression study of MARCKS, the expression vector pcDNA3.1(+) (Invitrogen) was cut by Hind III, and the FLAG oligonucleotide was inserted at the Hind III site to allow for specific immunodetection, thereby creating the pcDNA-FLAG vector. In order to analyze splicing and subsequent protein expression, we inserted wild-type and mutant MARCKS genomic DNA into expression vectors. To generate constructs P (MARCKS(WT)-splicing) and Q (MARCKS(−2)-splicing), exon 1 of MARCKS was obtained through PCR using genomic DNA of 293T and cloned into the yT&A cloning vector (RBC). Exon 2 of MARCKS was obtained from SNU C2A by PCR, since 2-bp monoallelic deletions in the cMNR exist in exon 2 of MARCKS in SNU C2A. These exon fragments were subcloned into the expression vector pcDNA-FLAG. Primer sequences are shown in Table S3.
HCT116 and HeLa cells (2 × 106) were transiently transfected using Lipofectamine 2000 (Invitrogen) with the specific construct plasmid and pSecTag-FLAG vector in a 60-mm plate. Cells were harvested 2 d later. Protein was purified from half of the cells using passive lysis buffer (Promega, Madison, Wisconsin, United States), and total RNA was purified from the other half using TRIzol Reagent (Invitrogen).
Whole lysates from cell lines were prepared using passive lysis buffer (Promega). Thirty micrograms of the total protein lysates were loaded into each lane, size-fractionated by SDS-PAGE, and then transferred to a PVDF membrane that was blocked with TBST containing 5% skim milk. Primary antibodies against GAPDH (Trevigen, Gaitherburg, Maryland, United States), FLAG (Sigma-Aldrich), hRad50 (Ab13B3; Gene Tex), MARCKS (Santa Cruz Biotechnology, Santa Cruz, California, United States), or hMSH6 and hMSH3 (BD Bioscience, Mountain View, California, United States) were incubated for 1 h at room temperature. After washing, membranes were incubated with HRP-conjugated secondary antibody (Santa Cruz Biotechnology), washed, and then developed with ECL-Plus (Amersham Pharmacia Biotech, Little Chalfont, United Kingdom).
RNA samples were extracted from each cell line using TRIzol Reagent (Invitrogen). RPA was carried out according to the instructions provided by the manufacturer (RiboQuant RPA kit; BD Biosciences) using 20 μg of total RNA per sample. The template antisense 32P-labeled RNA probes were specific for TGFBR2 (unprotected, 271 nt; protected, 254 nt), hRad50 (371, 353), hMSH6 (332, 312), and GAPDH (134, 120) mRNA. GAPDH was used as an internal control. The intensity of specific bands corresponding to individual riboprobes was determined by densitometry. All RPAs were evaluated at different exposures, and only bands that were within the linear range of the film were analyzed.
Total RNA was extracted from HCT116 with TRIZOL Reagent (Invitrogen). Full-length TGFBR2 cDNA was amplified by RT-PCR using 293T total RNA. Expression of the exogenous TGFBR2 construct mRNA was analyzed by Northern blot analysis using 10 μg of total RNA according to standard protocols.
Cultures were supplemented with 100-μg/ml cycloheximide for 24 h post-transfection and incubated for 5 min at room temperature. HeLa cells were washed three times with ice-cold PBS containing 100-μg/ml cycloheximide. HeLa cells were collected by scrapping in PBS, transferred to Eppendorf tubes for additional washes, and then lysed in lysis buffer (15 mM Tris-Cl [pH 7.4], 3 mM MgCl2, 10 mM NaCl, 0.5% Triton X-100, 100-μg/ml cycloheximide, 1-mg/ml heparin, and 200 U RNasin [Promega]). When indicated, puromycin (100 μg/ml) was added to the cultures 2 h prior to harvesting, and cycloheximide was omitted from the gradient. For each construct, lysates from two 100-mm dishes were pooled into a microcentrifuge tube and incubated for 10 min on ice with occasional mixing. Nuclei and debris were removed by centrifugation at 12,000g for 2 min. Then, 1 ml of each cytoplasmic lysate was layered onto an 11-ml 10%–50% sucrose gradient and centrifuged at 4 °C in an SW40 rotor (39,000 rpm) for 2 h. Sixteen fractions were collected from the top with concomitant measurement of absorbance at 254 nm, using a fraction collection system. RNA was extracted with TRIZOL Reagent and analyzed by RT-PCR.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession numbers for the gene products discussed in this paper are as follows: ABCF1 (NM_001090), ACVR2 (NM_001616), EIF4A3 (NM_014740), GAPDH (NM_002046), hMSH3 (NM_002439), hMSH6 (NM_000179), hRad50 (NM_005732), MARCKS (NM_002356), PRKWNK1 (NM_018979), RFC3 (NM_002915), SEC63 (NM_007214), TAF1B (NM_005680), TCF-4 (NM_030756), TGFBR2 (NM_003242), UPF1 (NM_002911), UPF2 (NM_015542), and Y14 (NM_005105).
|
10.1371/journal.pbio.1002245 | Face Patch Resting State Networks Link Face Processing to Social Cognition | Faces transmit a wealth of social information. How this information is exchanged between face-processing centers and brain areas supporting social cognition remains largely unclear. Here we identify these routes using resting state functional magnetic resonance imaging in macaque monkeys. We find that face areas functionally connect to specific regions within frontal, temporal, and parietal cortices, as well as subcortical structures supporting emotive, mnemonic, and cognitive functions. This establishes the existence of an extended face-recognition system in the macaque. Furthermore, the face patch resting state networks and the default mode network in monkeys show a pattern of overlap akin to that between the social brain and the default mode network in humans: this overlap specifically includes the posterior superior temporal sulcus, medial parietal, and dorsomedial prefrontal cortex, areas supporting high-level social cognition in humans. Together, these results reveal the embedding of face areas into larger brain networks and suggest that the resting state networks of the face patch system offer a new, easily accessible venue into the functional organization of the social brain and into the evolution of possibly uniquely human social skills.
| Primates have evolved to transmit social information through their faces. Where and how the brain processes facial information received by the eyes we now understand quite well. Yet we do not know how this information is made available to other brain areas so that a face can evoke an emotion, activate the memory of a person, or draw attention. Here, to identify brain regions interacting with face areas, we performed whole-brain imaging in macaque monkeys, whose face-processing system we know best. We find that the core face-processing areas are connected to several other brain areas supporting socially, emotionally, and cognitively relevant functions. Together, they form an extended face-processing network, similar to what has been proposed for humans. This extended face-processing network intersects with a second large-scale network, the so-called “default mode network”, in a pattern stunningly similar to that in the human brain. This intersection identifies selectively those brain regions that implement the most high-level forms of social cognition, such as understanding others’ thoughts and feelings. Thus, the results of this novel approach to understanding the functional organization of the social brain point to a deep evolutionary heritage of human abilities for social cognition.
| Primates are highly social animals who cope with the challenges posed by life in complex social groups through sophisticated mechanisms for the recognition, evaluation, and generation of social signals. To understand the neural circuits mediating primate social behavior, here we take a novel, bottom-up approach utilizing a particularly well-defined sensory circuit as our starting point: the neural machinery that processes faces. Faces transmit rich information about socially relevant dimensions such as personal identity, emotional expressions, and gaze direction [1]. To extract this multidimensional facial information, primates have evolved specialized brain areas [2], which are tightly and specifically interconnected [3], to form a face-processing system. Yet how this face-processing system is embedded, and thus how it makes face information available to other systems supporting social and cognitive functions, is largely unknown.
The connections between face processing and cognition are important to understand because behavioral and developmental studies show that faces occupy a special status among other, socially less relevant objects. Faces selectively draw spatial attention [4] and attract saccades much faster than other objects do [5], indicating privileged routing of facial information into attentional and eye movement control systems. Faces also drive specific mnemonic, emotional, and communicative responses [6], again suggesting specialized circuitry linking face areas to recipient, non-face-selective areas elsewhere in the brain.
To reveal with which parts of the brain the face-processing system can exchange information, we used resting state functional magnetic resonance imaging (rsfMRI) seeded in functionally defined face areas of the macaque monkey, the main animal model for face processing. rsfMRI noninvasively measures functional connectivity (FC) between brain areas based on spontaneous low frequency activity correlations, at high spatial resolution, and with full brain coverage [7]. We focus on FC because although FC generally shows good agreement with anatomical connectivity [8], the set of potential functional connections between brain areas is far greater than that of direct structural links, as it is not constrained to monosynaptic connections but also includes dynamic, polysynaptic connectivity [9]. A previous study of face patch connectivity using electrical microstimulation, a method that reveals primarily monosynaptic connectivity, found face patches to form a structurally closed system with few output connections [3], thus begging the question how the face patch system interacts with other systems. rsfMRI provides this type of complementary information and—due to its wide use in basic and clinical research in humans—confers the additional advantage of being readily comparable between species [10], thus enabling insight into the evolution of face-recognition systems [2,11,12].
Face area FC maps are not only essential for understanding the neural mechanisms of face recognition, they also provide a unique inroad into understanding high-level social cognition and its evolutionary heritage. This is because face areas are thought to constitute a major input into the so-called “social brain”, a set of brain areas devoted to the processing of social interactions [13]. In fact, almost 60% of the variance in our attitudes towards others can be explained by facial information alone [14]. In humans, one of the social brain’s core regions is the temporoparietal junction (TPJ). The TPJ is thought to be critical for high-level social cognition, in particular theory of mind (TOM) [15], the capacity to attribute mental states to ourselves and others. Apes and monkeys display basic forms of TOM such as understanding what others see or know [16,17]. However, the very existence of a TPJ homolog in monkeys has been debated since the days of Brodmann, in part because the high-level functions that are supported by human TPJ, such as understanding others’ false beliefs, may not be present in the macaque [15,18]. This uncertainty is in large part due to the difficulties in studying macaque social cognition in a controlled experimental setup [19,20]. Our approach sidesteps this issue and enables us to assess whether the kind of processing architecture enabling social cognition in humans already exists in the macaque: if this architecture was entirely absent, this would indicate that certain aspects of social cognition are indeed uniquely human. Conversely, if we could uncover similar brain networks in the macaque, this would suggest that at least a minimal scaffolding for high-level social cognition is already present in a primate whose evolutionary lineage split from ours some 25 million years ago [21].
To achieve this goal, we make use of the fact that the human social brain overlaps with another large scale network, the so-called default mode network (DMN) [22–24]. The DMN, readily identifiable with rsfMRI in both humans and monkeys [25,26], comprises a set of interconnected areas more active during rest than task performance and is thus thought to generate the brain’s default activity [27]. Importantly, the overlap between the human DMN and the social brain includes high-level social cognition areas like TPJ, medial posterior parietal cortex (PPC), and dorsomedial prefrontal cortex (dmPFC) [23,28]. We thus assessed, using the face patch resting state networks as a proxy for the social brain, whether and where a similar overlap exists in the macaque brain. This approach allows us to identify candidate homologs of human high-level social cognition brain areas.
To determine face patch resting state networks (FPRSNs), we first identified face patches in six awake macaque monkeys using standard face localizers. We presented pictures of faces, bodies, and other object categories and contrasted activation during face presentations with activations during the presentation of nonface stimuli to reveal face areas. We identified one orbitofrontal (prefrontal orbital, PO) [29] and five temporal (middle lateral, ML; middle fundus, MF; anterior lateral, AL; anterior fundus, AF; anterior medial, AM) [2] face patches in all six animals. For subsequent analyses, whenever possible, bilateral homolog pairs were joined into one region of interest (ROI). Subsequently, the same monkeys underwent scans for rsfMRI during light isoflurane anesthesia. Aligning face localizer with rsfMRI scans allowed us to extract the time courses of spontaneous activity from each of the face patches and from regions outside of the face patch system.
As a first step towards characterizing FPRSNs, we determined connections within the face patch system. Our goals were to (i) reveal the hitherto unknown FC between temporal and orbitofrontal face patches, (ii) determine whether the established organizational principles of the temporal lobe face patch system, i.e., hierarchical and parallel organization, can be recovered from rsfMRI, and (iii) validate the intra-face patch connectivity pattern our noninvasive methodology reveals with data from an invasive approach, i.e., electrical microstimulation. To this end, we performed ROI-to-ROI correlation analyses after regressing out motion, heartbeat, and breathing artifacts from the data. We found FC between all face patches, with average Pearson correlation coefficients ranging between 0.34 for ML-MF and 0.02 for ML-AM (Fig 1, Wilcoxon signed rank tests, one-sided, corrected for multiple comparisons using the False Discovery Rate (FDR) at q = 0.05). The frontal patch PO, whose connectivity pattern was previously unknown, showed significant FC with the temporal patches MF, ML, and AM. AM, the face patch residing at the top of the temporal face-processing hierarchy [30], showed significant FC with PO, a likely output structure for AM, and with AF and AL, two input areas to AM located at the preceding level of the processing hierarchy, but not ML and MF, which are one level further removed. Thus, rsfMRI FC patterns recover the first organizational principle of the face-processing system, i.e., its hierarchical organization along the posterior–anterior axis of the superior temporal sulcus (STS). The second main organizational principle of the face patch system is parallelism: two processing streams reside in different cytoarchitectonic subdivisions of the STS, one in the fundus and the other in the lower bank of the STS. Hence, we tested whether the strength of FC followed known anatomical patterns of connectivity [31], i.e., whether face patches residing either in the fundus or on the lip of the STS (MF-AF and ML-AL) are more strongly correlated with each other, or whether their connectivity was equally strong across the fundus and the lip of the STS (AF-ML and AL-MF). Indeed, we found stronger FC within cytoarchitectonic subdivisions than across (medianHL difference CI95 = [0.004 0.146]; Wilcoxon signed rank test, p = 0.03, two-sided), but this effect was mainly driven by the differential connectivity of ML. Anatomy also predicts a small but systematic bias for stronger interhemispheric connectivity between homolog than nonhomolog brain areas [32], and interhemispheric FC was in fact slightly higher between homolog face patches than between nonhomolog face patches (medianHL difference CI95 = [0.0037 0.2326]; Wilcoxon signed rank test, p < 0.02, one-sided). Finally, we compared the pattern of FC between the face patches from rsfMRI to previous results obtained with electrical microstimulation [3], which had revealed a highly specific set of connections between the STS face patches. As these specialized anatomical links provide the major scaffold for the functionally defined face processing network, we expected that rsfMRI should reveal a similar (although not identical) pattern of FC. A Spearman rank correlation between connection strengths indeed showed a high level of agreement between the two methods (r = 0.6314, p = 0.01; S1 Fig). The same results were obtained using the robust correlation method Shepherd’s Pi (r = 0.6352, p = 0.02) [33]. This substantiates the noninvasive rsfMRI approach with results from a method known for its specificity in identifying direct neuronal connections [34].
The second main goal of this study was to determine the embedding of the face patches into the rest of the brain. We first focused on connections with other cortical areas. To this end, we aligned and brought all functional data into a common surface space, preserving high specificity for cortical grey matter and anatomical landmarks despite slight smoothing (1.25 mm kernel). We then performed a fixed effects (FFX) General Linear Model (GLM) group analysis for each of the face patches, respectively. After correction for multiple comparisons, these analyses revealed a highly convergent set of connected areas for AL, AF, MF, ML, and PO (and the posterior lateral face patch, PL, which was identified only in a subset of animals, S2 Fig), as well as connections that were unique to individual face patches (Fig 2, S3–S9 Figs). The FC pattern common to all face patches can be summarized by a conjunction analysis using the minimum statistic from each of the five maps (Fig 2, center). Both temporal and orbitofrontal face patches were connected to the (i) lateral prefrontal cortex (Fig 2, light blue), including areas 12 and 46, where face-selective neurons have been located [36]; (ii) regions of premotor cortex (Fig 2, blue), including areas F2, F4, F5, and F7, involved in the visual guidance of movements [37]; (iii) inferior parietal areas, including areas 7a and 7b (Fig 2, blue); (iv) areas of the temporal lobe, including the lower bank, fundus, and upper bank of the STS and parts of area TE (Fig 2, green); and (v) early visual cortex, especially areas V3 and V4 (Fig 2, green). This pattern of results was highly consistent across hemispheres (S3 Fig). Within this common pattern of connectivity shared across face patches, we also found significant variation across face patches (S9 Fig): (i) connectivity to the insula was more prominent for AL, ML, and PO than for MF and AF, (ii) AL and AF connectivity extended more posteriorly on the dorsolateral surface towards the central sulcus than any of the other face patches, and (iii) only AM showed connectivity to medial temporal lobe structures (entorhinal cortex, perirhinal cortical areas 35 and 36).
Face-selective neurons have been found in several subcortical structures such as the pulvinar [40] and the amygdala [41], which are not accessible to surface-based analyses. To determine FC of the cortical face patch system with subcortical structures, we performed a volume-based whole-brain analysis. After smoothing (2 mm Gaussian kernel), a FFX GLM revealed subcortical FC of the face patch system with a subregion of the claustrum, the amygdala, and the pulvinar, as had previously been shown using microstimulation [3], and additionally with the geniculate nucleus of the thalamus, the caudate nucleus, and the hippocampal formation (Fig 3).
It has been suggested that visual categories represented in spatially disjunct parts of ventral visual cortex are associated with unique patterns of connectivity [42]. We thus tested which of the functional connections we observed were specific to the face patches. To this end, we isolated a patch in the anterior lip of the STS that responded to manmade objects during the localizer scans and then contrasted its FC to that of anatomically neighboring face patch AL (see Materials and Methods). AL showed stronger connectivity with the upper bank of the STS (including area TPO), the insula, lateral and medial parietal cortex (including area 7 and 23, respectively), lateral and medial prefrontal cortex (including areas F4, F5 and 6M, 9, respectively), as well as orbitofrontal cortex (area 13) (Fig 4). In contrast, the nearby object patch showed stronger connectivity with the inferotemporal cortex (including area TE) as well as the occipital cortex (including area VOT). Hence, faces and objects indeed display distinct patterns of FC.
Overall, the FC pattern of the face-patch system we identified includes areas like the lateral prefrontal cortex and the amygdala that contain face representations themselves, and others, like the premotor cortex, that likely do not. How strongly then does functional specialization for faces shape face patch FC? Specifically, we tested whether the strength of FC on the whole-brain level depended on the selectivity of the target voxels, i.e., whether face patches were more connected to other face-selective voxels than to object-selective voxels. First, as a measure of selectivity, we computed d’ between faces and objects for each voxel from the localizer data. Because rsfMRI connectivity falls off with distance, we also calculated the Euclidean distance from the voxel with peak selectivity within each respective face patch to the remaining voxels within the same hemisphere. We then matched voxels outside the face patch under consideration for selectivity and distance, weighing both factors equally (see Materials and Methods). Finally, we compared FC with matched face and object-selective voxels across 12 hemispheres, and found that for each face patch (AF, AL, MF, ML, PO), connection strength was higher for face than object voxels (mean differences: AF 0.04, AL 0.03, MF 0.04, ML 0.02, PO 0.03, paired t tests, all p < 0.03, one-sided). This shows that whole-brain rsfMRI can recover functional specificity within connection patterns, similar to what has been shown for structural connectivity in humans [43].
Can the neural systems that support the most advanced human sociocognitive skills be traced back to the macaque, a species with more limited sociocognitive abilities? In humans, social information processing networks, broadly defined, and the DMN overlap in the three cortical areas supporting the most high-level social-cognitive functions [22–24]. Should the two networks intersect in macaques as well, and should this intersection occur at anatomical locations corresponding to those in humans, this would support a scenario of deep evolutionary heritage of these sociocognitive abilities. We first determined the DMN according to its original definition [25], i.e., by seeding rsfMRI in a bilateral ROI placed in medial PPC (areas 31/PGm, see Materials and Methods). As in previous studies [25,26,44], we observed a network comprising PPC, medial prefrontal, and lateral temporoparietal cortex (Fig 5a). We then computed conjunction maps between the FPRSNs and the DMN. Fig 5c shows that there is significant overlap in area TPO in the posterior STS where the human TPJ resides. This result replicates for all face patches (including AM, S10 Fig). Furthermore, there is also overlap in areas 9M/10 in the dmPFC and, although less consistent in its precise location for each of the FPRSNs, areas PGm/23 in the medial PPC (Fig 5d), two further areas involved in high-level social cognition for which overlap with the DMN has been observed in humans [23]. Because the overlap critically depends on the statistical threshold at which it is evaluated, we calculated Jaccard indices, which can be interpreted as percent overlap between two networks, over a wide range of uncorrected thresholds and compared the empirically observed degree of overlap with distributions of Jaccard indices obtained from the overlap of the FPRSNs with randomly generated maps that had the same spatial and statistical properties as the DMN. After correction for multiple comparisons, we found significant overlap between the DMN and all FPRSNs, including that of AM, until thresholds were so conservative that the likelihood of overlap was minimal (S11 Fig). To assess whether the overlap with the DMN was specific to the FPRSNs, we also calculated the overlap between the object patch resting state network and the DMN. There was significantly more overlap between the FPRSN of AL and the DMN than between the object patch resting state network and the DMN at all thresholds (FDR-corrected, q = 0.01), and in fact, for most thresholds tested, there was no overlap between the object patch resting state network and the DMN at all (S12 Fig). Furthermore, voxels which showed overlap between the FPRSN of AL and the DMN in posterior STS, dmPFC, and medial PPC were more strongly connected with AL than with the nearby object patch, while the opposite was the case for a region around the occipitotemporal sulcus, an area in which FPRSN and DMN also overlapped but that is not considered part of the monkey DMN (Fig 5, grey inset). Together, these results show that the macaque FPRSNs and DMN prominently and specifically overlap in brain areas that support high-level social cognition in the human brain.
We find that the macaque face patches form a network linking highly face-selective regions in the temporal and orbitofrontal cortices. This face patch network is functionally embedded into a larger-scale, anatomically specific network of cortical and subcortical structures. This extended network significantly overlaps with the DMN, in particular in posterior STS, medial PPC, and dmPFC, which are involved in high-level social cognition in humans. The overlap is specific to the face patches, considering both the amount of overlap as well as the strength of FC in comparison to object resting state networks. Our results show the utility of a combined fMRI-rsfMRI approach in determining the embedding of functionally specific brain areas into larger-scale brain networks, and indicate that the face patch system may provide a unique inroad into understanding the complex organization of the social brain and a window into the evolution of primate social cognition.
Human face recognition has been proposed to rely on a “core system” consisting of interconnected face-selective areas and an “extended system” that utilizes inputs from the core system for cognitive, emotive, and mnemonic functions [6]. Our whole-brain FC maps of the macaque face-processing system (Figs 2 & 3) show how the face patches are embedded into a larger-scale network that shares, as we will discuss below, many of the properties of the proposed human extended system. In addition, these maps show how the face patches are nested into the general flow of information along the visual ventral stream. The ventral stream is organized along a main posterior–anterior axis, and along a dorsal–ventral axis with extensive lateral connectivity [45]. This connectivity pattern can account for the coupling of face patches to occipital and temporal areas we observed and likely reflects the input–output relationships directly relevant for visual face-processing. A subcortical face-processing system has been proposed to exist, consisting of the superior colliculus, the pulvinar, and the amygdala [46]. FC of the face-processing system with pulvinar and amygdala, as we found here, is compatible with this proposal of a separate, nonclassical set of subcortical inputs into the cortical face-processing network.
The face patch system strongly interconnects with lateral prefrontal cortex, one of the main recipients of ventral stream output [45]. Since connectivity from ventral stream face patches appears to include, but does not appear to be confined to, face specializations within the lateral prefrontal cortex [29,36] and is only partially specific to the face patches, facial information is likely made available for both face-domain-specific processing like face-specific working memory [36] and for domain-general cognitive processes in prefrontal cortex-like categorization [47] or attentional control [48]. Further structures guiding spatial attention and eye movements are the supplementary eye field (SEF) in area F7, PPC, the pulvinar, the amygdala, and the caudate nucleus [49], all of which we found to be connected with the core face-processing system. These connections may aid in relaying information about the direction of attention of others extracted from a visual analysis of eyes and faces into the attentional system [50], and stronger connectivity of face patches than of nearby object patches with areas such as the SEF may underlie the behavioral advantages in directing saccades [5] and drawing spatial attention towards faces [4]. Thus, the core face-processing system interfaces with attention and executive control systems through multiple functional routes.
Extensive further connectivity of the face patch system to executive systems beyond those for oculomotor control was evidenced by face patch-specific rsfMRI connectivity to several parts of premotor cortex. This came as a surprise to us, since it was not predicted by classical anatomy. Current anatomical evidence for direct connections between the parts of the STS that contain the face patches is limited to area F7 [51,52]. Polysynaptic projections, however, from the STS to area F2 have recently been identified [53]. The latter may arise from relay through the ventrolateral prefrontal cortex [53] or through well-established connectivity between parietal areas 7b and S2, which provide input to areas F4 and F5 [37]. These premotor areas all contain visually responsive neurons and have been shown to be involved in the visual guidance of movements of face, eyes, and upper limbs; in particular, area F5 contains mirror neurons for socially relevant movements of the mouth such as lip smacks [54]. Thus, connectivity between face and premotor areas may be stronger than previously thought and may support social communicative functions.
One of the main sets of functions proposed for the extended face-processing system lie within the emotional domain. We found the face patches to be connected with the amygdala, orbitofrontal cortex, and the insula, three structures implicated in the processing of emotions, which, in humans, have been shown to be involved in evaluating faces on social dimensions such as trustworthiness [55]. The amygdala in particular also processes facial expression and gaze direction [56], two of the most important facial cues for social interactions. The orbitofrontal cortex, tightly interconnected with the amygdala, is thought to support the assignment of valence [57]. Thus, functional links between the face patches and core structures of the emotional brain exist, which may serve the utilization of facial information for the generation of emotional responses.
The fourth main set of connections we observed linked the face patch system, and in particular AM, the area at the top of the face-processing hierarchy [30], to structures supporting long-term memory, notably entorhinal cortex, the hippocampal formation, and the claustrum. The latter has been shown to be heavily connected to the temporal lobe and has been hypothesized to act as a relay for sensory inputs to mediotemporal memory areas [58]. Mediotemporal areas, including the hippocampus, contain face-selective cells thought to encode episodic memories [59]. Thus, FC of the most anterior–ventral face area to mediotemporal lobe structures exists, possibly supporting the encoding and retrieval of memories of familiar individuals. Taken together, we find evidence for the core face-processing system to be functionally connected to areas that are known to support cognitive, emotional, communicative, and mnemonic functions. Macaques thus appear to possess an extended face-processing system as proposed for humans [6].
Neuroimaging studies investigating FC of human face-processing areas have found connectivity patterns that are broadly consistent with those we obtained in the macaque. In both species, correlations between the more posterior face-processing regions (the occipital face area (OFA) and the fusiform face area (FFA) in humans, and MF, ML, AF, and AL in the monkey) are stronger than with more anterior face-processing regions [60]. FC outside the core face-processing system also displays a similar pattern in monkeys and humans, with overlap in the occipital, temporal, and frontal lobes [61,62], as well as subcortically, including the hippocampus, amygdala, caudate nucleus, and thalamus [63]. Thus, internal and external FC of core face-processing areas is similar across the two primate species, consistent with the hypothesis of a deep evolutionary heritage of human face recognition abilities. The core face-processing systems in humans and monkeys are composed of multiple face areas. Establishing their homologies based on criteria like relative location and functional specialization [12] has proven difficult. Connectivity is a third, strong, and independent criterion for homology. The differences in face patch connectivity we found, in particular with mediotemporal lobe areas and the insula (S9 Fig), provide clear predictions for human combined fMRI/rsfMRI studies that localize all major face areas and map their FC. Currently available data on human face area connectivity, although not entirely consistent, points to a differentiation between dorsal and ventral face areas. Specifically, it has been found that the posterior STS displays stronger FC to premotor cortex than ventral face-processing areas, in particular the FFA [61,64] (but see [62]), in line with the finding that the STS face areas, but not the FFA, show structural connectivity to the ventrolateral prefrontal cortex [65]. If this differentiation between dorsal and ventral face areas is confirmed, it would indicate homology of the entire macaque face-processing network with the human dorsal face areas, a quite radical view that has previously been put forward on other grounds [11]. More data, in particular from the more variable human brain, will be needed to fully exploit the potential that these FC patterns hold in establishing homologies. Differential FC of the macaque face patches as we found hints at functional differentiations within the system and specialized roles these areas might play in social behavior. Comparing human and monkey rsfMRI connectivity networks to establish homologies as illustrated for the case of face areas is an approach that can be taken even further to understand the networks supporting complex cognitive functions.
It has been a long-standing question whether certain high-level sociocognitive skills, e.g., the ability to reason about the contents of other persons’ mental states, are uniquely human [15,18]. For example, monkeys do not infer the belief that someone has about his/her own state of mind, something humans do routinely [66]. There is behavioral evidence that monkeys display basic forms of TOM such as understanding what others, including human agents, see or know [17]. However, even this interpretation remains contested, since such behavior may also arise from reasoning about the observable behavior of others without explicitly representing the others’ mental state, i.e., without a TOM [20]. Even more so, there is uncertainty about the neural basis that supports high-level social cognition. A case in point is human area TPJ, for which an old-world monkey homolog has been outrightly rejected [67], or proposed to reside either in parietal area 7a [68] or the posterior STS [69]. While a whole battery of tasks is available to characterize high-level social cognition in humans, it has remained difficult to study the sociocognitive abilities of nonhuman primates experimentally. Here, we bypassed this issue and used a novel mapping strategy that utilizes the overlap of social brain areas and the DMN to identify putative homologs of human high-level social cognition systems. In humans, the main overlap between these two networks localizes to areas TPJ, PPC, and dmPFC. These areas have been strongly implicated in high-level social cognition, e.g., TOM in area TPJ and PPC, as well as the understanding of triadic interactions in the dmPFC [15,70,71]. We now find that a similar overlap between FPRSNs and DMN exists in the macaque. The overlap is specific to FPRSNs and prominently includes area TPO in the dorsal posterior STS where the human TPJ resides. Little is known about the role of TPO in social cognition, but it shares several functional characteristics with human TPJ: Like TPJ, TPO is a polysensory area [72] that responds to biological motion [73] and action observation [74] and is involved in attention [75,76]. Together with the connectivity overlap we observed, this suggests that a TPJ precursor or homolog exists in the macaque dorsal posterior STS and where it exists. Interestingly, this area is distinct from a more anterior STS region that has been shown to correlate with social network size [77].
Furthermore, we also find face-patch-specific overlap in areas 9M/10 in the dmPFC and areas PGm/23 in the medial PPC, which are well-established components of the human social brain and anatomically homologous in both species. Electrophysiological recordings during action- and error-monitoring of others [78,79] implicate the dmPFC in at least some aspects of social cognition in the monkey, which may form a precursor for the high-level sociocognitive functions that are supported by the dmPFC in humans [70]. Furthermore, the human dmPFC is also involved in more basic forms of social processing that explicitly rely on facial information, such as gaze following [80], possibly a consequence of functional integration with face processing. The medial PPC, which has been linked to understanding social interactions [81], inferring other people’s thoughts [82], and attributing mental states to others [83] in humans, has been shown to be active during action observation in monkeys [84], a basic ingredient for understanding the intentions of others. Thus, in addition to location and connectivity overlap, the functional properties of these areas are suggestive of a role in social cognition in the macaque.
Fig 5c also shows overlap around the occipitotemporal sulcus, including parts of areas V4, TFO, and TEO. In contrast to the other three regions, overlap in this area was not face-specific and connectivity to an STS object patch was even stronger than to a nearby face patch. This region has been found to be part of the human DMN [85] and is known to be anatomically connected both to the STS [45] and medial PCC [86,87], where our seed regions were located. Thus, connections exist that link this location to the DMN and face processing, rendering overlap in this region plausible. However, they do not suggest a specialized role of posteroventral cortex in social cognition.
Taken together, we find a pattern of overlap between the FPRSN and the DMN that includes the very areas that are selectively active in humans conducting the complex mental operations of TOM in macaque monkeys at rest. Hence, tapping into the social brain via an easily accessible sensory route uncovers similar regions as going through explicitly social cognition tasks in humans. This offers the exciting possibility to uncover the neural computations underlying sociocognitive operations. Since this overlap is present even under anesthesia, it is unlikely to reflect “mentalizing” as a default mode of processing [23], but it suggests a connectivity basis upon which a social default mode could have arisen. Our results thus point to a deep evolutionary heritage of a brain network composed of at least three areas for high-level social cognition.
We used fMRI and rsfMRI to noninvasively assess FC within the face patch system and the embedding of the face patches into larger brain networks. Our results demonstrate that face-processing areas interconnect with each other and with a set of nonface areas involved in cognitive, emotional, and mnemonic functions, forming an extended face-processing network. Importantly, we can also show that this extended network seeded in the face patches exhibits a similar overlap with the DMN as areas involved in high-level social cognition in humans, which allows us to localize a putative TPJ homolog in the macaque STS. This suggests that the face patch system offers an easily accessible, sensory venue into studying the social brain in monkeys, and thus into the evolution of possibly uniquely human social skills.
All animal procedures met the National Institutes of Health Guide for Care and Use of Laboratory Animals, and were approved by the local Institutional Animal Care and Use Committees of The Rockefeller University (protocol number 12585-H) and Weill-Cornell Medical College (protocol number 2010–0029), where MR scanning was performed. Data were acquired in six male, pair-housed macaque monkeys (5 Macaca mulatta, 1 M. fascicularis, 5.4–7.3 kg, age 3–5 yr).
Implantation of MR-compatible headposts (Ultem; General Electric Plastics), MR-compatible ceramic screws (Rogue Research), and acrylic cement (Grip Cement, Caulk; Dentsply International, and/or Palacos, Heraeus Kulzer GmbH) followed standard anesthetic, aseptic, and postoperative treatment protocols [88].
To localized face-selective ROIs, we used a standard face localizer [3]. In short, subjects fixated on a white dot at the center of the screen while we presented images of human and/or monkey faces, human and/or monkey body parts and/or headless bodies, manmade objects, and fruits, intermixed with baseline periods in which only the fixation dot was shown in a block design. Each block lasted 24–30 s. Fluid reward was delivered after variable periods of time (2–4 s), during which the subject maintained fixation within 2 degrees of the fixation dot. Only runs in which the subjects reached at least 90% fixation stability were used for analyses. Visual stimulation and reward were controlled using in house software (Visiko, M. Borisov). Stimuli were projected on a back-projection screen using a video projector (NEC NP3250, refresh rate 60 Hz, resolution 1024 × 768 pixel) with a custom lens. Eye position was measured at 120 Hz using a commercial eye monitoring system (ISCAN).
Data were acquired on a 3 T scanner (Siemens TIM Trio). Functional data were acquired with an AC88 gradient insert (Siemens) and a custom 8-channel phased-array receive surface coil with a horizontally oriented single loop transmit coil (L. Wald, MGH/HST Martinos Center for Biomedical Imaging) while the monkeys were in sphinx position. Before scanning, the contrast agent ferumoxytol (8–10 mg of Fe per kg body weight) was injected into the femoral vein to increase the signal-to-noise ratio (SNR). For the face localizer experiments, we acquired between 16 and 51 runs of functional (T2*-weighted) gradient-echo echoplanar imaging (EPI) data per animal. Each run consisted of 196 volumes of 54 horizontally oriented slices (field of view [FOV] 96 mm, voxel size 1 × 1 × 1 mm, repetition time [TR] = 2 s, echo time [TE] = 16 ms, echo spacing [ESP] = 0.63 ms, bandwidth [BW] = 1,860 Hz/Px, flip angle [FA] = 80°, no gap) acquired in interleaved order with phase partial Fourier 7/8, and two times generalized autocalibrating partially parallel acquisitions (GRAPPA) acceleration, covering the whole brain. Additionally, we obtained field maps that allowed subsequent EPI undistortion [89]. For the resting state scans, we acquired 12 runs of 300 volumes of EPI data per animal, using the same sequences as in the localizer experiments. After induction with ketamine and dexmedetomidine hydrochloride, monkeys were lightly anesthetized with isoflurane (0.5%–0.6%) and placed in an MR-compatible monkey chair. The use of anesthesia follows the original definition of the monkey DMN [25] and conferred several technical advantages, including the elimination of motion artifacts and the ability to record cardiac and respiratory signals. Although anesthesia can affect systemic physiology, neural activity, vasoactive signal transmission and/or vascular reactivity [90], it has been shown that anesthesia preserves the correlation structure that is observed when the subjects are awake [91,92], that significant changes in correlation patterns occur only under much deeper levels of isoflurane anesthesia (>1.5%) [93] than the one we used (0.5%–0.6%), and that anesthetized monkey resting state networks, including the DMN, are strikingly similar to the same networks observed in awake humans [25,94]. Electrocardiogram (sampling rate 400 Hz) and breathing rate (sampling rate 50 Hz) were acquired together with the imaging data. Anatomical images were obtained in a separate session using a T1-weighted magnetization-prepared rapid gradient echo (MPRAGE) sequence (FOV 128 mm, voxel size 0.5 × 0.5 × 0.5 mm, TR = 2.53 s, TE = 3.07 ms, ESP = 7.3 ms, BW = 190 Hz/Px, FA = 7°, 240 slices) and a custom 1-channel receive coil (L. Wald, MGH/HST Martinos Center for Biomedical Imaging) while the monkeys were anesthetized (isoflurane 1.5%–2%) and positioned in an MR-compatible stereotactic frame (Kopf Instruments).
Data were analyzed in Freesurfer (v5.1, https://surfer.nmr.mgh.harvard.edu) and Matlab (R2011b, The Mathworks) using custom code. The first five volumes of each functional run were excluded to prevent T1 saturation effects. Preprocessing included slice scan time correction, motion correction, and geometric distortion correction by means of a field map. Outliers in the time courses were detected semiautomatically based on a threshold of median absolute deviation = 3.5 [95] in the mean whole-brain time course and later excluded from analyses. To create inflated cortical surface reconstructions, the gray–white matter boundary in the skull-stripped anatomical scans was segmented, reconstructed, smoothed, and inflated separately for each hemisphere [96,97].
|
10.1371/journal.ppat.1001092 | Cyclin-Dependent Kinase-Like Function Is Shared by the Beta- and Gamma- Subset of the Conserved Herpesvirus Protein Kinases | The UL97 protein of human cytomegalovirus (HCMV, or HHV-5 (human herpesvirus 5)), is a kinase that phosphorylates the cellular retinoblastoma (Rb) tumor suppressor and lamin A/C proteins that are also substrates of cellular cyclin-dependent kinases (Cdks). A functional complementation assay has further shown that UL97 has authentic Cdk-like activity. The other seven human herpesviruses each encode a kinase with sequence and positional homology to UL97. These UL97-homologous proteins have been termed the conserved herpesvirus protein kinases (CHPKs) to distinguish them from other human herpesvirus-encoded kinases. To determine if the Cdk-like activities of UL97 were shared by all of the CHPKs, we individually expressed epitope-tagged alleles of each protein in human Saos-2 cells to test for Rb phosphorylation, human U-2 OS cells to monitor nuclear lamina disruption and lamin A phosphorylation, or S. cerevisiae cdc28-13 mutant cells to directly assay for Cdk function. We found that the ability to phosphorylate Rb and lamin A, and to disrupt the nuclear lamina, was shared by all CHPKs from the beta- and gamma-herpesvirus families, but not by their alpha-herpesvirus homologs. Similarly, all but one of the beta and gamma CHPKs displayed bona fide Cdk activity in S. cerevisiae, while the alpha proteins did not. Thus, we have identified novel virally-encoded Cdk-like kinases, a nomenclature we abbreviate as v-Cdks. Interestingly, we found that other, non-Cdk-related activities reported for UL97 (dispersion of promyelocytic leukemia protein nuclear bodies (PML-NBs) and disruption of cytoplasmic or nuclear aggresomes) showed weak conservation among the CHPKs that, in general, did not segregate to specific viral families. Therefore, the genomic and evolutionary conservation of these kinases has not been fully maintained at the functional level. Our data indicate that these related kinases, some of which are targets of approved or developmental antiviral drugs, are likely to serve both overlapping and non-overlapping functions during viral infections.
| The eight human herpesviruses are ubiquitous pathogens associated with a wide spectrum of disease, and are grouped into three families (alpha-, beta-, and gamma-) based on sequence homology and tissue tropism. They encode sixteen kinase proteins that are grouped into three recognized families. Eight of these kinases, one from each virus, are found at homologous positions in their respective viral genomes and display limited but discernable amino acid similarity. These proteins have been designated the CHPKs for conserved herpesvirus-encoded protein kinases. We found that the beta and gamma CHPKs, but not the alpha CHPKs, share at least a subset of activities with the cellular cyclin-dependent kinases (Cdks) that are key components of cell cycle regulation. Furthermore, we discovered that the CHPKs, although evolutionarily-conserved, showed a somewhat surprising non-conservation of biological activities not associated with cellular Cdks. Knowledge of functional as well as sequence and positional conservation is essential for understanding the evolutionary relationships among viral, and between viral and cellular, kinases. Importantly, because viral kinases are the targets of approved and emerging antiviral treatments, cataloging both the common and unique activities of these proteins may help predict or explain successes or failures during antiviral drug development.
| Kinases catalyze the transfer of phosphate groups onto targeted substrates. These phosphorylation events are fundamental to proper cellular function and viability and help control enzyme activity, signaling cascades, and protein trafficking, as well as a multitude of other pathways and processes [1]. The enzymatic activity of many kinases can be inhibited both potently and specifically with small molecules, making them suitable drug targets for clinical applications. In fact, kinases are now targets for lung and breast cancer chemotherapies [2], and importantly, for herpesviral infections as well [3], [4].
The human herpesviruses (HHVs) are large, enveloped viruses with double-stranded DNA genomes. Based on sequence homology and cellular tropisms, they are divided into three families, alpha, beta, and gamma. The alpha-herpesviruses are Herpes Simplex Virus type 1 (HSV-1, HHV-1), type 2 (HSV-2, HHV-2), and Varicella Zoster Virus (VZV, HHV-3). These viruses complete productive (lytic) replication cycles in epithelial cells, and establish life-long latency in sensory neurons. They cause recurrent and sometimes painful skin lesions and, in rare cases, meningitis [5], [6]. The beta-herpesviruses are Human Cytomegalovirus (HCMV, HHV-5), and the roseolaviruses Human Herpesviruses 6A and 6B (HHV-6), and Human Herpesvirus 7 (HHV-7). These viruses are classified as lymphotropic because lymphocytes likely serve as a latent reservoir, but they replicate productively in many cell types. Beta-herpesviruses cause severe disease in patients with immature or compromised immune function, and may exacerbate certain chronic ailments in otherwise healthy patients [7], [8]. The gamma-herpesviruses are Epstein-Barr Virus (EBV, HHV-4) and Kaposi's Sarcoma Associated Herpesvirus (KSHV, HHV-8). These viruses are also characterized by a lymphocyte tropism, but can also infect epithelial and endothelial cells. They are causally associated with human cancers [9], [10].
As obligate intracellular parasites, viruses must manipulate cellular processes to facilitate their own replication. Because protein phosphorylation events are so crucial to cellular activity, it is not surprising that they are key targets of regulation during viral infections. While many (perhaps all) families of viruses manipulate the activity of cellular kinases, only two families, the poxviruses [11], [12] and the herpesviruses [13], [14], [15] encode viral proteins with confirmed kinase activity. However, rotavirus, a member of the Reoviridae family of double stranded RNA viruses, encodes a protein called NSP5 that may be a functional kinase [16].
At least eighteen human herpesvirus proteins are reported to possess protein kinase activity. Sixteen of these are grouped into three distinct families, (Us3, UL13 and thymidine kinase) based on amino acid sequence homology (Table 1). The other two human herpesvirus proteins reported to have kinase activity are HHV-2 ICP10 [17] and HHV-5 pp65 [18]. These proteins do not appear to be members of any known viral kinase family. Among the conserved kinase families, only alpha-herpesviruses encode the Us3 family of kinases [14] that, among other functions, prevent apoptosis [19], [20] and disrupt the nuclear lamina [21], [22], [23]. Both the alpha- and gamma-herpesviruses encode thymidine kinase family members that, as their name implies, phosphorylate nucleosides, including thymidine [24]. Importantly, viral thymidine kinases can also phosphorylate unnatural nucleoside analogs (such as ganciclovir and its derivatives) that act as chain terminators for viral DNA replication, and constitute an important subset of anti-herpesviral drugs [24]. The third family of herpesviral kinases was originally termed the UL13 family [25], [26]. However, because these represent the only homologous kinases (at the level of genome position and amino acid sequence) found in every human herpesvirus, they were renamed the CHPKs for conserved herpesvirus-encoded protein kinases [13], [27]. The individual members of the human CHPKs are named UL13 (HHV-1 and -2), ORF47 (HHV-3), BGLF4 (HHV-4), UL97 (HHV-5), U69 (HHV-6 and -7), and ORF36 (HHV-8) (Table 2).
The CHPKs are not absolutely required for viral replication in cell culture, but deletion mutants are severely attenuated for viral growth [28], [29], [30], [31], [32]. They are expressed with early-late kinetics and are incorporated into virions [33], [34], [35], [36], [37]. Demonstrated or postulated functions for the CHPKs during viral replication include tegument disassembly [38], [39], modulation of gene expression [28], [31], [40], stimulation of viral DNA replication [31], [41], [42], [43], and facilitating capsid nuclear egress in part through disruption of the nuclear lamina [31], [44], [45]. Recently, the CHPK encoded by the UL97 gene of HHV-5 (HCMV) was shown to directly phosphorylate the cellular retinoblastoma (Rb) tumor suppressor protein both in vivo and in vitro, on residues that are normally targeted by the cellular cyclin-dependent kinase (Cdk) proteins that control cell cycle progression [46], [47]. UL97 was also found to phosphorylate lamin A/C proteins in vitro on Cdk phosphorylation sites [45], and to rescue the G1-to-S cell cycle defect of Saccharomyces cerevisiae cells lacking Cdk function. Significantly, this yeast complementation assay demonstrated that UL97 can functionally substitute for cellular Cdks [46], indicating that the kinase has Cdk activity, and marking UL97 as the first identified v-Cdk, an abbreviation for the term virally-encoded Cdk-like kinase.
Here we show that the CHPKs encoded by the beta- and gamma-herpesviruses are all capable of inducing Rb phosphorylation in vivo on residues that inactivate the cell cycle inhibitory and tumor suppressor function of this protein. They can also induce lamin A phosphorylation and disrupt the nuclear lamina. Importantly, all the beta- and gamma-herpesvirus CHPKs, with the exception of the HHV-8 (KSHV) ORF36 protein, displayed authentic Cdk function in the yeast complementation assay. The alphaherpesvirus CHPKs were unable to phosphorylate Rb or lamin A, efficiently disrupt the nuclear lamina, or act as Cdks in S. cerevisiae. When we assayed all eight kinases for additional non-Cdk functions against cellular proteins that were previously reported for UL97, we found that despite the evolutionary conservation of these proteins, functional conservation was poor. Altogether, our study identifies a subset of the CHPKs as viral Cdk-like kinases (v-Cdks), but also indicates that the positional homology and amino acid similarity of these protein kinases does not always translate into common substrates or similar biological activities for these proteins.
The CHPKs are grouped as a kinase family based on their conserved genome location and limited sequence homology. While some common functions for select members of the CHPK family have been identified, a thorough functional comparison of this group of kinases has not been reported. Upon the revelation that the HHV-5 (HCMV) CHPK, the UL97 protein, was a viral Cdk ortholog [46], we initiated experiments to determine if the other CHPKs were also v-Cdks. We obtained an expression plasmid for the HHV-2 CHPK which, upon transfection into mammalian cells, produces an active kinase that is tagged with the hemagglutinin (HA) epitope [48]. We then generated individual plasmids that express HA epitope-tagged derivatives of the other seven CHPKs (Table 2). Previous reports indicate that epitope-tagging does not eliminate the kinase activity of the seven CHPKs for which such an analysis has been conducted [48], [49], [50], [51], [52], [53], [54], [55]. The activity of an epitope-tagged CHPK of HHV-7 has not been previously examined. Verification of the correct CHPK sequence by direct analysis (data not shown) and our finding that each tagged kinase scored positive in at least one activity assay (Table 3) provide confidence that our expression plasmids produced active kinases. Although we cannot rule out changes in substrate specificity or specific activity due to the epitope tag, we do note that tagged CHPKs can complement the growth defects of HHV-4 and HHV-5 mutant viruses lacking the untagged (wild type) CHPK gene [52], [56]. Importantly, the epitope tag allows us to monitor the expression and localization of these eight different kinases with the same antibody. Individual transfection of these eight plasmids into human U-2 OS cells allowed for the production of the viral proteins which, when detected on Western blots with an HA antibody, migrated near the predicted molecular weight (Table 2) of the full-length protein [34], [35], [57], [58], [59], [60], [61]. Adjusting the amount of transfected plasmid DNA allowed us to identify experimental conditions under which the steady state protein levels achieved for six of these eight different proteins was consistent and comparable. However, the CHPKs encoded by HHV-2 and HHV-5 often accumulated to lower levels. Therefore, Western blot expression controls are presented for each individual experiment.
The sub-cellular localization of the transfected CHPKs in U-2 OS cells was examined by fluorescence microscopy (Fig. 1A) and quantitated (Fig. 1B) by visually determining the percentage of cells with primarily nuclear localization of the kinase, primarily cytoplasmic localization, or localization to both cellular compartments. All of the kinases were found in both the nucleus and cytoplasm in at least 25% of the transfected cells. The alpha-herpesvirus CHPKs (HHV-1, -2, and -3) generally showed both strong cytoplasmic and faint nuclear staining. Certain members of the beta- and gamma-herpesvirus CHPKs showed predominantly nuclear staining (HHV-4, -5, and -6), while others (HHV-7 and -8) were more likely to be found in both cellular compartments. These observed sub-cellular localizations are in agreement with previously published data [34], [35], [59], [60], [62], [63].
The retinoblastoma (Rb) and lamin A/C proteins are phosphorylated by the cyclin-dependent kinases (Cdks) that control cell cycle progression. Rb is a tumor suppressor responsible for regulating the G1/S cell cycle checkpoint [64]. Phosphorylation of Rb on specific Cdk-consensus sites results in the dissociation of protein complexes between Rb, histone deacetylases (HDACs), and E2F transcription factors and allows for the expression of E2F-responsive genes that drive progression through G1 and entry into the S-phase of the cell cycle [64]. Certain E2F-responsive genes also play critical roles in nucleotide biosynthesis and DNA replication. Many DNA viruses partially rely on cellular machinery for the replication of their genomes, and therefore target Rb for inactivation during infection [65], [66]. For example, during HHV-5 (HCMV) infection, the UL97 CHPK is necessary and sufficient for Rb inactivation through phosphorylation on Cdk consensus sites [46]. Rb inactivation appears to be a critical function of UL97 for efficient viral replication [67].
In vivo phosphorylation of Rb by an ectopically-expressed kinase can be easily and reliably determined only in human Saos-2 cells. These Rb-null osteosarcoma cells fail to phosphorylate ectopically-expressed Rb unless a cyclin protein (cellular or viral) or UL97 is included in the transfection [46], [68], [69]. Rb phosphorylation in Saos-2 cells can be observed on Western blots by an electrophoretic mobility shift of the protein to higher molecular weights, as well as with phospho-specific antibodies that detect Rb proteins only when phosphorylated on certain Cdk consensus sites. We found that the CHPKs encoded by all of the beta- and gamma-herpesviruses (HHV-4, -5, -6, -7, and -8) were able to phosphorylate Rb when co-transfected into Saos-2 cells (Fig. 2A). Phosphorylation was detected by both the shift in molecular weight and with three independent phospho-specific antibodies that detect Rb only when phosphorylated on specific Cdk consensus sites that regulate association with the E2F proteins [70], [71]. Interestingly, the beta-herpesvirus CHPKs induced the phosphorylation of Rb-inactivating residues to a substantially higher degree than the gamma-herpesvirus proteins (Fig. 2A). The alpha-herpesvirus proteins (HHV-1, -2, and -3) were unable to phosphorylate Rb (Fig. 2A), as judged by both electrophoretic mobility and the lack of reactivity with the phospho-specific antibodies. The ability of the beta- and gamma-herpesvirus CHPKs to phosphorylate Rb required that each protein was an active kinase. Substitution of the catalytic lysine to create kinase-inactive mutants inhibited the ability of the CHPKs to phosphorylate Rb, shown by both molecular weight shift as well as with the phospho-Rb-specific antibodies (Fig. 2B).
There was no correlation between steady state expression level and the ability to phosphorylate Rb (Fig. 2A). CHPKs that accumulated to high levels either did (e.g. HHV-4 and HHV-7) or did not (e.g. HHV-3) phosphorylate Rb. Likewise, CHPKs that accumulated only to low levels either did (e.g. HHV-5) or did not (e.g. HHV-2) phosphorylate Rb. To further explore any potential relationship between expression level and Rb phosphorylation, we decreased the amount of HHV-4 and HHV-7 CHPK expression plasmid transfected in an attempt to equalize, as much as possible, the steady-state expression levels of these Rb-phosphorylating kinases to the non-Rb-phosphorylating HHV-2 kinase. We found that lower steady state levels of the HHV-4 and HHV-7 CHPK still led to Rb phosphorylation (Fig. 2C), as seen by the shift in apparent molecular weight. While substantial kinase overexpression may result in the phosphorylation of non-physiologic substrates (false-positives), and low expression levels may prevent detection of the phosphorylation of true substrates (false-negatives), the data we present here are consistent with the conclusion that the beta- and gamma-herpesvirus CHPKs, but not the alpha-herpesvirus CHPKs, result in Rb phosphorylation when ectopically expressed.
Like Rb, lamin A/C is also a Cdk substrate phosphorylated by the CHPKs of HHV-4 and HHV-5 on Cdk consensus sites [45], [72]. Lamins are intermediate filament proteins that line the inner nuclear membrane as part of the nuclear lamina [73]. During mitosis, the nuclear lamina is disassembled after phosphorylation of lamin A/C by Cdk1 [73]. During viral infections, the nuclear lamina likely represents a physical barrier to herpesviral capsids that must leave the nucleus through envelopment at the inner nuclear membrane. Thus, it has been proposed that in order to gain access to the inner nuclear membrane, herpesviruses must disrupt the nuclear lamina [74]. In cells infected with HHV-1, HHV-4, and HHV-5, lamina disruption is achieved through phosphorylation of lamin A/C on Cdk consensus sites [23], [45], [72].
We found that beta- and gamma-herpesvirus CHPKs (HHV-4, -5, -6, -7, and -8), but not those of the alpha-herpesviruses (HHV-1, -2, and -3), were able to efficiently disrupt the nuclear lamina in transfected U-2 OS cells (Fig. 3A). The nuclear lamina was visualized by fluorescence microscopy in cells co-transfected with expression plasmids for the CHPKs and human lamin A fused to GFP. Western blots of the ectopically-expressed proteins are shown in Figure S1. Non-confocal fluorescent images of the transfected cells show a dim GFP signal throughout the body of the nucleus that is brighter where lamin A/C concentrates at the nuclear periphery (see the panel with empty vector co-transfected cells, Fig. 3A, bottom right). Co-expression (visualized by indirect immunofluorescence with an HA antibody) of the beta- and gamma-herpesvirus CHPKs, and to a lesser extent the HHV-2 CHPK, caused a redistribution of GFP from the ring-like signal around the nuclear perimeter into large punctate spots found throughout the nucleus (Fig. 3A) in the majority of cells imaged. For these CHPKs, the percentage of kinase-positive cells showing disrupted nuclear lamina was statistically different than cells transfected with GFP-lamin A and an empty vector control (Fig. 3B). The HHV-1 and HHV-3 alphaherpesvirus CHPKs did not visibly disrupt the nuclear lamina at a frequency distinguishable from empty vector transfected cells.
We also examined the ability of ectopically-expressed CHPKs to direct the phosphorylation of the endogenous lamin A protein at serine-22. Phosphorylation of this and other lamin A residues by cellular Cdk1 initiates the process of lamina breakdown during mitosis [75]. Matching our results from the GFP-lamin experiments (Fig. 3A and 3B), the alpha-herpesvirus CHPKs were unable to induce serine-22 phosphorylation (Fig. 3C and 3F), whereas the beta- (Figure 3D and 3F) and gamma-herpesvirus CHPKs (Fig. 3E and 3F) did induce serine-22 phosphorylation. Kinase-inactive mutants of the beta- and gamma-herpesvirus CHPKs were unable to induce serine-22 phosphorylation. Western blots of the ectopically-expressed CHPKs are shown in Figure S2, and quantitation of the data is shown in Figure 3F.
The HHV-2 CHPK showed low level lamina disruption that was statistically different from an empty vector control (Fig. 3B). A similar level of serine-22 phosphorylation was observed, although the difference from the empty vector control did not reach statistical significance. Thus, although the HHV-2 CHPK may retain some ability to phosphorylate lamin A, (perhaps more efficiently in the tail region) and to disrupt the nuclear lamina [76], we conclude that these activities are, in general, absent from the alpha-herpesvirus CHPKs, but present in the beta- and gamma-herpesvirus CHPKs. Although experiments of this type are often cited as evidence that CHPK-mediated lamin A/C phosphorylation directly leads to lamina disruption, we note that they cannot rule out other mechanisms for lamina disruption upon CHPK expression or herpesvirus infection (such as lamin B or lamin receptor phosphorylation) in addition to or instead of lamin A/C phosphorylation.
The first human Cdk was cloned by genetic complementation of Cdk mutant yeast cells [77]. Unlike higher eukaryotes, S. cerevisiae encodes only a single Cdk, the CDC28 gene. Yeast cells harboring a temperature-sensitive mutant allele of this gene (cdc28-13) exhibit growth arrest in G1 phase upon a shift to the restrictive temperature [78]. Expression of a functional human Cdk in cdc28-13 cells rescues this arrest phenotype [79]. Expression of the HHV-5 (HCMV) CHPK, the UL97 protein, in cdc28-13 yeast cells also rescued the cell cycle arrest phenotype at the restrictive temperature, indicating that it has authentic Cdk activity, and identifying UL97 as the first known viral functional ortholog of a Cdk [46].
We used the same assay to determine if the other CHPKs were also genuine Cdks. Asynchronously growing cdc28-13 yeast cells harboring plasmids expressing the CHPKs under the control of the GAL1 promoter were shifted to the restrictive temperature to inactivate the sole yeast Cdk, and grown in the presence of galactose to induce the expression of the CHPK. Western blots of the ectopically-expressed CHPKs are shown in Figure S3. Five hours later, cell cycle progression through G1 was verified by determining the number of budded cells. In the absence of an ectopic Cdk, cdc28-13 cells grown at the restrictive temperature arrested in the G1 phase of the cell cycle as unbudded cells [46], [78], (Fig. 4). However, as was previously observed [46], ectopic expression of human Cdk1, and to a lesser extent the HHV-5 CHPK permitted cell cycle progression through G1 and into S phase of cdc28-13 cells grown at the restrictive temperature, as evidenced by an increase in the number of budded cells (Fig. 4). This result is indicative of authentic Cdk function. Expression of the other beta-herpesvirus CHPKs (HHV-6 and -7) and the CHPK from the HHV-4 gamma-herpesvirus also rescued the cell cycle defect of cdc28-13 yeast grown at the restrictive temperature (Fig. 4), indicating that they also have genuine Cdk activity. The HHV-8 gamma-herpesvirus CHPK, the ORF36 protein of KSHV, failed to produce a statistically significant increase in cdc28-13 budded cells at the restrictive temperature (Fig. 4), although it did display substantial Cdk activity in one of the three experiments performed. Interestingly, KSHV is the only human herpesvirus to encode a viral cyclin [80]. The KSHV cyclin pairs with and activates human Cdks, perhaps obviating the need for Cdk function of the viral CHPK. As expected from their inability to phosphorylate Rb or lamin A/C, the alpha-herpesvirus CHPKs did not display Cdk activity in this assay. Note that we cannot rule out the possibility that the alpha-herpesvirus proteins might not be active when expressed in yeast cells. In total, the Rb, lamin, and yeast experiments functionally define the beta- and gamma-herpesvirus CHPKs as viral cyclin-dependent-like kinases, or v-Cdks.
While the CHPKs are grouped as a family based upon their positional homology and limited amino acid similarity, our data indicate that only the beta- and gamma-herpesvirus CHPKs are v-Cdks. These observations indicate that the evolutionary relationship of these eight kinases does not translate to a complete conservation of Cdk-like function. To determine if other functions of these kinases may have been more widely conserved throughout evolution, we screened each kinase for their ability to disrupt PML-NBs, nuclear aggresomes, and cytoplasmic aggresomes, all Cdk-unrelated activities that have been previously attributed to UL97, the HHV-5 (HCMV) CHPK [47].
Promyelocytic Leukemia Nuclear Bodies (PML-NBs) are sub-nuclear structures that appear as multiple punctate spots within the nucleus and are involved in a wide array of cellular activities [81]. During infections with HHV-1, new PML-NBs are generated at the sites where infecting viral genomes enter the nucleus [82]. HHV-5 genomes also co-localize with PML-NBs shortly after infection [83]. Proteins that localize to PML-NBs institute an intrinsic cellular defense against these herpesviruses by silencing viral immediate early (IE) gene expression [84], [85]. Viral proteins delivered to cells upon virion entry and/or expressed as IE genes inactivate this defense and disrupt PML-NBs [86], [87], [88].
We used U-2 OS cells and indirect immunofluorescent detection of the PML protein (Fig. 5A) to monitor the effects of ectopically-expressed CHPKs on PML-NBs. Western blots of the ectopically-expressed CHPKs are shown in Figure S4. As expected, we found that the number of PML-NBs in U-2 OS cells resembles a Gaussian distribution between 4 and 24 (Fig. 5B), with an average of 11 per cell (Table 3). Our average agrees well with previously reported data [63]. We used HCMV IE1 as a positive control for PML-NB disruption [86], and counted the number of PML-NBs in 100 CHPK-positive nuclei in three separate experiments. With the exception of HHV-1 and HHV-3, expression of all of the other CHPKs resulted in a perceptible shift in the distribution of PML-NBs per cell towards lower numbers (Fig. 5B). Most of the shifts were small, although the HHV-5 and HHV-8 CHPKs showed more substantial effects. However, only the HHV-4 CHPK caused a drop in the average number of PML-NBs per cell that was statistically different than control cells (Table 3).
High synthesis rates, cellular stresses, and other stimuli can lead to protein aggregation and the formation of aggresomes. Ectopic expression of the HHV-5 (HCMV) CHPK (UL97) in uninfected cells has been shown to disrupt aggregates of viral proteins as well as nuclear and cytoplasmic aggresomes formed by GFP-GCP170*, an artificial stimulator of aggresome formation [47], [53]. Conversely, GFP-GCP170* aggregates have been detected in HHV-1 infected cells [89], implying that the HHV-1 CHPK may not disrupt aggresomes.
To determine if aggresome disruption was a general feature of the CHPKs, we tested the ability of the kinases to disrupt nuclear or cytoplasmic aggresomes formed by the ataxin-1 protein. Ataxin-1 (Atx1) has a poly-glutamine tract that, when expanded beyond 40 amino acids, produces a protein with a non-native conformation that forms aggregates, is cytotoxic, and causes spinocerebellar ataxia type 1 [90]. We co-transfected cells with the CHPKs and FLAG-tagged derivatives of either Atx1(Q82) or Atx1(Q82)-K772T that form aggregates in the nucleus and cytoplasm, respectively [90]. Western blots of the ectopically-expressed proteins are shown in Figure S5. The presence of at least one aggregate or the total absence of aggregates in ataxin-expressing cells (Fig. 6A and 6C) was determined for at least 200 CHPK-positive cells in at least three independent experiments. We found that the HHV-2 and HHV-5 CHPKs resulted in a statistically significant decrease in the number of cells in which ataxin-1 could form both nuclear (Fig. 6A and 6B) and cytoplasmic (Fig. 6C and 6D) aggregates even though these were the CHPKs expressed to the lowest level (Figure S5). CHPKs encoded by HHV-1, -3, -4, and -6 caused a statistically significant decrease in cells with either nuclear or cytoplasmic aggresomes, but not both (Fig. 6). While the HHV-4 CHPK caused a substantial decrease in the percentage of cells containing nuclear aggresomes, the wide variability in the numbers accumulated over three independent experiments resulted in data that did not achieve statistical significance (Fig. 6B). Our data confirm previous results demonstrating that the HHV-5 CHPK can either prevent or disrupt protein aggregation [47], but also indicate that this ability is not well conserved among the eight CHPKs.
In summary, we found that the most conserved activity of the CHPKs is their Cdk-like function. However, this is only found in the beta-and gamma-herpesvirus members, and is absent in the alpha-herpesvirus proteins. The non-Cdk-like activities reported for the HHV-5 CHPK, the HCMV UL97 protein, are poorly conserved among the other family members. Thus the functional conservation between these eight kinases as a whole, much like their amino acid sequence similarity, is limited.
Cellular substrates of selected CHPKs that are normally phosphorylated by the Cdks in uninfected cells have been identified. These include Rb [46], [47], lamin A/C [45], [72], translation elongation factor 1 delta [49], the carboxy-terminal domain of RNA Polymerase II [91], the helicase complex component MCM4 [92], the condensin protein involved in chromatin packaging [93], and the Cdk inhibitor p27 [51]. However, only the CHPK encoded by HHV-5, the HCMV UL97 protein, had been shown to be a functional Cdk ortholog [46]. In this study, we examined the remainder of the CHPKs for Cdk-like activity, as well as non-Cdk-like activities previously reported for UL97. We employed the yeast complementation assay because it is the most basic and stringent test for Cdk activity, and the Saos-2 assay because it is the most widely-accepted assay for demonstrating Rb phosphorylation in vivo by a co-transfected protein. For all other experiments, we used U-2 OS cells because they transfect well, do not express a transforming viral oncogene, and represent a generic human cell (the differing cellular tropisms of the eight human herpesviruses prevented us from using a more physiologically relevant cell type). Effects on cellular targets or processes (as opposed to viral ones) were analyzed so that activity against identical (as opposed to just homologous) substrates could be monitored.
Our results indicate that Cdk-like activity is absent in the alpha-herpesvirus CHPKs, as they were unable to induce Rb or lamin A phosphorylation, lamina disassembly, or yeast cdc28-13 cell cycle progression. These results agree well with the observations that Rb is not phosphorylated in cells infected with HHV-1, HHV-2, or HHV-3 [66], but is phosphorylated in cells infected with HHV-4 and HHV-5, [94], [95], representative members of the beta- and gamma-herpesvirus families. Our lamin results are also mostly consistent with previously published data. Although it is clear that lamin A/C becomes phosphorylated in cells infected with representative members of the alpha- [23], beta- [45], and gamma- [72] herpesviruses, different kinase families appear to be responsible for these phosphorylation events. A recent analysis [72] found that, in addition to the HHV-4, HHV-5 [96], and HHV-8 CHPKs, the HHV-1 CHPK was able to disrupt the nuclear lamina in HeLa cells [72], however, we detected no activity of the HHV-1 protein in two independent lamin assays (Fig. 3). It is unclear if expression of the papillomavirus E7 oncoprotein in HeLa cells, a known stimulator of Cdk activity [97] that is not present in the U-2 OS cells used here, could explain this difference between the two studies. Thus, although previous studies [72], [76] and our own data with the HHV-2 CHPK (Fig. 3) indicate that the alpha-herpesvirus CHPKs may retain some ability to phosphorylate lamin A/C and to disrupt the nuclear lamina, we suspect, as others have shown, that the members of the alpha-herpesvirus-specific Us3 family of kinases are largely responsible for this activity in virus-infected cells [21], [22], [23], [98]. Thus, we conclude that the alpha-herpesvirus CHPKs do not mimic cellular Cdk activity. Interestingly, cellular Cdks are active during, and required for efficient alpha-herpesvirus replication [66]. The critical substrates for cellular Cdks in alpha-herpesvirus infected cells remain to be determined.
In contrast, the beta- and gamma-herpesvirus CHPKs were found to possess Cdk-like activity. All of these proteins phosphorylated Rb and disrupted the nuclear lamina, and all but one rescued cell cycle progression in the yeast complementation assay. It is presently undetermined what (if any) advantage encoding and expressing their own Cdk-like protein affords these viruses that utilization of cellular Cdk activity would not. Nevertheless, this analysis, along with our previous study [46] identifies the HHV-4 (EBV-BGLF4), HHV-5 (HCMV-UL97), HHV-6 (U69), HHV-7 (U69), and HHV-8 (KSHV-ORF36) CHPKs as virally-encoded cyclin-dependent-like kinases, or v-Cdks.
Unlike cellular Cdks, the HHV-5 CHPK (HCMV UL97), the first identified v-Cdk, lacks many of the regulatory features that restrict kinase activity during certain stages of the cell cycle or under physiological stresses [46]. For example, UL97 lacks conserved sequences for cyclin binding and appears to have cyclin-independent kinase activity. UL97 has an amino acid substitution at the Cdk site (Thr 160 of Cdk2) of phosphorylation by Cdk activating kinase (CAK), and is active under conditions of CAK inhibition. UL97 lacks amino acid residues important for binding both the Cip/Kip and INK classes of cyclin-dependent kinase inhibitors, and is immune to inhibition by p21WAF1/CIP1 in vivo and in vitro. Finally, UL97 has a phenylalanine substitution for a tyrosine residue (Tyr 15 of Cdk2) found in Cdks which, when phosphorylated in G2 phase or in response to DNA damage, attenuates kinase activity.
A sequence alignment of the other v-Cdks showed that, like UL97, none of these kinases contain the cyclin-binding, cyclin-dependent kinase inhibitor-binding, or CAK phosphorylation motifs found in the cellular Cdks (Fig. 7). Thus, the newly-identified v-Cdks lack many of the amino acids and protein domains responsible for regulating the activity of their cellular functional orthologs. Interestingly, all of the v-Cdks except UL97 and KSHV ORF36 (the HHV-8 CHPK) do contain the tyrosine residue that, in cellular Cdks, allows for the attenuation of kinase activity upon phosphorylation. Whether or not this amino acid substitution affects kinase activity during viral infection remains to be examined.
In the past, designation as a Cdk required not only sequence and functional homology to known Cdks, but also experimental evidence that the activity of the kinase was dependent upon association with a cyclin regulatory subunit [99]. However, a recent nomenclature adjustment has grouped all human and murine kinases with sequence and functional similarity to Cdks into the Cdk family, even if cyclin binding has not been observed or is not expected [100]. Thus, while the beta- and gamma-herpesvirus CHPKs do not appear to require cyclin binding for activity, describing these proteins as v-Cdks is in accordance with the accepted classification system because they are related to true Cdks by amino acid sequence (Fig. 7), and because they mimic Cdk activity (Fig. 2, Fig. 3, Fig. 4, Table 3) even though they are likely not regulated in an identical fashion to cellular Cdks.
The lack of conservation of Cdk activity in the alpha-herpesvirus CHPKs was surprising based on the strong evolutionary relationship of this family of proteins, and prompted us to ask if non-Cdk-like functions were more conserved throughout the CHPK family. Interestingly, we found that non-Cdk-like functions were even less well conserved than Cdk activity (Table 3). Only the HHV-4 CHPK mediated a decrease in PML-NB numbers per cell that reached statistical significance, although other kinases, most notably the HHV-5 and HHV-8 CHPKs, clearly appeared to affect these structures. This finding varies considerably from previous reports, perhaps because of different experimental approaches. Our study used a version of the HHV-4 CHPK tagged at the amino-terminus with the 9 amino acid HA epitope tag, while a previous examination of this protein that failed to demonstrate PML-NB disruption [63] used a 69 amino acid SPA tag at the carboxy-terminus. The differences in tag size and location may affect kinase activity, which was not directly examined in the other study [63], but has been demonstrated here (Fig. 2). A study that concluded the HHV-5 CHPK disrupted PML-NBs [47] was conducted in Cos7 cells that express the simian virus 40 (SV40) tumor (T) antigen, and monitored Sp100 localization as a measure of PML-NBs. Our study visualized PML, the sole protein required for PML-NB formation, in U-2 OS cells that do not express a transforming viral oncogene. It is currently unknown which if any of these differences may explain why the subtle effects on PML-NB distribution that we observed upon expression of the HHV-5 CHPK failed to reach statistical significance.
Proteins that efficiently disrupt PML-NBs (HCMV IE1 and EBV Zta) are expressed within infected cells in vitro prior to the time when the CHPKs are expressed [101], [102], so the ability of the HHV-4 and HHV-5 CHPKs to decrease the numbers of PML-NBs when ectopically expressed may have limited relevance during viral infection. Furthermore, because the proteins that localize to PML-NBs and not the PML-NB structures themselves appear to be the mediators of an intrinsic antiviral defense [84], [85], the significance of the ability of these viral proteins to reduce the number of, but not eliminate PML-NBs from cells is unclear, especially since it is unknown whether or not any PML-NB resident proteins are direct substrates of these kinases. Therefore, perhaps unsurprisingly, the ability to disrupt PML-NBs appears to be an activity not well conserved throughout the CHPK family.
Disruption of protein aggregates termed aggresomes is another non-Cdk function attributed to the HHV-5 CHPK. Aggresomes have been hypothesized to represent a cellular anti-viral defense that identifies, initially sequesters, and ultimately disposes of viral proteins to inhibit viral replication [103]. Thus, aggresome disruption could potentially enhance viral infections. However, aggresome formation is also thought to sequester misfolded proteins and perhaps facilitate their degradation by the proteasome or through autophagy [103]. In addition, viruses may commandeer aggresome formation pathways to help form replication or assembly compartments [103]. In these respects, aggresome formation would appear to enhance virus replication.
Different human herpesviruses appear to present examples where aggresome formation has opposing effects on viral replication. Structures morphologically resembling aggresomes form in both the nuclei and cytoplasm of cells infected with HHV-1 and -2 (the herpes simplex viruses) or HHV-5 (HCMV). At early times after HHV-1 infection, aggregates containing viral proteins, cellular chaperones, and proteasomes form in the nucleus [104]. These virus-induced chaperone enriched (VICE) domains are proposed to be sites where misfolded proteins are sequestered away from virus replication compartments, perhaps to facilitate either their proper folding or degradation, as well as to prevent the misfolded proteins from triggering cellular processes deleterious to virus infection such as apoptosis or the unfolded protein response [105]. At late times after HHV-2 infection, similar structures form in the cytoplasm [106]. Disruption of these cytoplasmic aggresome-like structures correlates with decreased HHV-2 yields [106], indicating that the integrity of these structures may positively influence viral infection. In contrast, the disruption of aggresomes may facilitate HHV-5 replication. Cells infected with an HHV-5 mutant lacking its CHPK contain nuclear and cytoplasmic aggresome-like structures that are absent during wild type-infection [53], and produce fewer infectious progeny virions than wild-type infection [32]. These findings have led to the hypothesis that aggresomes may represent a cellular anti-viral defense, and that one role for the HHV-5 CHPK during viral infection is to prevent their formation [47].
Only one CHPK from each family (alpha, HHV-2; beta, HHV-5; and gamma, HHV-4) disrupted cytoplasmic aggresomes. However, disruption of nuclear aggresomes was the one activity that could conceivably be considered as conserved among these kinases, even though the magnitude of this disruption was minimal in some cases. All of the three alpha-herpesvirus CHPKs scored positive for this assay, as did two (HHV-5 and HHV-6) of the three beta-herpesvirus CHPKs. While our statistical analysis argues that both gamma-herpesvirus proteins are devoid of this activity, the HHV-4 CHPK clearly appeared to have an effect on these structures.
Because, for the most part, the non-Cdk-like activities we examined do not segregate to specific viral families (as Cdk activity does), we suggest that cellular proteins that are substrates for individual CHPKs, but not the Cdks, may represent a more divergent set of proteins than might be expected based upon the evolutionary relationship of these kinases. Furthermore, our data suggest that the v-Cdks may have an expanded repertoire of substrates as compared to the alpha-herpesvirus CHPKs. This may not be surprising because the alpha-herpesvirus-specific Us3-family kinases appear to assume at least one v-Cdk role, lamina disruption [21], [22], [23], during viral infection. Whether or not a comprehensive analysis of kinase activity against homologous viral proteins would reveal more conservation of function throughout the CHPK family awaits further examination.
An alternative strategy to assay for functional conservation among the CHPKs would be to determine if other members of the kinase family could complement the growth defect observed in viruses lacking their endogenous CHPK. The few experiments that have been performed support the contention that the v-Cdks have a partially overlapping but also extended substrate range when compared to the alpha-herpesvirus CHPKs. Expression of the HHV-5 CHPK from an HHV-1 genome in which its CHPK was deleted resulted in a complete restoration of progeny virion formation at both high and low multiplicities of infection [107], indicating that this v-Cdk can perform all the necessary functions for viral replication normally carried out by the HHV-1 CHPK. However, when expressed from a recombinant adenovirus, the HHV-1 CHPK was unable to rescue the growth of an HHV-5 CHPK null-mutant virus [108]. Interestingly, that same series of experiments demonstrated that the HHV-4 CHPK could partially complement the growth defect of the HHV-5 CHPK mutant virus, indicating that there is substantially more functional overlap within the v-Cdks than between the v-Cdks and the alpha-herpesvirus CHPKs. A comprehensive genetic analysis of inter-virus CHPK complementation should increase our understanding of the evolutionary path that these viral kinases have traveled, as well as helping to define their critical roles during viral infection.
In all eight human herpesviruses, the CHPK gene is found in a conserved genomic block and is flanked on either side by homologous genes [109], [110], likely indicating that the current CHPK genes are descended from a single primordial precursor. Over time, the functions of this putative CHPK precursor appear to have diversified differently among these eight viruses. Thus while they represent a single historical kinase family, it may be more practical to consider the CHPK family as representing two individual clades, the UL13s and the v-Cdks. An evolutionary tree diagram of the sixteen shared human herpesvirus kinases (Fig. 8) clearly shows their segregation into different families and clades. Whether the putative primordial CHPK was a Cdk whose Cdk-like activity was lost in the alpha-herpesvirus lineage, or whether it was a kinase that evolved to acquire Cdk-like function in the beta- and gamma-herpesviruses is an interesting topic for speculation and debate.
U-2 OS human osteosarcoma cells were grown in a 5% CO2 atmosphere at 37°C in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% (vol/vol) fetal bovine serum (Gemini), 100 U/ml penicillin, 100 ug/ml streptomycin, and 0.292 mg/ml glutamine (Gibco). Cells (2×106) were seeded on 10 cm plates five hours prior to transfection by the calcium phosphate co-precipitation method. After an overnight incubation, the cells were washed twice with medium and then re-fed with serum-containing media (time zero). Different amounts of CHPK expression plasmid DNA were transfected (HHV-2, -5, -7, 15 µg; HHV-1 and -8, 10 µg; HHV-3, 5 µg; HHV-4, 2µg; HHV-6, 1µg) in order to approximately equalize steady state protein levels. Total DNA levels in transfections were balanced with pGEM7 (Promega). Saos-2 cells were grown as above, seeded at a density of 8×105 cells per 60 mm dish, and transfected 5 hours later with TransIT-2020 (Mirus) according to the manufacturer's instructions. Along with 1µg of Rb expression plasmid, different amounts of CHPK expression plasmid DNA were transfected (HHV-2, -5, -5KD, 1.5 µg; HHV-1, -7, -7KD, -8KD, 1 µg; HHV-8, 0.15 µg; HHV-3, -4KD, -6, -6KD, 0.1 µg; HHV-4, 0.05 µg). Transfection conditions for the experiment in Figure 2C were the same except 0.01 µg was transfected for HHV-4, and 0.2 µg for HHV-7.
Kinase alleles (except HHV-2) were amplified by PCR, sequence verified (see Table 2), and subsequently cloned into the pCGN vector that adds an N-terminal hemagglutinin (HA) epitope tag and expresses genes from the HCMV major immediate early promoter. HA-tagged alleles for all of the CHPKs except HHV-5 were also cloned into the yeast expression vector pMSS78 under the control of the yeast GAL1 promoter [111]. The plasmid that expresses V5 epitope-tagged HHV-5 CHPK in yeast has been previously described [46]. Kinase-deficient (KD) mutants (Table 2) were created by standard mutagenesis approaches and confirmed by complete sequencing of the mutant allele. PCR templates were as follows: HHV-1, viral genomic DNA strain KOS, a gift from Curtis Brandt; HHV-3, plasmid pCAGGS ORF47.12 [50], a gift from Charles Grose; HHV-4, plasmid pcDNA3.1-FLAG BGLF4 [112], a gift from Thomas Stamminger; HHV-5, ppUL97-V5 [53], a gift from Mark Prichard; HHV-6 and HHV-7, cosmids pMF147-19 [113] and I6 (unpublished) respectively, gifts from Steven Dewhurst; HHV-8, plasmid pND ORF36 [55], a gift from Paul Luciw. Plasmid pSG5-UL13 3′ HA [48] kindly provided by Lynda Morrison, was used to express a kinase-active C-terminally tagged HHV-2 CHPK from the SV40 promoter.
The sequence comparison was created with a ClustalW alignment in MEGA 4 [114], and utilized only the kinase domains of the indicated proteins. The parameters were as follows: Pairwise alignment - gap opening penalty 2, gap extension penalty 0.05; multiple alignment - gap opening penalty 5, gap extension penalty 1, utilizing the Blosum matrix. The alignment was heuristically adjusted to align kinase domain III. The phylogenic tree was constructed by the neighbor-joining method, in MEGA 4 [114].
Equal amounts of protein (determined by Bradford assay) from cell lysates prepared in radioimmunoprecipitation assay (RIPA) buffer with protease inhibitors were analyzed by Western blotting (WB) as previously described [115]. Cells grown on glass coverslips were processed for indirect immunofluorescence (IF) as previously described [116]. Images were produced with a Nikon Eclipse TE2000-S or Zeiss Axiovert 200M microscope. The following antibodies were from commercial sources: HA (IF: Roche 3F10; WB: Covance MMS-101P), PML (Santa Cruz sc-966), Total Rb (Cell Signaling 9309), Rb p-Ser807/811 (Cell Signaling 9308), Rb p-Ser780 (Cell Signaling 9307), Rb p-Thr821 (Biosource 44-582G), Lamin A/C (Novocastra NCL-Lam-A/C) Lamin A p-Ser22 (Cell Signaling 2026), Flag (Sigma F1804), V5 (Invitrogen R960-25) Tubulin (DM 1A Sigma), PGK (Invitrogen 459250). The HCMV IE1 antibody 1B12 has been previously described [117].
Rb Phosphorylation in Saos-2 cells and Cdk activity in S. cerevisiae cdc28-13 cells were determined as previously described [46]. For the nuclear lamina disruption assays, U-2 OS cells grown on coverslips were co-transfected with expression plasmids for the indicated CHPK and pEGFPhLA-WT [72] that expresses a GFP-lamin A fusion protein (a kind gift of David Gilbert). Coverslips were harvested 24h later, stained for the CHPK, and then visualized by fluorescence microcopy. At least 300 CHPK- and GFP-lamin A-positive cells for each co-transfection were analyzed in three independent experiments. For the lamin A/C serine-22 phosphorylation assay, U-2 OS cells grown on coverslips were transfected with expression plasmids for the indicated CHPK. Cells were then incubated in media containing 0.1% FBS for 36 hours before harvesting the coverslips, staining for the CHPK and lamin A phosphorylated at serine-22, and visualization by fluorescence microscopy. At least 200 (wild type) or 100 (kinase-deficient) CHPK-positive cells were analyzed in three separate experiments. For PML-NB disruption assays, U-2 OS cells grown on coverslips were transfected with an expression plasmid for the indicated CHPK or pCGN-IE1 [117]. Coverslips were harvested 24h later, stained for PML and the CHPK, and then visualized by fluorescence microscopy. The number of PML-NBs in 100 individual CHPK- or IE1-positive cells was counted in each of three independent experiments. For the aggresome disruption assays, U-2 OS cells grown on coverslips were co-transfected with expression plasmids for the indicated CHPK and either p3PK-Flag-Ataxin-1 or p3PK-Flag-Ataxin-1 K722T [90] that express, respectively, nuclear or cytoplasmic versions of the ataxin-1 Q82 protein that spontaneously forms large aggregates fused to GFP (kindly provided by Harry Orr). Coverslips were harvested 24h later, stained for the CHPK, and visualized by fluorescence microscopy. At least 200 (nuclear) or 300 (cytoplasmic) CHPK- and ataxin-positive nuclei were analyzed in three (nuclear) or four (cytoplasmic) separate experiments.
|
10.1371/journal.pgen.1004829 | Germline Signals Deploy NHR-49 to Modulate Fatty-Acid β-Oxidation and Desaturation in Somatic Tissues of C. elegans | In C. elegans, removal of the germline extends lifespan significantly. We demonstrate that the nuclear hormone receptor, NHR-49, enables the response to this physiological change by increasing the expression of genes involved in mitochondrial β-oxidation and fatty-acid desaturation. The coordinated augmentation of these processes is critical for germline-less animals to maintain their lipid stores and to sustain de novo fat synthesis during adulthood. Following germline ablation, NHR-49 is up-regulated in somatic cells by the conserved longevity determinants DAF-16/FOXO and TCER-1/TCERG1. Accordingly, NHR-49 overexpression in fertile animals extends their lifespan modestly. In fertile adults, nhr-49 expression is DAF-16/FOXO and TCER-1/TCERG1 independent although its depletion causes age-related lipid abnormalities. Our data provide molecular insights into how reproductive stimuli are integrated into global metabolic changes to alter the lifespan of the animal. They suggest that NHR-49 may facilitate the adaptation to loss of reproductive potential through synchronized enhancement of fatty-acid oxidation and desaturation, thus breaking down some fats ordained for reproduction and orchestrating a lipid profile conducive for somatic maintenance and longevity.
| Much is known about how increasing age impairs fertility but we know little about how reproduction influences rate of aging in animals. Studies in model organisms such as worms and flies have begun to shed light on this relationship. In worms, removing germ cells that give rise to sperm and oocytes extends lifespan, increases endurance and elevates fat. Fat metabolism and hormonal signals play major roles in this lifespan augmentation but the genetic mechanisms involved are poorly understood. We show that a gene, nhr-49, enhances worm lifespan following germ-cell removal. NHR-49 is increased in animals that lack germ cells by conserved longevity proteins, DAF-16 and TCER-1. NHR-49, in turn, increases levels of genes that help burn fat and convert saturated fats into unsaturated forms. Through synchronized enhancement of these processes, NHR-49 helps eliminate excess fat delegated for reproduction and converts lipids into forms that favor a long life. NHR-49 impacts these processes during aging in normal animals too, but using different regulatory mechanisms. Our data helps understand how normal lipid metabolic processes can be harnessed to adapt to physiological fluctuations brought on by changes in the reproductive status of animals.
| Many studies have documented the apparent trade-off between aging and reproduction as reduced fertility is associated with increased lifespan in several species [1]–[3]. However, reproductive fitness also confers distinct physiological benefits [4], [5]. A growing body of evidence underscores the complex interactions between aging and reproduction [6]–[9] but the mechanisms underlying this dynamic relationship remain obscure.
Aging and reproduction are both inextricably connected to the energetics of fat metabolism. Reproduction is an energy-intensive process that relies heavily on lipid supplies and is influenced by lipid homeostasis. Epidemiological data indicate that obesity and low-body weight together account for ∼12% of female infertility [10]. Reproductive senescence in women and other female mammals is marked by re-organization of body fat and frequently associated with weight gain [11]. Similarly, obesity not only increases the susceptibility to a host of age-related diseases such as diabetes and CVD, it may also directly accelerate the aging clock by hastening telomere attrition [12]. Thus, it would appear that lipid metabolism influences both reproduction and the rate of aging and may provide the basis for the impact of these processes on each other. These molecular underpinnings are poorly understood and identifying them has relevance for multiple aspects of human health, procreation and longevity.
In recent years, the nematode Caenorhabditis elegans has provided unique insights into the effect of reproductive status on the rate of organismal aging [7]–[9]. In C. elegans, sperm and oocytes are generated from a population of totipotent, proliferating germline-stem cells (GSCs) whose removal increases lifespan and enhances stress resistance [13], [14]. This phenomenon is not just a peculiarity of a hermaphroditic worm, since similar lifespan extension is exhibited by Drosophila melanogaster and other insect and worm species following germline removal [15]–[17]. Moreover, ovarian transplantation experiments in mice [18] and studies in human populations [19] suggest that the reproductive control of lifespan may be widely prevalent in nature.
The longevity of germline-ablated C. elegans is entirely dependent upon the presence of the conserved, pro-longevity FOXO-family transcription factor, DAF-16 [13]. DAF-16 is part of a transcriptional network that is activated in intestinal cells when the germline is eliminated [20]. DAF-16 is a shared longevity determinant that increases lifespan in response to multiple stimuli, including reduced insulin/IGF1 signaling (IIS) [21]. On the other hand, TCER-1, the worm homolog of the conserved, human transcription elongation and splicing factor, TCERG1 [22], specifically promotes longevity associated with germline loss [23]. Other components of the intestinal transcriptional network include regulators of cellular processes such as autophagy (PHA-4, HLH-30) [24], [25], heat-shock response (HSF-1) [26], oxidative stress (SKN-1) [27] and transcriptional co-factors (SMK-1) [28]. In addition to these proteins, a steroid signaling cascade that includes the nuclear hormone receptor (NHR), DAF-12, and components of a lipophilic-hormonal pathway that synthesize the DAF-12-ligand, dafachronic acid (DA), enhance the lifespan of germline-ablated animals ([29]; reviewed in [7], [9]). DAF-12 mediates the up-regulation of another NHR, NHR-80, that is in turn required for the increased expression of fatty-acid desaturases that catalyze the conversion of stearic acid (SA, C18:0) to oleic acid (OA, C18:1n9) [30]. DAF-12 also promotes DAF-16 nuclear localization in intestinal cells following germline ablation [31]. Several lines of evidence suggest that DAF-16-mediated lifespan extension relies on modulation of fat metabolism, at least in part, and involves lipophilic signaling [32], [33]. However, the mechanism through which DAF-16 orchestrates these lipid-metabolic changes is not known. NHR-80 and DAF-16 function in parallel pathways and NHR-80-mediated SA-to-OA conversion is not sufficient to overcome the loss of DAF-16 [30]. Other lipid regulators, including NHRs, which may act in the DAF-16 pathway to alter fat metabolism following germline removal are yet to be identified.
DAF-12 and NHR-80 are two members of a family of ∼284 NHRs represented in the worm genome, most of which have been derived from a hepatocyte nuclear factor 4 alpha (HNF4α) ancestor [34]. Many NHRs are lipid-sensing factors that respond to fatty acid and steroid ligands to alter gene expression. One such factor, NHR-49, shows sequence similarity to HNF4α, but performs functions undertaken in vertebrates by peroxisome proliferator-activated receptor alpha (PPARα). PPARα is a member of the PPAR family of proteins which plays essential roles in vertebrate energy metabolism and it operates at the hub of a regulatory complex that impacts fatty-acid uptake, lipoprotein transport and mitochondrial- and peroxisomal β-oxidation [35]. In worms, NHR-49 regulates of mitochondrial- and peroxisomal β-oxidation and fatty-acid desaturation during development and under conditions of food scarcity [36], [37]. nhr-49 mutants exhibit metabolic abnormalities, shortened lifespan and reduced survival upon nutrient deprivation [36], [37]. NHR-49 expression is also essential in a small group of GSCs that can survive long periods of starvation to re-populate the gonad and restore reproductive potential when the animal encounters food [38]. It is conceivable that this protein has a pervasive role in promoting organismal survival in diverse physiological contexts that induce metabolic flux and require the restoration of lipid homeostasis.
Despite the identification of several genes that encode lipid-modifying enzymes, how lipid homeostasis is re-established following germline loss, and how this translates into enhanced survival of the animal remains recondite. In this study, we identify a group of NHRs required for the longevity of germline-less C. elegans. We describe a role for one of these, NHR-49, in enhancing lifespan through modulation of specific lipid-metabolic pathways. We demonstrate that NHR-49 is transcriptionally up-regulated by DAF-16 and TCER-1 in the soma upon germline removal. NHR-49 causes the increased expression of multiple genes involved in fatty-acid β-oxidation and desaturation, triggering a metabolic shift towards lipid oxidation and an unsaturated fatty acid (UFA)-rich lipid profile. NHR-49 is critical for young germline-less adults to maintain their lipid reserves and de novo fat synthesis, and overexpression of the protein in fertile adults increases their lifespan modestly. nhr-49 single mutants display similar biochemical and age-related lipid deficits but not the widespread reduction in β-oxidation genes’ expression seen in germline-less mutants. NHR-49 expression during normal aging is DAF-16 and TCER-1 independent. It is also dispensable for the lifespan extension mediated by reduced insulin/IGF1 signaling (IIS), a DAF-16-dependent longevity pathway, suggesting that the DAF-16- and TCER-1-directed elevation of NHR-49 is especially important for the metabolic and lifespan changes induced by germline loss. Our results suggest that through the concerted enhancement of fatty-acid oxidation and desaturation, NHR-49 may mediate the breakdown of fats designated for reproduction and restore lipid homeostasis. Together, they provide evidence for an important role for NHR-49 in adapting to loss of reproductive potential and augmenting longevity.
Lipid signaling and fat metabolism play important roles in the reproductive control of aging [7]–[9]. Hence, to identify components of the DAF-16/TCER-1 pathway that confer lifespan extension upon germline loss, we focused on NHRs. These transcription factors are activated by lipid ligands and many of them modulate lipid-metabolic pathways. From the two large-scale, feeding RNAi libraries that cover a majority of the worm genome [39], [40], we derived a focused ‘NHR-library’ to perform an RNAi screen. Our ‘NHR-library’ included RNAi clones targeting 259 of the 283 worm NHRs. We used temperature-sensitive glp-1 mutants, a widely used genetic model for the longevity resulting from germline removal [41]. Previously, we had identified a GFP reporter, Pstdh-1/dod-8::GFP that is jointly up-regulated by DAF-16 and TCER-1 in intestinal cells of long-lived glp-1 mutants [23]. We used this strain (glp-1;Pstdh-1/dod-8::GFP) to screen our ‘NHR library’ for clones that prevented the up-regulation of GFP in young adults at 25°C.
We identified 22 RNAi clones, targeting 19 nhr genes, which prevented Pstdh-1::GFP up-regulation. 16 of these clones (targeting 13 NHRs) reproducibly reduced GFP expression (S1 Table) and also shortened the extended lifespan of glp-1 mutants, albeit with variable efficiency (11–48% suppression; S2 Table). We found that two independent RNAi clones targeting nhr-49 completely abrogated the longevity of glp-1 mutants (Fig. 1A, S2 Table). We chose to focus on nhr-49 because of these strong phenotypes, and because it's functional similarity to PPARα provided an avenue for investigating the mechanisms that link fat metabolism and longevity.
To substantiate the nhr-49 RNAi phenotype, we examined the effect of nhr-49 mutation on the extended lifespan of glp-1 mutants. We found that nhr-49(nr2041), a mutant that carries an 893 bp deletion, caused a suppression of glp-1 longevity similar to that caused by nhr-49 RNAi (Fig. 1B, S3 Table). The mutant also had a shorter lifespan compared to wild-type worms, as previously reported (Fig. 1B, S3 Table). Surprisingly, nhr-49 was not essential for the longevity of daf-2 mutants that live long due to impaired IIS and represent another DAF-16-dependent longevity pathway [21]. nhr-49 mutation had no impact on the extended lifespan of daf-2(e1368) mutants in two of three independent trials and caused a small suppression in longevity in the third (Fig. 1C and S4A Table). Similarly, results were obtained with nhr-49 mutants carrying another daf-2 allele, e1370, (Fig. 1C and S4A Table) and upon RNAi-inactivation of daf-2 in nhr-49 mutants (S4B Table). In C. elegans, lifespan is also enhanced by perturbations to mitochondrial electron transport chain activity through a distinct regulatory pathway that is daf-16 independent [42]. We found that RNAi treatment against cco-1 and cyc-1, genes that encode components of mitochondrial electron transport chain, elicited a similar lifespan extension in nhr-49 mutants as in wild-type worms (Fig. 1D, S4B Table). These observations suggest that nhr-49 has variable degrees of relevance for different physiological alterations that influence aging. It is critical for the longevity mediated by reproductive signals but is not central to the lifespan changes resulting from reduced IIS or deficient mitochondrial electron transport.
To address the role of nhr-49 in the reproductive control of aging, we first examined NHR-49 localization in worms. We generated transgenic worms expressing GFP tagged to a full length NHR-49 transgene driven under control of its endogenous promoter from extra-chromosomal arrays (Pnhr-49::nhr-49::gfp, henceforth referred to as NHR-49::GFP). Animals expressing NHR-49::GFP showed widespread fluorescence throughout embryonic and larval development (S1A Figure). In adults, it was visible in all somatic tissues (Fig. 1E–H), localized to both nuclei and cytoplasm, with highest expression in intestinal cells (Fig. 1H). Expectedly, the transgene was silenced upon nhr-49 RNAi except in neuronal cells (S1B Figure). To test if the NHR-49::GFP transgene was functional, we asked if it could rescue the shortened lifespan of nhr-49;glp-1 double mutants. In two independent trials, NHR-49::GFP completely rescued the longevity of nhr-49;glp-1 double mutants (Fig. 1I; S3 Table), whereas the rescue was 77% in a third trial (with strains generated by injecting the transgene at a lower concentration). This demonstrated that NHR-49::GFP is a functional protein that recapitulates the expression and function of the wild-type version.
Intestinal DAF-16 nuclear localization and TCER-1 transcriptional up-regulation are important molecular hallmarks associated with germline loss-dependent longevity [20], [23]. We asked if NHR-49 was similarly affected by germline removal. Germline depletion resulted in increased NHR-49::GFP, especially in intestinal cells (Fig. 2A, B, E, F). Next, we used the NHR-49::GFP reporter to test if this increased expression was dependent upon DAF-16 and/or TCER-1. In glp-1 mutants carrying the daf-16 null allele, mu86, GFP expression was dramatically and uniformly reduced in all tissues (Fig. 2C, E). daf-16 knockdown by RNAi caused a similar but less marked reduction in NHR-49::GFP in glp-1 mutants (Fig. 2F, striped bars). In tcer-1;glp-1 double mutants, GFP expression pattern was unevenly affected. In most animals, some intestinal cells showed no GFP whereas others showed high GFP expression (Fig. 2D, E). It is not clear if the mosaic expression observed in tcer-1;glp-1 mutants is a result of partial loss of function (the tcer-1 allele, tm1452, is a 392 bp deletion coupled to a 10 bp insertion that is predicted to disrupt three of five transcripts produced by the gene) or reflects a spatial aspect of regulation by TCER-1. glp-1 mutants subjected to tcer-1 RNAi had reduced GFP expression as well (Fig. 2F, striped bars). In addition to these observations, an independent line of investigation supported the regulation of NHR-49 by DAF-16 and TCER-1. In an RNA-Sequencing (RNA-Seq) analysis designed to map the transcriptomes dictated by DAF-16 and TCER-1 upon germline ablation, we identified nhr-49 as one of the genes jointly up-regulated by these two proteins (Amrit et al., manuscript in preparation). Using Q-PCR assays, we confirmed that germline removal produced a significant increase in nhr-49 mRNA, and this increase was repressed in daf-16;glp-1 and tcer-1;glp-1 mutants (the tcer-1 mutant did not achieve statistical significance; Fig. 2G). Interestingly, DAF-16 and TCER-1 up-regulated nhr-49 expression only in germline-ablated worms. In daf-16 and tcer-1 mutants alone, we did not observe a reduction in nhr-49 mRNA during adulthood (Fig. 2H and S2 Figure). RNAi knockdown of these genes also did not reduce NHR-49::GFP levels (Fig. 2F, solid bars) suggesting that NHR-49 is differentially regulated depending on the reproductive status of the animal. Together, our experiments show that both mRNA and protein levels of NHR-49 are elevated in somatic cells upon germline removal. These changes are strongly dependent on DAF-16, at least partially dependent on TCER-1, and indicate that DAF-16 and TCER-1 mediate the transcriptional up-regulation of NHR-49 when the germline is eliminated.
Since NHR-49 expression is increased in glp-1 mutants, we asked if elevating levels of the protein in normal animals could circumvent the need for germline removal and directly enhance longevity. We used the NHR-49::GFP animals that overexpressed the protein due to the presence of multiple extra-chromosomal arrays of the transgene (Fig. 3A). Indeed, we found that wild type, fertile worms overexpressing NHR-49 lived ∼15% longer than their non-transgenic siblings and wild-type controls (Fig. 3B, S3 Table). Intriguingly, the lifespan enhancement was greater when NHR-49 was overexpressed in an nhr-49 mutant background. Not only was the lifespan of nhr-49 mutants rescued to wild-type levels, it was augmented even further (Fig. 3C). These long-lived worms did not display any obvious fertility defects (S3 Figure). NHR-49 overexpression in glp-1 mutants caused a small additional increment in their longevity as well (Fig. 3D). This lifespan increment was dependent on both daf-16 and tcer-1 (Fig. 3E, F and S5A Table). These data show that elevating NHR-49 levels can increase lifespan modestly without compromising fertility.
During development and in response to food deprivation, NHR-49 regulates the expression of multiple genes predicted to function in mitochondrial- and peroxisomal- β-oxidation (Fig. 4A) as well as fatty-acid desaturation pathways (Fig. 4B) [36], [37]. Strikingly, along with nhr-49, many of these genes, were also identified as DAF-16 and TCER-1 targets in the RNA-Seq analysis mentioned above (S4 Figure; Amrit et al., manuscript in preparation). This led us to ask (a) if the expression of these genes was enhanced in glp-1 mutants, and (b) whether their up-regulation was dependent upon nhr-49. We focused on the mitochondrial β-oxidation genes. In Q-PCR assays, the mRNA levels of 12 genes we tested were all elevated in long-lived, glp-1 mutants as compared to wild-type worms, although to variable degrees (Fig. 4C–N). Of these, the up-regulation of seven genes was significantly reduced or abolished in nhr-49;glp-1 mutants (Fig. 4C-N). These genes encode enzymes that participate in different steps of mitochondrial β-oxidation including: i) acyl CoA synthetases (ACS; acs-2 and acs-22) that catalyze the conversion of fatty-acids to acyl CoA ii) carnitine palmitoyl transferases (CPT; cpt-2, cpt-5) that transport activated acyl groups from the cytoplasm into the mitochondrial matrix and iii) acyl CoA dehydrogenases (ACDH; acdh-9, acdh-11), enoyl CoA hydratases (ech-1.1, ech-7) and thiolase (acaa-2) whose combined activities result in the shortening of fatty-acid moieties and generation of acetyl CoA (Fig. 4A) [43]. Thus, NHR-49 mediates the increased expression of genes involved in different steps of mitochondrial β-oxidation following germline loss. We did not observe a similar, conspicuous difference in the expression of these genes on comparing wild-type worms with nhr-49 single mutants. The expression of two genes, acs-2 and ech-1.1, was reduced in nhr-49 mutants (Fig. 4C, E) and acdh-9 was elevated (Fig. 4G), but the others were not significantly altered (Fig. 4 and S5 Figure). Surprisingly, worms overexpressing NHR-49 also did not exhibit a consistent change in the mRNA levels of these genes, although they are longer lived and it was conceivable that they may have elevated β-oxidation gene expression (S6 Figure). These observations suggest that germline removal may provide the impetus for a perspicuous up-regulation of β-oxidation genes by NHR-49.
To test if these gene expression changes had any relevance on the lifespan extension observed in germline-less animals, we examined the effect of RNAi knockdown of each of these genes on the longevity of glp-1 mutants. RNAi was initiated with the onset of adulthood to circumvent developmental requirements. We found that RNAi knockdown of seven of nine genes shortened glp-1 longevity to variable degrees (7–48%; Table 1 and S6 Table). On the other hand, RNAi knockdown of the same genes in a control strain with wild-type lifespan either had no statistically significant effect (7/9 genes tested) or an inconsistent lifespan reduction (2/9 genes) (Table 1 and S6 Table). These results underscore the singular importance of the β-oxidation genes we tested to the longevity of germline-less animals. Together, our data defined a functional role for the NHR-49-mediated up-regulation of mitochondrial β-oxidation genes in response to germline removal. Moreover, they suggested that in germline-less animals, NHR-49 triggers an increase in fatty-acid β-oxidation and this metabolic shift is critical for the consequent lifespan extension.
Germline loss results in increased triglyceride (TAG) storage in C. elegans [44]. Based on the observations described above, we asked if inhibiting mitochondrial β-oxidation affected the elevated fat stores of glp-1 mutants. As a preliminary test, we compared the lipid levels of glp-1 and nhr-49;glp-1 mutants by staining the animals with the dye Oil Red O (ORO) that labels TAGs and whose estimation closely matches biochemically detected TAG levels [44]. Since β-oxidation is a lipolytic pathway, it was conceivable that nhr-49;glp-1 mutants would exhibit a further increase in TAGs due to their impaired mitochondrial β-oxidation gene expression profile. However, we found that ORO levels were indistinguishable between glp-1 and nhr-49;glp-1 day 1 adults (Fig. 5A, B and I). Surprisingly, by day 2, nhr-49;glp-1 adults showed a small but significant reduction in ORO staining as compared to glp-1 mutants (Figs. 5C, D, I and S6A Figure). As the animals aged, this difference became more pronounced. By days 6–8 of adulthood, nhr-49;glp-1 mutants underwent a striking loss of ORO staining (Fig. 5E–I). By comparison, glp-1 mutants continued to show high ORO staining from day 2 till at least day 18 of adulthood (S6B Figure). To obtain a direct measure of lipid levels, we used gas chromatography/mass spectrometry (GC/MS) and found that TAG levels are indeed significantly reduced in nhr-49;glp-1 mutants as compared to glp-1 adults (Fig. 5J).
To understand why nhr-49;glp-1 mutants exhibited reduced TAG levels despite diminished expression of β-oxidation genes, we explored the possibilities that (a) they consumed less food than glp-1 mutants, and/or (b) they also experienced a simultaneous reduction in fatty-acid synthesis. We observed no difference in the pharyngeal pumping rates of the two strains, indicating that they ate similar quantities of food (S6C Figure). Next, we compared the dietary fat absorption and de novo fat synthesis between glp-1 and nhr-49;glp-1 mutants using a previously described 13C isotope fatty-acid labeling assay [45]. de novo fat synthesis was substantially reduced in nhr-49;glp-1 mutants (Fig. 5K). We confirmed that this reduction was not due to repressed transcription of genes involved in initiation of fat synthesis or those mediating conversion to stored fats. mRNA levels of pod-2 {that encodes acyl CoA carboxylase (ACC), the rate-limiting enzyme required for initiation of fat synthesis} and fasn-1 {that encodes fatty-acid synthase (FASN-1), another key regulator of fat synthesis} were not reduced by nhr-49 reduction of function (S7A and B Figure). Similarly, nhr-49 did not affect the expression of dgat-2, a gene that encodes a rate-limiting enzyme diacylglycerol acyl transferase (DGAT) needed for conversion of diglycerides (DAGs) into TAGs (S7C Figure). Thus, our experiments showed that germline-less animals require nhr-49 for de novo lipid synthesis and to retain their high TAG levels during adulthood. They suggest that impairing NHR-49 may impact other important metabolic processes besides β-oxidation such as lipid synthesis, storage and maintenance.
Since nhr-49 mutants are short-lived compared to wild-type controls [36], [46], we asked if they also exhibited age-related fat phenotypes, and if NHR-49 played an analogous role during normal aging. nhr-49 mutants have been reported to have higher fat. These studies predominantly relied on staining live larvae with the dye Nile Red [36], [47]–[49], an inaccurate technique for labeling fats as the dye is trafficked to the lysosome-related organelle in live animals [44], and in some cases these observation have not been corroborated by other methods [50]. Using ORO labeling, we did not observe a significant difference between wild-type, day 2 animals and age-matched nhr-49 mutants (Fig. 5L and S6A Figure). However, while wild-type worms underwent increased fat accumulation with age, nhr-49 mutants, similar to nhr-49;glp-1, exhibited a progressive loss of fat (Fig. 5L). Intriguingly, a similar age-related loss of ORO staining was also observed in worms overexpressing NHR-49 (Fig. 5L). These phenotypes could not be explored biochemically in the reproductively active day 2 adults of these strains due to the confounding effects of eggs and progeny (see Materials and Methods). Hence, we used late L4 larvae/early day 1 adults to compare the lipid profiles of wild-type worms and nhr-49 mutants. GC/MS data showed that, at least in late L4/early day 1 adults, lipid levels were the same between the two strains (Fig. 5M). Since the biochemical analyses could not be extended to adults, it is formally possible that nhr-49 mutants have elevated fat. But, our experiments strongly indicate that nhr-49 loss of function does not increase fat accumulation. Instead, in both germline-less and normal adults, it causes an age-related loss of stored lipids. On comparing the dietary fat absorption and de novo lipid synthesis profiles between late L4/early day 1 nhr-49 mutants and wild-type worms, we noticed fatty-acid specific differences. Some fatty acids were synthesized at a higher level in nhr-49 mutants as compared to wild-type (eg., OA) whereas the synthesis of others (eg., Vaccenic Acid, C18:1n7) was reduced (Fig. 5N and S8A Figure). Together, these experiments showed that NHR-49 is required for the maintenance of TAG stores during normal aging and its absence causes de novo lipid synthesis abnormalities. The similarities between the phenotypes of nhr-49 and nhr-49;glp-1 mutants suggest a shared role for the gene in the two contexts. However, the gene expression and lifespan studies described in the previous section also point towards mechanistic and possibly functional differences in NHR-49's modulation of these processes in fertile vs. germline-less adults (see Discussion).
NHR-49 regulates both fatty-acid β-oxidation and desaturation during development and nutrient deprivation [36], [37] so we asked if it impacted desaturation in the glp-1 mutant context and/or during normal aging. The genes fat-5, fat-6 and fat-7 encode desaturase enzymes that catalyze the conversion of SFAs to MUFAs. FAT-5 converts palmitic acid (PA, C16:0) to palmitoleic acid (POA, C16:1n7) while FAT-6 and -7 function redundantly to convert stearic acid (SA, C18:0) to oleic acid (OA, C18:1n9) (Fig. 4B) [51], [52]. Under control of NHR-80, FAT-6/7 mediated conversion of SA to OA is necessary for the longevity of glp-1 mutants; OA supplementation completely rescues the short lifespan of glp-1;fat-6;fat-7 mutants to glp-1 level [30]. We found that, similar to NHR-80, NHR-49 was also required for the changes in levels of fat-5, fat-6 and fat-7 observed in glp-1 mutants (Fig. 6A–C). However, OA supplementation did not rescue the shortened lifespan of nhr-49;glp-1 mutants (S7 Table) indicating other critical functions for NHR-49. On comparing the lipid profiles of nhr-49;glp-1 with glp-1 mutants through GC/MS, we observed an increased SA:OA ratio in the former, as expected (Fig. 6D and S9A Figure). In addition, the ratio of PA:POA was enhanced as well (Fig. 6E and S9B Figure). Overall, nhr-49;glp-1 mutants exhibited a widespread decline in MUFAs and increased SFAs in both the neutral and phospholipid pools (Figs. 6F, G and S9C, D Figure). Desaturation is coupled to the elongation of fatty-acid chains that is mediated by elongase enzymes (encoded by the ‘elo’ genes in C. elegans). Our Q-PCR assays showed that nhr-49 was not required for the up-regulation of elo-1 and elo-2 in glp-1 mutants (S7D, E Figure) implying a selective role for the gene in desaturation. Overall, these data showed that NHR-49, similar to NHR-80, is required for SA-to-OA conversion in glp-1 mutants. In addition, it also promotes the desaturation of other SFAs to MUFAs and PUFAs to ensure an UFA-rich lipid profile. Hence, while NHR-80 influences desaturation alone, NHR-49 modulates both desaturation and β-oxidation and has a broader effect on lipid composition. This may also explain the more severe phenotypes associated with nhr-49 reduction-of-function.
In nhr-49 single mutants, the levels of fat-5 and fat-7 mRNAs were reduced, whereas the effect on fat-6 was inconsistent and statistically insignificant (Figs. 6A-C and S5 Figure). Despite these weak gene-expression effects, the fatty-acid profile of late L4/early day 1 nhr-49 mutants showed increased SA:OA ratio (PA:POA ratio was increased only in neutral lipids; Figs. 6H, I and S10A, B Figure) and an increased accumulation of SFAs with a concomitant reduction in MUFAs (Figs. 6J, K and S10C, D Figure) indicating a role for NHR-49 in establishing a MUFA-rich lipid profile in normal animals too. Overall, the multiple fat phenotypes of nhr-49;glp-1 mutants, the role of nhr-49 in enhancing fatty-acid β-oxidation as well as desaturation and our biochemical and functional data together suggest that through the coordinated enhancement of β-oxidation and desaturation, NHR-49 helps establish lipid homeostasis that is critical for the survival of germline-less animals, and may also impact normal aging.
In this study, we show that NHR-49 promotes the longevity of germline-less adults through the increased expression of genes that mediate mitochondrial β-oxidation and fatty-acid desaturation. Our data strongly suggest that germline-removal enhances fatty-acid oxidation and desaturation through NHR-49 activity. We propose that through the synchronized up-regulation of these ostensibly disparate lipid-metabolic pathways, NHR-49 facilitates the adaptation to loss of reproductive potential (by eliminating fats designated for reproduction) and helps establish a lipid profile that favors increased lifespan (by converting SFAs into UFAs that are more conducive for cellular maintenance (S11 Figure) [53]–[56].
A key finding of our study is the identification of multiple genes predicted to function in fatty-acid β-oxidation whose expression is up-regulated following germline loss, and the strong dependence on NHR-49 for this up-regulation. These genes encode enzymes that together cover all the catalytic reactions of β-oxidation including some that are specific to the process (e.g., CPTs) [43]. While we cannot rule out the possibility that they function together in a different pathway, the simplest interpretation of our data is that these genes enhance mitochondrial β-oxidation. These data imply that germline removal causes a shift towards fatty-acid metabolism. Lipid oxidation confers several advantages over glucose metabolism such as more efficient energy production and reduced reactive oxygen species (ROS) generation [57]. In C. elegans, fatty-acid oxidation provides energy in other situations where stored lipids are used for long-term survival such as dauer diapause and caloric restriction [47], [58]. However, in these contexts, the animal is food deprived and not faced with the hazard of large-scale lipid accumulation due to thwarted procreation. Following germline loss, a metabolic shift towards increased β-oxidation coupled to lipid repartitioning may allow the animal to eliminate fats normally delegated for reproduction and restore lipid homeostasis, thus averting the negative consequences of loss of fertility. Such a metabolic shift can also explain the extraordinary dependence of germline-ablated animals on the presence of NHR-49, a key mediator of both oxidation and desaturation. Fatty-acid oxidation and desaturation, although independent processes, are intimately interlinked and inter-dependent. Deficiency of the mouse desaturase, SCD1, inhibits β-oxidation in cardiac cells [59]. Alternatively, impaired β-oxidation impacts lipid composition and is implicated in human dyslipidemias [60]. A coordinated up-regulation of these processes would be especially relevant for germline-less animals, since they face the dual challenges of eliminating superfluous fat and transforming their lipid profile in adaptation to an altered physiological status.
We were intrigued by the progressive depletion of stored fats, despite impaired expression of β-oxidation genes, in nhr-49;glp1 mutants. While the precise reason for this is unknown, we postulate that it may be due to the simultaneous inhibition of β-oxidation and desaturation that causes accumulation of free fatty acids (FFAs) [61]. FFAs stimulate insulin release and serve as key signaling molecules. But their chronic accrual causes deregulated insulin secretion and apoptosis in pancreatic β cells [62] and insulin resistance in muscle and liver cells [63]. Impaired fatty-acid oxidation and non-metabolized SFAs are implicated as the primary agents underlying lipotoxicity [64]. We observed a significant increase in such SFAs in nhr-49;glp-1 mutants. Hence, it is conceivable that in nhr-49;glp-1 mutants inadequate mobilization of fat stores and impaired desaturation together cause FFA accretion and an energy imbalance that may lead to early death. Further studies will be needed to test this hypothesis and unravel the molecular basis of this intriguing phenotype.
The requirement of NHR-49 for enhancement of both β-oxidation and desaturation following germline removal distinguishes the protein from other regulators such as NHR-80 which influences desaturation, especially SA to OA conversion. In our experiments, NHR-49 also had a wider impact on the fatty-acid composition of glp-1 mutants. Besides SA, nhr-49;glp-1 mutants exhibited reduced desaturation of multiple fatty acids including PA conversion to POA. They displayed overall reduction in MUFAs and PUFAs and a concomitant increase in SFAs (Fig. 6). These data suggest a broader role for NHR-49 in the increased fatty-acid desaturation associated with germline-loss.
Two independent approaches led us to the identification of DAF-16 and TCER-1 as regulators of nhr-49: the RNAi screen described here and an RNA-Seq study designed to identify DAF-16 and TCER-1 targets (Amrit et al., manuscript in preparation). NHR-49::GFP confirmed the RNA-Seq and Q-PCR data. Loss of daf-16 almost completely abolished NHR-49::GFP in glp-1 mutants but had no impact in fertile adults (Fig. 2). TCER-1 also specifically enhanced NHR-49 in a glp-1 background. These observations provide clues as to how reproductive stimuli may modulate somatic metabolism. Since germline loss triggers intestinal nuclear relocation of DAF-16 and elevated TCER-1 expression [20], [23], it is possible that these two events stimulate increased nhr-49 transcription. But, it is not clear at present if nhr-49 is a direct DAF-16 target because we did not find a canonical DAF-16-Binding Element (DBE) [65] in the promoter used in our study. The strong DAF-16-dependence in glp-1 mutants also distinguished NHR-49 from NHR-80 whose up-regulation in germline-ablated animals is largely DAF-16 independent [30]. While the lifespan of daf-16;glp-1 mutants is increased by NHR-80 [30], NHR-49 overexpression in these animals did not rescue longevity significantly (S5B Table).
The short lifespan of nhr-49 mutants led us to explore its role during normal aging. nhr-49 loss causes similar age-related fat loss and biochemical deficits in both germline-less and wild-type adults, but we also noticed mechanistic and regulatory differences between the two paradigms. For instance, nhr-49;glp-1 mutants exhibited a consistent reduction in the de novo synthesis of OA, an important determinant of glp-1 longevity [30], whereas, nhr-49 mutants appeared to synthesize more of it, at least at late L4/early day 1 stage. Similarly, DAF-16 and TCER-1 mediated increased NHR-49 expression in glp-1 mutants but were not needed for the basal expression in wild-type adults. Further, NHR-49 was required for the up-regulation of multiple mitochondrial β-oxidation genes in glp-1 mutants, whereas, their levels were largely unchanged by its depletion in fertile adults. RNAi knockdown of these genes also impacted glp-1 longevity selectively (Table 1). These data suggest that germline-less animals experience enhanced β-oxidation and are more dependent upon it for survival, whereas basal levels are maintained in young, fertile adults. In the light of these differences, the similarities in age-related fat loss and fatty-acid composition defects between nhr-49 and nhr-49;glp-1 mutants are intriguing. One possible explanation for these contradictory observations is that nhr-49 controls the same pathway in the two situations but through the regulation of different targets, a premise supported by the considerable redundancy observed in C. elegans mitochondrial β-oxidation genes. It is also plausible that NHR-49 influences a different process in fertile adults whose inactivation also leads to progressive fat loss. Interestingly, other longevity-promoting genes exhibit similar phenotypes. For instance, HSF-1 is needed for the longevity of daf-2 mutants and their enhanced stress resistance. But, its depletion also shortens lifespan and increases stress-susceptibility in wild-type worms [26]. Similarly, DAF-16 and SKN-1 are both essential for daf-2 longevity and stress-resistance and they are also critical for normal worms’ ability to mount a response against oxidative stress, pathogen attack and other noxious stimuli [66]–[68]. Mutations in both genes shorten lifespan in wild-type worms [66], [67], though not to the extent seen in hsf-1 and nhr-49 mutants. These similarities may reflect a common mechanism by which normal cellular and metabolic pathways are leveraged and enhanced by an organism to cope with major physiological changes, and how this may in turn lead to a change in the length of life. Our results suggest that increased mitochondrial β-oxidation and transformation of the lipid profile into one enriched in UFAs may not only allow adaptation to germline loss but also be beneficial to normal aging animals.
All strains were maintained by standard techniques at 20°C. Lifespan experiments were conducted as described previously and have been discussed in detail elsewhere [69]. For all lifespan assays that involved the glp-1 genetic background, eggs were incubated at 20°C for 2–6 h, transferred to 25°C to eliminate germ cells, then shifted back to 20°C on day 1 of adulthood (∼72 h later) for the rest of their lifespan. For fer-15;fem-1 lifespans, eggs were similarly transferred to 25°C to induce sterility and left at the same temperature for life. For lifespans with daf-2 mutants, worms were grown at 15°C till L4 stage and then transferred to 20°C for life. All other lifespan assays were performed at 20°C. In all cases, the L4 stage was counted as day 0 of adulthood. Fertile strains were transferred every other day to fresh plates until progeny production ceased. For lifespans performed with transgenic strains, eggs were transferred to fresh plates and after 48 h scored under a Leica M165FC microscope with a fluorescence attachment (Leica Microsystems, Wetzlar, Germany). Transgene-carrying, fluorescent L4 larvae (day 0) were separated from their age-matched, non-transgenic siblings. The latter were used as internal controls in the same experiment. For whole-life RNAi experiments, worms were exposed to RNAi clones from hatching by transferring eggs to RNAi plates. For adult-only RNAi lifespans, the worms were grown on E. coli OP50 till day 0 and then transferred to freshly-seeded RNAi plates for the rest of adulthood. pAD12, an empty vector plasmid without an RNAi insert [42] was used as the control in all RNAi lifespans along with pAD43 [42] and tcer-1 RNAi constructs to knock-down daf-16 and tcer-1, respectively. Data from animals that crawled off the plate, exploded, bagged, or became contaminated were censored on the day the abnormality was observed. Stata 10.0, 8.2 (Stata Corporation, Texas, USA) and OASIS (Online Application of Survival Analysis, http://sbi.postech.ac.kr/oasis) were used for statistical analysis. P-values were calculated using the log-rank (Mantel–Cox method) test. The complete genotypes and pertinent details of all the strains used in this study are given in S8 Table.
According to Wormbase WS239, 283 genes are annotated as nuclear hormone receptors (www.wormbase.org). Of these, we could isolate 429 clones targeting 259 nhr genes from the feeding RNAi feeding libraries created by the laboratories of Julie Ahringer and Marc Vidal [39], [40]. This ‘sub-library’ was screened to identify RNAi clones that suppressed the up-regulation of Pstdh-1/dod-8::gfp in a long-lived glp-1 mutant using the strain CF2573 [23]. Briefly, RNAi clones were inoculated overnight at 37°C in LB medium containing 100 µg/ml ampicillin. 100 µl culture of each clone was seeded onto NGM plates with ampicillin (100 µg/ml) and supplemented with 1 mM IPTG. Synchronized eggs of CF2573 were isolated by hypochlorite treatment and seeded onto freshly-seeded RNAi plates. After 4–6 h at 20°C the plates were moved to 25°C for ∼70–72 h and then screened under a Leica M165FC microscope with a fluorescence attachment (Leica Microsystems, Wetzlar, Germany). In addition to pAD12, multiple random clones were also used as baseline negative controls (since pAD12 causes a modest, non-specific reduction in fluorescence in all GFP-expressing strains). All screen plates were independently examined by two observers. Clones identified by both observers were tested in three subsequent trials (S1 Table). All confirmed suppressor (and some enhancer) RNAi clones were confirmed by sequencing (M13-forward primer) and before any experiment, RNAi clones were tested by PCR (T7 primers).
To generate the Pnhr-49::nhr-49::gfp construct, 6.6 kb region of nhr-49 gene (4.4 kb comprising the coding region covering all nhr-49 transcripts and 2.2 kb sequence upstream of the first nhr-49 exon) was amplified with primers modified to introduce PstI and SalI restriction sites (forward 5′ gctagCTGCAGgaccagaaagagcaagagccaatattct 3′; reverse 5′ taagcaCCCGGGtcgagcatatgattattctgctcactg 3″). The amplified product was cloned into the GFP expression vector pPD95.77 (Addgene plasmid 1495). The full-length nhr-49 fragment was inserted upstream of, and in frame with, GFP at the PstI and SalI sites (pAG4). To generate the NHR-49::GFP expressing worms, Pnhr-49::nhr-49::gfp (pAG4) was injected at a concentration of 25 ng/µL or 100 ng/µL with 3.75 ng/uL or 15 ng/µL of Pmyo-2::mCherry co-injection marker, respectively. Three to six independent stable transgenic lines were generated for each of the genetic backgrounds in which the transgene was injected. Transgenic strains were maintained by picking fluorescent worms in each generation. The strains generated for this study are listed in S8 Table.
For GFP assays involving NHR-49::GFP, eggs were transferred to freshly-seeded E. coli OP50 or RNAi plates, incubated at 20°C for 2–6 h, transferred to 25°C (to eliminate germ cells in strains containing glp-1 mutation), then shifted back to 20°C on day 1 of adulthood. GFP assays were conducted on day 2 of adulthood, using the Leica MZ16F stereomicroscope. All assays were performed blind after initial familiarization with GFP levels in control plates by the experimenter.
For imaging purposes, 6 to 10 worms were immobilized in 35 mm optical glass bottomed dishes (MatTek Corporation, Ashland, MA) with 6 µl of 0.1 mM sodium azide in PBS. Confocal images were taken using a Leica TCS SP8 microscope. GFP fluorescence was illuminated using a 488 nm argon laser line with a 63×1.4NA oil Apochromat CS2 objective. Fluorescence was captured using a spectral HyD detector over ∼100 Z-planes. Confocal images were visualized, rendered and analyzed using Volocity Visualization Software (v 5.4, PerkinElmer).
ORO staining was done as described in earlier [44]. Briefly, 0.5 g ORO (Sigma-Aldrich St. Louis, MO) was dissolved in 100 mL isopropanol and the solution was equilibrated for four days. One day before staining, the stock solution was diluted to 60% with water and filtered twice on the day of the experiment through a 0.22 µm filter. 30–40 adults were picked into a 1.5 mL tube containing 1×PBS, washed twice with 1×PBS pH 7.4 and settled by spinning at 2000 rpm for 1 min. The worms were then re-suspended in 120 µL PBS to which an equal volume of 2×MRWB buffer was added. Samples were rocked gently for 1 h at room temperature and centrifuged at 2000 rpm for 1 min. The buffer was aspirated, worms washed with PBS, re-suspended in 60% isopropanol and incubated for 15 minutes at room temperature. After 15 minutes, the 60% isopropanol solution was removed and worms were then incubated overnight with rocking in 1 ml filtered 60% ORO stain. Next day the dye was removed after allowing worms to settle, and 200 µL of 1×PBS 0.01% Triton X-100 was added. Animals were mounted and imaged with using a Leica M165FC microscope mounted with a Retiga 2000R camera (Q Imaging, Burnaby, British Columbia, Canada). Images were captured with the QCapture Pro7 software (Q Imaging, Burnaby, British Columbia, Canada) and quantified using ImageJ software (NIH).
To perform Q-PCRs, total RNA was isolated using mirVana miRNA Isolation Kit (Ambion, Austin, TX) from approximately 5,000 day 2 worms of each strain grown under identical conditions. RNA was treated with DNase I, (Sigma-Aldrich St. Louis, MO) and cDNA was prepared from 1 µg of total RNA in a 20 µL reaction using a ProtoScript first strand cDNA synthesis kit (New England Biolabs, Beverly, MA, USA). For comparing mRNA levels from strains carrying extra-chromosomal transgenes, fluorescent worms were picked on to a 10 cm NGM plate seeded with E. coli OP50 and allowed to lay eggs that were maintained at 20°C. On day 1 of adulthood, worms were washed with M9 and transferred to a NGM plates seeded with E. coli OP50 to prevent starvation. Transgenic worms were isolated on day 2 using a Leica MZ16F stereomicroscope (Leica Microsystems, Wetzlar, Germany) with standard fluorescence filters. For each strain, approximately 200 worms were used for RNA isolation. Total RNA was extracted with TRIzol (Ambion, Austin, TX) and converted to cDNA as described above. Q-PCRs were performed using an ABI 7000 machine (Applied Biosystems). PCR reactions were undertaken in 96-well optical reaction plates (ABI PRISM N8010560). A 25 µL PCR reaction was set up in each well with 12.5 µL SensiMix SYBR Hi-ROX Kit (Bioline, USA), 1/20th of the converted cDNA and 0.25 M primers. For every gene at least three independent biological samples were tested, each with three technical replicates. Primers used in this study are listed in S9 Table.
The pharyngeal pumping assay was done as previously described [70]. Briefly, age matched glp-1 and nhr-49;glp-1 worms were obtained by picking eggs on to E. coli OP50 seeded plates. On the day of counting one worm was transferred to a freshly seeded E. coli OP50 plate and allowed to recover for 2–5 minutes. Pumping rate was determined by counting the contraction of the terminal bulb of the pharynx for 30 seconds under a dissecting microscope. The counting was repeated 4 more times to get the average. After the fifth replicate, the worm was moved to a freshly seeded E. coli OP50 plate. Pumping rate was measured on day 2, 4 and 6 of adulthood.
To assess reproductive health, brood size, percentage of hatching and oocyte ratio were calculated, using at least 10 worms per strain, per biological replicate, as described previously [70].
For each strain, gravid adults were bleached to obtain approximately 15,000 eggs that were transferred to NGM plates seeded with E. coli OP50 for growth. The plates were incubated for 2 hours at 20°C and then transferred to 25°C for growth to desired stage. For experiments performed with day 2 sterile adults, the plates were transferred back to 20°C after 72 hours for another 18 hours of additional growth and then transferred to prepared stable isotope plates for 6 hours. The same protocol was followed harvesting worms at late L4/early day 1 adults, except for harvesting after 52 hours (N2 and glp-1) or 64 hours (nhr-49 and nhr-49;glp-1). The additional growth time was provided for nhr-49 mutant strains to compensate for their developmental delay under large-scale growth conditions. By day 2 of adulthood, the altered de novo synthesis and fatty-acid composition profiles of nhr-49;glp-1 mutants were similarly changed when compared to nhr-49;glp-1 mutants harvested together with glp-1 (simultaneous 96 hour harvest) or after a 12 hour delay (108 hour harvest) (S12 Figure). Larvae and adult animals utilize their fatty acids differently. In growing larvae, large quantities of lipids are used to build membranes and establish lipid stores, whereas in fully-grown adults, they are utilized to fulfill the demands of reproduction. Using wild-type N2 and other fertile strains in lipid assays confounds the results as we begin to see the metabolic profile of the progeny skew the data as early as day 2 (Shaw Wen-Chen and Carissa Olsen, unpublished data). To circumvent this, we used late L4 larvae/early day 1 adults for our lipidomic studies with fertile strains.
The stable isotope plates were prepared as previously described; in short, each plate was seeded with a 1∶1 ratio of 12C-bacteria and 13C-bacteria grown respectively in LB or Isogro media (98.5% 13C-enriched media, Sigma-Aldrich). The animals were harvested, washed in M9 three times, and frozen in a dry ice/ethanol bath before being stored at −80°C until processed. Total lipids were extracted as previously described with the following modifications [45]. Briefly, standards were added to each sample (1,2-diundecanoyl-sn-glycero-3-phosphocholine, Avanti Polar Lipids, for PLs and tritridecanoin, Nu-Chek Prep, for TAGs) before the start of the extraction procedure. The lipids were extracted with 2∶1 chloroform:methanol for 90 minutes at room temperature while shaking continuously. Residual carcasses were pelleted by centrifugation and the extracted lipids were transferred to fresh tubes and dried under a constant nitrogen stream. Dried lipids were re-suspended in 1 mL chloroform and loaded onto a pre-equilibrated solid phase exchange (SPE) columns (100 mg capacity, Fisher Scientific). Lipid classes were eluted from the column in the following order: neutral lipids in 3 ml of chloroform, glycosphingolipids in 5 ml of acetone:methanol (9∶1) and then phospholipids in 3 ml of methanol. Purified lipids were dried under nitrogen, re-suspended in methanol/2.5% H2SO4 and incubated for 1 h at 80°C to create fatty acid methyl esters (FAMEs) that were analyzed by gas chromatography/mass spectrometry (GC/MS) (Agilent 5975GC, 6920MS).
The relative abundance of fatty acids in each class was determined for all the major fatty acid species in the nematode as previously described [45]. To quantify TAG and PL yields, total PL and TAG abundance were normalized using the added standards, and data were presented as a TAG:PL ratio. de novo synthesis was calculated through a series of described equations which allow for the quantification of the amount of each fatty acid species generated from synthesis by determining the abundance of each isotopomer [45]. The synthesis numbers reported here represent the amount of 13C-labeled fatty acids derived from synthesis when compared to the total amount of fatty acids newly incorporated into the animal.
|
10.1371/journal.pntd.0003109 | Epidemiology of Disappearing Plasmodium vivax Malaria: A Case Study in Rural Amazonia | New frontier settlements across the Amazon Basin pose a major challenge for malaria elimination in Brazil. Here we describe the epidemiology of malaria during the early phases of occupation of farming settlements in Remansinho area, Brazilian Amazonia. We examine the relative contribution of low-density and asymptomatic parasitemias to the overall Plasmodium vivax burden over a period of declining transmission and discuss potential hurdles for malaria elimination in Remansinho and similar settings.
Eight community-wide cross-sectional surveys, involving 584 subjects, were carried out in Remansinho over 3 years and complemented by active and passive surveillance of febrile illnesses between the surveys. We used quantitative PCR to detect low-density asexual parasitemias and gametocytemias missed by conventional microscopy. Mixed-effects multiple logistic regression models were used to characterize independent risk factors for P. vivax infection and disease.
P. vivax prevalence decreased from 23.8% (March–April 2010) to 3.0% (April–May 2013), with no P. falciparum infections diagnosed after March–April 2011. Although migrants from malaria-free areas were at increased risk of malaria, their odds of having P. vivax infection and disease decreased by 2–3% with each year of residence in Amazonia. Several findings indicate that low-density and asymptomatic P. vivax parasitemias may complicate residual malaria elimination in Remansinho: (a) the proportion of subpatent infections (i.e. missed by microscopy) increased from 43.8% to 73.1% as P. vivax transmission declined; (b) most (56.6%) P. vivax infections were asymptomatic and 32.8% of them were both subpatent and asymptomatic; (c) asymptomatic parasite carriers accounted for 54.4% of the total P. vivax biomass in the host population; (d) over 90% subpatent and asymptomatic P. vivax had PCR-detectable gametocytemias; and (e) few (17.0%) asymptomatic and subpatent P. vivax infections that were left untreated progressed to clinical disease over 6 weeks of follow-up and became detectable by routine malaria surveillance.
| Despite decades of control efforts, malaria remains a major public health concern in Brazil. A large proportion of the 243,000 cases diagnosed per year originate from areas of recent colonization in the densely forested Amazon Basin. This population-based longitudinal study addresses the epidemiology of malaria during the early stages of colonization of frontier settlements in Remansinho area, rural Amazonia. We documented a major decline in the prevalence of P. vivax infection, from 23.8% to 3.0%, between March–April 2010 and April–May 2013. Up to 73.1% of the P. vivax infections were missed by microscopy as malaria transmission declined and most (56.6%) of these infections caused no clinical signs or symptoms. Few (17.0%) asymptomatic P. vivax infections that were left untreated eventually progressed to clinical disease, becoming detectable by routine malaria surveillance, over 6 weeks of follow-up. Moreover, nearly all P. vivax infections that were undetected by microscopy had gametocytes, the parasite's blood stages responsible for malaria transmission to mosquito vectors, detected by molecular methods. These findings indicate that apparently healthy carriers of low-density parasitemias, who are often missed by conventional microscopy, contribute significantly to ongoing P. vivax transmission and may further complicate residual malaria elimination in Remansinho and similar endemic settings.
| Malaria is one of the major tropical infectious diseases for which decades of intensive control efforts have met with only partial success in Brazil [1]. With nearly 243,000 slide-confirmed infections, this country contributed 52% of all malaria cases reported in the Region of the Americas and the Caribbean in 2012 [2]. Most transmission in Brazil occurs in open mining enclaves, logging camps and farming settlements across the Amazon Basin, a region that currently accounts for 99.9% of the country-wide malaria burden [3].
Since the early 1970s, official and informal colonization projects in densely forested areas of Amazonia have attracted migrant farmers from the malaria-free South and Southeast regions, originating a number of new frontier agricultural settlements [4], [5]. Initial land clearing for slash-and-burn agriculture and extensive logging can induce major changes in vector biology, by creating or expanding mosquito breeding habitats, as well as in vector species composition, with a marked increase in the abundance of the highly competent local malaria vector Anopheles darlingi [6]–[10]. Not surprisingly, recent frontier settlements, where ongoing deforestation and the immigration of non-immune pioneers favor transmission, constitute malaria hotspots until these communities become more stable and endemicity declines [6].
Plasmodium falciparum and P. vivax infections are widespread across Amazonia, with rare and focal P. malariae transmission [11]–[14]. A clear change has been recently observed in the relative proportion of the two main species. Similar proportions of slide-confirmed infections were due to P. falciparum and P. vivax until 1990, but transmission of the latter species maintained an upward trend while that of P. falciparum declined steadily throughout the next decade [15]. Plasmodium vivax now accounts for 85% of the malaria burden in Brazil [2]. These trends may be explained by factors such as the presence of dormant liver stages (hypnozoites) and the early circulation of sexual stages (gametocytes) in peripheral blood, which may render P. vivax less responsive than P. falciparum to available control strategies based on early diagnosis and treatment of blood-stage infections [16], [17].
Here we describe the epidemiology of malaria and associated risk factors during the early phases of occupation of frontier agricultural settlements in the Amazon Basin of Brazil. We observed a major decline in P. vivax prevalence, with vanishing P. falciparum transmission, over 3 years of malaria surveillance. Risk of both infection and P. vivax-related disease decreased with increasing cumulative exposure to malaria, consistent with anti-parasite and anti-disease immunity being acquired by this population. We discuss the challenges of controlling and eliminating malaria, especially that caused by the resilient parasite P. vivax, in low-endemicity areas where most infections are asymptomatic and parasite densities are often below the detection threshold of conventional microscopy.
Once a sparsely populated rubber tapper settlement (seringal) situated in southern Amazonas state, northwestern Brazil, Remansinho (average population, 260) now comprises five farming settlements (Figure 1). The main settlement is situated along the final 40 km of the Ramal do Remansinho, a 60 km-long unpaved road originating from the BR-364 interstate highway, while the other four are situated along secondary roads (known as Ramal da Linha 1, Ramal da Castanheira, Ramal dos Seringueiros, and Ramal dos Goianos) originating from this main unpaved road (Figure S1). The farming settlements along Ramal da Linha 1 and Ramal da Castanheira were opened in the late 1990s, whereas the colonization of the other areas started only in 2007. Most houses have complete or incomplete wooden walls and thatched roofs; just a few of them have brick walls and are covered with asbestos, cement or zinc shingles. With a typical equatorial humid climate (annual average temperature, 26.4°C), Remansinho receives most rainfall between November and March (annual average, 2,318 mm), but malaria transmission occurs year-round. The main local malaria vector is the highly anthropophilic and exophilic An. darlingi [18]. Most families currently living in Remansinho have resettled from other areas within Amazonia, and are now involved in subsistence agriculture and logging. There is a single government-run health post in Remansinho, which provides free malaria diagnosis and treatment, but a small proportion of locally acquired infections are diagnosed and treated in the nearest village (Nova Califórnia; population, 2,600), situated along the BR-364 highway, about 60 km south of Remansinho (Figure 1). There is no electricity or piped water supply in the area.
A population-based prospective cohort study was initiated in March 2010 to estimate the prevalence and incidence of malaria parasite carriage in Remansinho, by combining microscopy and molecular diagnosis, and to characterize risk factors for malaria infection and clinical disease in the local population. This ongoing study comprises periodic cross-sectional malaria prevalence surveys of the entire population, every four months between March 2010 and March 2011 and every six months thereafter, complemented with clinical and laboratory surveillance of febrile illness episodes between the cross-sectional surveys.
Here we analyze data collected from March 2010 to May 2013. During this period, we enrolled 584 participants belonging to 205 households. Dwellings were geo-localized using a hand-held 12-chanel global positioning system (GPS) receiver (eTrex Personal Navigator, Garmin, Olathe, KS), with a positional accuracy within 15 m. Nearly all (98.8%) study subjects were recruited during house-to-house visits in Ramal do Remansinho (376 or 65.2%), Ramal da Castanheira (85 or 14.7%), Ramal da Linha 1 (57 or 9.9%), Ramal dos Goianos (32 or 5.5%), and Ramal dos Seringueiros (27 or 4.7%); only 7 (1.2%) subjects, who were enrolled at the local health post, had their settlement of origin undetermined.
Each cross-sectional survey comprised a population census and the entire population found during the census was considered eligible to participate in the study. During the first (March–May 2010), 165 inhabitants identified during the census in Ramal do Remansinho and Ramal dos Goianos were invited to participate. Subsequent surveys, which also included subjects living in the other three settlements, were carried out between May and July 2010 (survey 2), October and November 2010 (survey 3), March and April 2011 (survey 4), October and November 2011 (survey 5), April and May 2012 (survey 6), October and November 2012 (survey 7), and April and May 2013 (survey 8). Most surveys were carried out either at the beginning or the end of the rainy season, except for survey 2, which took place during the dry season. Total numbers of subjects present in the study area during each survey are given in Table 1.
A baseline questionnaire was applied to all study participants in March–May, 2010, to collect demographic, health, behavioral and socioeconomic data. Cumulative exposure to malaria was estimated using the duration of residence in Amazonia as a proxy. We used a structured questionnaire [19] to determine the presence and intensity of 13 malaria-related signs and symptoms (fever, chills, sweating, headache, myalgia, arthralgia, abdominal pain, nausea, vomiting, dizziness, cough, dyspnea, and diarrhea) up to seven days prior to the interview. Information on selected household assets, access to utilities, infrastructure, and housing characteristics was used to derive a wealth index, from which socioeconomic status was estimated. We combined discrete (i. e., yes or no) ownership information (for power generator, chainsaw, radio, sofa set, shotgun, bicycle, car, motorcycle, and well) and continuous data (i. e., total number of items, for beds, rooms and bedrooms present in the household, and number of pigs, cattle, chickens, ducks, and horses owned). Principal component analysis, carried out using statistical software STATA 12.1, was used to weight each variable [20]. The first principal component explained 18% of the variability and gave the greatest weights to ownership of beds, number of rooms, number of bedrooms, sofa set, and chickens. Lowest weights were given to ownership of horses, ducks or cattle. The scores were summed to give a wealth index for each household. Wealth indices were then used to stratify households into quartiles in increasing order (first quartile, 25% poorest). A shorter version of the baseline questionnaire was used in all subsequent cross-sectional surveys to update demographic and clinical data.
All inhabitants in the study area aged more than 3 months were invited to contribute either venous (5-ml) or finger-prick blood samples for malaria diagnosis, irrespective of any clinical symptoms, Duffy blood group genotyping, and other laboratory assays, such as hemoglobin measurements and ABO and Rh blood group typing. The participation rates ranged between 96.3% in survey 1 (159 of 165 inhabitants) and 70.3% in survey 5 (204 of 290) (Table 1). Nearly all study participants provided venous blood samples in all but one survey; the exception was survey 3, during which finger-prick capillary blood was preferentially collected from all participants for logistic reasons. Reasons for not providing blood samples included temporary absence from the study area, age below 3 months, inability to perform venous puncture, and refusal to participate. Given the high mobility of the study population, only 21 subjects (3.6% of the study population) contributed blood samples in all cross-sectional surveys; 529 subjects (90.6%) participated in two or more surveys. All study participants, either symptomatic or not, who provided either venous or finger-prick blood samples during cross-sectional surveys and had malaria diagnosis confirmed by onsite microscopy were treated according to the malaria therapy guidelines published by the Ministry of Health of Brazil in 2010 [21]. Briefly, P. vivax infections were treated with chloroquine (total dose, 25 mg of base/kg over 3 days) and primaquine (0.5 mg of base/kg/day for 7 days), while P. falciparum infections were treated with a fixed-dose combination of artemether (2–4 mg/kg/day) and lumefantrine (12–24 mg/kg/day) for 3 days. Infections that were missed by onsite microscopy but later confirmed by polymerase chain reaction (PCR) were left untreated because the results of molecular diagnosis were not available at the time of the cross-sectional surveys.
To quantify clinical malaria episodes diagnosed between the cross-sectional surveys, we examined all records of slide-confirmed infections diagnosed between March 2010 and November 2013 at the government-run health posts in Remansinho and in the nearest village, Nova Califórnia. Local malaria control personnel performed both active and passive detection of febrile cases during the study period. Blood samples were collected and examined for malaria parasites whenever febrile subjects visited the health posts in Remansinho or Nova Califórnia or were found during monthly house-to-house visits carried out by field health workers in Remansinho. This strategy is assumed to detect virtually all clinical malaria episodes in cohort subjects between the cross-sectional surveys, since there are no other public or private facilities providing laboratory diagnosis of malaria in the area. Microscopic diagnosis is required to obtain antimalarial drugs in Brazil, which are distributed free of charge by the Ministry of Health and cannot be purchased in local drugstores.
Laboratory diagnosis of malaria was based on microscopic examination of thick smears and PCR. A total of 1,541 thick blood smears were stained with Giemsa in our field laboratory in Acrelândia (120 km southwest of Remansinho). At least 100 fields were examined for malaria parasites, under 1000× magnification, by two experienced microscopists, before a slide was declared negative. We additionally used quantitative real-time PCR (qPCR) that target the 18S rRNA genes [22] to detect and quantify P. vivax and P. falciparum blood stages in 1,476 clinical samples (Methods S1). Because microscopy is poorly sensitive for detecting circulating gametocytes [23], we used a quantitative reverse transcriptase PCR (qRT-PCR) that targets pvs25 gene transcripts [24], [25] to detect and quantify mature gametocytes in 55 laboratory-confirmed P. vivax infections diagnosed during cross-sectional surveys 4, 5, and 6 (Methods S1).
Since co-infection with multiple parasite clones has been suggested to either increase or reduce the risk of clinical falciparum malaria, we sought to determine whether the presence of multiple-clone P. vivax infections was associated with malaria-related morbidity. To this end, we amplified two highly polymorphic single-copy markers, msp1F1 (a variable domain of the merozoite surface protein-1 gene) [26] and MS16 (a P. vivax-specific microsatellite DNA marker with degenerate trinucleotide repeats) [27], using the nested PCR protocols of Koepfli and colleagues [28]. DNA samples from 85 qPCR-confirmed P. vivax infections (all of them isolated from venous blood samples) were tested for the presence of multiple clones; 47 were from asymptomatic and 38 from symptomatic parasite carriers. PCR products were analyzed by capillary electrophoresis on an automated DNA sequencer ABI 3500 (Applied Biosystems), and their lengths (in bp) and relative abundance (peak heights in electropherograms) were determined using the commercially available GeneMapper 4.1 (Applied Biosystems) software. The minimal detectable peak height was set to 200 arbitrary fluorescence units. We scored two alleles at a locus when the minor peak was >33% the height of the predominant peak. Plasmodium vivax infections were considered to contain multiple clones if one or both loci showed more than one allele.
Since Duffy blood group polymorphisms modulate the ability of P. vivax merozoites to invade human red blood cells (reviewed by [29]), we used TaqMan assays (Applied Biosystems) to genotype two major Duffy polymorphisms: the T-33C substitution in the red blood cell-specific GATA1 transcription factor binding motif (rs2814778), which suppresses Duffy expression on the erythrocyte surface (Fy phenotype, associated with FY*BES allele homozygozity), and the G125A polymorphism (rs12075), which defines the FY*B (wild-type) and FY*A (mutated) alleles associated with the Fyb and Fya phenotypes, respectively. The primers and probes (labelled with VIC and FAM) were designed and synthesized by Applied Biosystems (assay ID, C__15769614_10 and C__2493442_10) [30]. We used a Step One Plus Real-Time PCR System (Applied Biosystems) for genotyping, with a template denaturation step at 95°C for 10 min, followed by 40 cycles of 15 sec at 95°C and 1 min at 60°C, with a final step at 60°C for 30 sec. DNA samples from 487 study participants were genotyped.
A laboratory-confirmed malarial infection was defined as any episode of parasitemia detected by thick-smear microscopy, qPCR, or both. Subpatent or submicroscopic infections were defined as infections confirmed by qPCR but missed by microscopy. We defined clinical malaria as a laboratory-confirmed infection, irrespective of the parasite density, diagnosed in a subject reporting one or more of the 13 signs and symptoms investigated, at or up to seven days before the interview. No attempt was made to calculate pyrogenic thresholds of parasitemias in our heterogeneous group of study participants. Subjects with laboratory-confirmed infection, irrespective of the parasite density, who reported no signs or symptoms at or up to seven days prior to the interview, were classified as asymptomatic carriers of malaria parasites.
A database was created with SPSS 17.0 software (SPSS Inc., Chicago, IL) and all subsequent analyses were performed with R statistical software [31]. For the purposes of explanatory data analysis, proportions were compared using standard χ2, Mantel-Haenzel χ2 for stratified data, or χ2 tests for linear trends. Correlations between parasite densities, which had an overdispersed distribution in the population, and other continuous variables were evaluated using the nonparametric Spearman correlation. Median parasitemias were compared with the nonparametric Mann-Whitney U test. Statistical significance was defined at the 5% level and 95% confidence intervals (CI) were estimated whenever appropriate.
Separate regression models were built to describe independent associations between potential risk factors and two outcomes: (a) P. vivax infection and (b) clinical (i. e., symptomatic) vivax malaria. Due to the small number of P. falciparum infections detected in the community no attempt was made to characterize risk factors for infection with this species. Dependent variables were assumed to follow a binomial distribution with a logit link function, being fitted with a logistic regression. We considered the nested structure of the data, intrinsic to the study design, when building regression models; we have repeated observations (up to 8 observations over 3 years of study; grouping variable, “survey”) nested within subjects (grouping variable, “individual”) who are clustered within households (grouping variable, “household”). This clustered sampling scheme introduces dependency among observations that can affect model parameter estimates. Consequently, we used mixed-effects regression models that include the grouping variables as random factors to handle nested observations.
Our modeling strategy further considered the hierarchical levels of independent variables (Methods S1). The effects of distal determinants, such as demographic, social and environmental factors, on malaria risk are often not direct, but mediated by more proximate determinants, such as occupational and behavioral factors [32]. Variables within each level of determination were introduced in the model in a stepwise approach, and only those that were associated with the outcome at a significance level of at least 20% were retained. Most subjects with missing observations were excluded from the final model, except those with missing values for the following four variables: Duffy genotype, wealth index, whether bedroom windows were left open at night, and main occupation. These were maintained in the model by creating a new missing-value category. All models were adjusted for the timing of the survey (months elapsed since the beginning of the study in March 2010). Three variables in the model were time-dependent: age, years of residence in Amazonia, and timing of the survey. The final models comprised 1,242 observations from 442 individuals grouped into 159 households (outcome: P. vivax infection), and 1,237 observations from 438 individuals grouped into 158 households (outcome: vivax malaria).
Alternative logistic models, which excluded Duffy-negative subjects (88 observations from 31 subjects), examined the association between Duffy-positive genotypes (FY*AFY*BES, FY*AFY*A, FY*AFY*B, FY*BFY*BE, and FY*BFY*B) and risk of P. vivax infection and vivax malaria. To account for the hypothesis that age at the beginning of exposure to malaria affects the rate at which antimalarial immunity is acquired by migrants [33], we further tested for an interaction between age and years of residence in Amazonia. In addition, we fitted mixed-effects Poisson regression models to the data, but the random-effects variances associated with the estimates were substantially higher than those obtained with the logistic models described above. As a consequence, here we only present results derived from the logistic regression analysis.
In addition, we used a mixed-effects Cox proportional hazards model [34] to compare the risk of slide-positive vivax malaria between the surveys in two sub-cohorts of asymptomatic subjects: (a) carriers of subpatent P. vivax infections at baseline that were left untreated and (b) control subjects who were parasite-negative at baseline by both microscopy and qPCR. Subjects who were symptomatic but parasite-negative at baseline were excluded from the uninfected sub-cohort because they might harbor ongoing low-grade infections, causing malaria-related symptoms, which were missed by our laboratory methods. At each survey, eligible study participants were assigned to either sub-cohort and followed up until the next survey at which their clinical and infection status was reassessed. Time at risk was defined as either the interval between two consecutive surveys in which the subjects participated (the first survey in the pair was defined as the baseline survey) or the interval between the baseline survey and the date when subjects left the study, whatever came first. Analysis was adjusted for subjects' age (stratified as <15 years and ≥15 years), Duffy blood group negativity, and years of residence in Amazonia. The clustering of repeated observations within individuals was modeled as a random effect [34].
As required for all observational studies published by PLoS Neglected Tropical Diseases, this article includes the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) checklist to document its compliance with STROBE guidelines (Checklist S1).
Study protocols were approved in early 2010 by the Institutional Review Board of the University Hospital of the University of São Paulo (1025/10) and by the National Human Research Ethics Committee of the Ministry of Health of Brazil (551/2010). The ethical clearance has been renewed annually by the Institutional Review Board of the University Hospital of the University of São Paulo. Written informed consent was obtained from all study participants or their parents/guardians.
Of 584 people living in Remansinho who participated in at least one cross-sectional survey, 333 (57.0%) were male and 251 (43.0%) were female, with a median age of 23.0 years. Nearly all (94.3%) adult subjects aged more than 18 years were migrants, 42.2% of them originating from malaria-free areas outside Amazonia. Only 31 subjects (6.4%) were homozygous FY*BES carriers, with the P. vivax-refractory Duffy-negative (Fy) phenotype; 127 (26.1%) had the Fya phenotype (70 FY*AFY*BES heterozygotes and 57 FY*A FY*A homozygotes), 142 (29.2%) had the FyaFyb phenotype (FY*A FY*B heterozygotes), and 187 (38.4%) had the Fyb phenotype (91 FY*BFY*BES heterozygotes and 96 FY*B FY*B homozygotes).
Polyethylene bed-nets treated with 2% permethrin (Olyset Net, Sumitomo Chemical, London, United Kingdom) were distributed, free of charge, to the entire study population in August 2012, as a component of malaria control activities in Brazilian Amazonia. In October–November 2012 (survey 7), 74.4% of the study participants reported having slept the previous night under an Olyset net; the corresponding figure for April–May 2013 (survey 8) was 84.5%. No other insecticide-treated bed nets were available in the community.
A total of 1,541 blood samples were examined for malaria parasites by microscopy, qPCR, or both. Of these, 141 (9.1%) were positive (by one or both methods) for P. vivax, 40 (2.6%) for P. falciparum and 10 (0.6%) for both species. Over the entire study period, 191 (12.4%) samples examined tested positive for malaria parasites; 10 P. vivax and 2 P. falciparum infections were only diagnosed by microscopy, since DNA samples were not available for qPCR or qPCR yielded negative results. In addition, 61.8% of all infections diagnosed by qPCR, regardless of the infecting species, and 49.6% of the qPCR-confirmed single-species P. vivax infections, were missed by conventional microscopy and thus defined as subpatent. The last P. falciparum infections in Remansinho were diagnosed (by qPCR only) in March–April 2011.
These figures, however, changed over time. The numbers of malaria infections, either symptomatic or not, diagnosed by conventional microscopy and qPCR in each cross-sectional survey are shown in Table 1. The proportions of qPCR-confirmed single-species P. vivax infections that were subpatent varied widely across surveys, ranging from 73.1% in the surveys with the lowest P. vivax prevalence rates (surveys 4, 6, 7, and 8 combined; 26 qPCR-confirmed infections) to 43.8% in those with the highest prevalence rates (surveys 1, 2, 3, and 5 combined; 105 qPCR-confirmed infections; Yates' corrected χ2 = 6.02, 1 degree of freedom [df], P = 0.014). The numbers of P. falciparum and mixed-species infections were too small for a similar comparison. Microscopy thus had a better diagnostic performance for vivax malaria when overall parasite prevalence rates were higher, consistent with a recent meta-analysis of P. falciparum data showing lower proportions of submicroscopic infections in areas with greater malaria transmission [35].
Overall, 17.1% of the study subjects (ranging between 12.3% in survey 6 and 39.6% in survey 1) interviewed during the cross-sectional surveys reported one or more malaria-related signs and symptoms up to seven days prior to the interview (Table 1). However, reported clinical signs and symptoms were neither sensitive nor specific for malaria diagnosis. On the one hand, almost two thirds (64.5%) of all qPCR-confirmed malaria infections by any species, and 56.6% of those due to P. vivax, were asymptomatic; on the other hand, only 26.7% of subjects reporting symptoms had a malaria infection (by any species) confirmed by microscopy, qPCR, or both. All carriers of mixed-species infections (all of them confirmed by qPCR but missed by microscopy) were asymptomatic (Table 1).
Most P. vivax-infected subjects harbored few parasites, with densities estimated by qPCR on 129 samples ranging between 2.1 and 38,390 parasites/µL (median, 49.1 parasites/µL; interquartile range, 10.0–483.1 parasites/µL; data were missing for 2 qPCR-confirmed infections). We found no evidence for decreasing P. vivax densities with increasing cumulative exposure to malaria in this population. In fact, individual P. vivax parasitemias did not show a negative correlation with the subjects' length of residence in Amazonia, a proxy of cumulative exposure to malaria (Spearman correlation coefficient rs = −0.046, P = 0.600), or with their age (rs = −0.068, P = 0.427). We next tested whether differences in the diagnostic sensitivity of conventional microscopy across cross-sectional surveys might be explained by higher average parasite densities found at times of increased malaria transmission [35]. Parasitemias appeared slightly higher in qPCR-positive samples (P. vivax only) obtained during surveys 1,2,3 and 5 (high prevalence), with a median of 55.7 parasites/µL (interquartile range, 10.4–597.6 parasites/µL; n = 103), than in those obtained during surveys 4,6,7, and 8 (low prevalence), with a median of 19.8 parasites/µL (interquartile range, 5.8–65.4 parasites/µL; n = 26), although the difference did not reach statistical significance (Mann-Whitney U test, P = 0.057).
The proportion of symptomatic P. vivax infections correlated positively with increasing parasite density (χ2 for trend = 7.99, 1 df, P<0.005). Only 30.6% of the subjects carrying less than 10 parasites/µL, but 73.9% of those carrying more than 1,000 parasites/µL, reported one or more malaria-related symptoms (Figure 2). Consistent with previous observations from Amazonia [36], [37], more than half (53.9%) of the asymptomatic infections with this species confirmed by qPCR were missed by conventional microscopy (Table 1). Overall, 32.8% of the 131 single-species, qPCR-confirmed P. vivax infections for which complete data were available were both subpatent and asymptomatic (Figure 3). Only one P. vivax infection was diagnosed by qPCR, but missed by conventional microscopy, among 88 samples collected from Duffy-negative study participants during the 8 cross-sectional surveys. The only reported symptom during this subpatent P. vivax infection in a Duffy-negative subject was a chronic myalgia; parasite density was very low (9.9 parasites/µL of blood).
To estimate the relative contribution of asymptomatic parasite carriage to the total P. vivax biomass in the host population, we summed up all individual qPCR-derived P. vivax densities and calculated the fraction corresponding to asymptomatic infections. Assuming that, on average, asymptomatic and symptomatic subjects have similar whole blood volumes, we concluded that most (54.4%) P. vivax blood stages circulating in Remansinho at the time of the surveys were found in apparently healthy subjects who were unlikely to have their infection diagnosed through case detection strategies targeting only febrile subjects. The number of P. falciparum and mixed-species infections was too small for similar analyses.
Next, we examined whether P. vivax gametocyte carriage was similarly frequent in symptomatic and asymptomatic infections. We detected pvs25 gene transcripts, consistent with mature P. vivax gametocytes circulating in the bloodstream, in all 32 symptomatic infections, and in 21 of 23 (91.3%) asymptomatic infections from which cryopreserved blood samples were available for RNA extraction. Interestingly, 25 of 27 (92.6%) subjects with subpatent P. vivax parasitemia, and all 28 subjects with patent infection, had pvs25 gene transcripts detected by qRT-PCR. Therefore, qRT-PCR failed to detect pvs25 transcripts in only 2 (3.6%) of 55 samples tested (Figure 3), both of them collected from asymptomatic carriers of low parasitemias (6.3 and 11.0 parasites/µL). Not surprisingly, the number of pvs25 transcripts per µL of blood, measured by qRT-PCR, correlated positively with the qPCR-derived overall parasite density (rs = 0.445, P<0.0001).
The mixed-effects logistic regression model showed that the risk of P. vivax infection decreased with increasing cumulative exposure to malaria, consistent with anti-parasite immunity being acquired in this population (Table 2). Each additional year of residence in Amazonia decreased the average odds of being infected by 2% (Figure 4). There was no significant interaction between age and years of residence in Amazonia (P = 0.9008), suggesting that the subjects' age when exposure started did not affect, in this migrant population, the rate of decline in P. vivax infection risk with increasing time of residence in Amazonia. Calendar time was also a major determinant of infection risk; each month elapsed since March 2010 was associated with a 7% decrease in the odds of being infected (Table 2). Moreover, the grouping variable “survey” accounted for 99.9% of the random effect variance in the mixed-effects model, with minor contribution of individual- and household-level grouping.
Interestingly, adjusting for more proximate determinants affected the association between age and risk of infection with P. vivax in multivariate models. Age under 15 years was a protective factor of borderline significance (partially adjusted OR = 0.616; 95% CI, 0.33–0.95, P = 0.057) in the first model, which also adjusted for sex, years of residence in Amazonia, Duffy blood group genotype, months elapsed since the beginning of the study, and wealth index quartiles. However, after adjusting for main occupation, the effect of age on infection risk became non-significant (fully adjusted OR = 1.169; 95% CI, 0.63–2.19, P = 0.624). These results indicate that young age per se is not protective, but young subjects are less likely to engage in activities such as logging and fishing in the fringes of the rain forest, which are potentially associated with increased risk of infection (Table 2).
Not surprisingly, Duffy-negativity emerged as a protective factor against P. vivax infection in this community (Table 2). However, additional logistic regression models including only Duffy-positive subjects showed that, compared to FY*A FY*B heterozygotes, neither FY*A FY*BES heterozygotes (OR = 0.864; 95% CI, 0.43–1.73) and FY*A FY*A homozygotes (OR = 0.921; 95% CI, 0.48–1.75) were protected against P. vivax infection, nor FY*B FY*BES heterozygotes (OR = 1.226; 95%, 0.69–2.17) and FY*B FY*B homozygotes (OR = 0.588; 95% CI, 0.32–1.09) were at increased risk of infection. These results are consistent with a protective role of FY*BES heterozygosity, but not of FY*A allele carriage, against P. vivax infection in this population.
The risk of clinical P. vivax malaria decreased with increasing cumulative exposure to malaria (Table 2); each additional year of residence in Amazonia decreased the odds of having vivax malaria by 3%, again with no significant interaction between age and length of residence in Amazonia (P = 0.863). These findings are consistent with similar exposure-dependent rates of acquisition of anti-parasite and anti-disease immunity in this community. Calendar time was the only other major determinant of malaria risk; each month elapsed since the beginning of the study was associated with an 8% decrease in the odds of having clinical vivax malaria (Table 2). Due to the small sample size, Duffy-negativity emerged as a protective factor of borderline significance (OR = 0.16, 95% CI, 0.02–1.29, P = 0.084) against clinical vivax malaria.
In Brazil, malaria is only treated if blood smear microscopy is positive; subpatent malaria parasitemia as determined with qPCR is not accepted as the basis for treatment. Of 53 asymptomatic subpatent P. vivax infections diagnosed at baseline, 9 (17.0%) progressed to clinical malaria over the following 6 weeks, being diagnosed by onsite microscopy and treated (Figure 5). During this 6-week period, only 2.5% of the subjects in the uninfected cohort experienced an episode of slide-confirmed vivax malaria, but at the end of the follow-up period similar proportions of subjects in each sub-cohort had experienced vivax malaria episodes confirmed by microscopy (Figure 5). A Cox proportional hazards model revealed no significant difference, between the two sub-cohorts, in overall risk of vivax malaria episodes, after controlling for potential confounders (hazard ratio = 1.07; 95% CI, 0.52–2.22, P = 0.840). Most subpatent asymptomatic infections cleared spontaneously (or, at least, became undetectable by qPCR), since only 5 of 44 (11.4%) carriers who remained untreated were again P. vivax-positive in the next survey. Therefore, few asymptomatic and subpatent P. vivax infections eventually became patent and symptomatic (and therefore detectable by routine malaria surveillance) over the following weeks. We conclude that untreated, low-density, and asymptomatic P. vivax parasitemias may persist for several weeks without progressing to clinical disease, and thus constitute a major infectious reservoir for continued transmission in the community.
By typing two highly polymorphic markers, we found more than one genetically distinct clone in 25 of 85 (29.4%) P. vivax infections analyzed. Although multiple-clone infections were more frequent in symptomatic (13 of 38, 34.1%) than asymptomatic (12 of 47, 25.5%) carriers, this difference did not reach statistical significance (Yates' corrected χ2 = 0.762, 1 df, P = 0.382). Because average P. vivax densities were lower in asymptomatic infections and detecting minority clones may be more difficult in samples with low-level parasitemias, we re-analyzed the data after stratifying parasite densities into quartiles. Again, stratified analysis yielded negative results (Mantel-Haenzel χ2 = 0.004, 1 df, P = 0.991). Therefore there was no observable association between multiplicity of P. vivax infection and the presence of symptoms in this community.
This longitudinal study in newly opened frontier settlements provides further evidence that carriers of low-density parasitemias, who are often missed by conventional microscopy, contribute significantly to ongoing P. vivax transmission in rural Amazonia. Results from this and other studies in Amazonia [12]–[14], [36], [37] challenge the often persisting view that subjects in low malaria transmission settings are unlikely to harbor low parasitemias, due to the lack of acquired immunity. To the contrary, average parasite densities decreased, with higher proportions of P. vivax infections being missed by microscopy, as malaria prevalence decreased in the community. Interestingly, our findings for P. vivax are consistent with a recent meta-analysis of 106 P. falciparum prevalence studies worldwide that combined microscopy and molecular methods [35]. Because the risk of P. vivax infection (confirmed by microscopy, qPCR, or both) correlated negatively with cumulative exposure to malaria, we suggest that our study population has developed over time some degree of anti-parasite immunity, in line with recent findings from traditional riverine communities in Amazonia [13], [37]. Finally, we show that nearly all subpatent blood-stage P. vivax infections comprise mature gametocytes detected by a highly sensitive molecular technique [24]. We thus conclude that subpatent infections constitute a major P. vivax reservoir in rural Amazonia and possibly in other low-transmission settings.
Our findings also challenge classical views regarding asymptomatic infections in low-endemicity populations. Prior to the molecular diagnosis era, nearly all laboratory-confirmed episodes of malarial infection, even those with low parasite densities, were thought to elicit clinical disease in pioneer settlements across the Amazon Basin [38]–[40]. More recent surveys, however, demonstrated that subclinical infections are common in agricultural settlements [12], [14] and traditional riverine communities [13], [36], [37], [41], but most of them are missed by microscopy. Interestingly, the high proportion of infections found to be asymptomatic in the present study must be interpreted as a conservative estimate. We may have misclassified some episodes of parasite carriage in subjects reporting any of the 13 symptoms investigated, which may or may not be caused by the current infection, as symptomatic malaria infections, overestimating the proportion of symptomatic infections. Not surprisingly, however, we found very low P. vivax densities in most subclinical infections in Remansinho. Conventional microscopy missed 54% of them, suggesting that previous microscopy-based studies failed to detect asymptomatic parasite carriage in rural Amazonians because they missed a large proportion of low-density infections. Mathematical models identified asymptomatic infections as a crucial target for P. falciparum malaria eradication efforts in Africa [42], but no similar analyses are available for other endemic areas and other human malaria parasite species [43]. The following findings argue for a major role of asymptomatic infections in maintaining ongoing P. vivax transmission in Remansinho: (a) apparently healthy subjects accounted for half of the total P. vivax biomass found in the local population, (b) nearly all asymptomatic infections comprised mature gametocytes, and (c) few untreated asymptomatic infections became symptomatic (and thus detectable by routine surveillance) over the next few weeks of follow-up. We were unable to measure the average duration of untreated, asymptomatic infections in our population; there is a recent estimate of 194 days of duration for untreated P. falciparum infections in Ghana [44], but no comparable estimate is currently available for P. vivax. Specific studies to quantify the transmissibility of subpatent parasitemia to vector mosquitos via direct and membrane feeding assays are ongoing (JMV and colleagues, unpublished data).
Who are at risk of malaria in Remansinho? Migrants from malaria-free areas (54.5% of the adults in the community) constitute a major risk group, with each year of residence in Amazonia decreasing their risk of P. vivax infection and clinical vivax malaria by 2–3%. In some Amazonian communities, malaria has been associated with forest-related activities such as logging, fishing and mining, which typically involve young male adults [12], [39], [45], [46]. However, we show that housekeeping and forest-related activities were associated with similar risks for infection and disease in Remansinho. We hypothesize that nearly all adolescents and adults of both genders engage to some extent in farming activities, especially harvesting, in the forest fringes close to their dwellings, although only young males are often involved in logging and land clearing in more densely forested areas. We are currently using high-resolution satellite images to measure the distance between dwellings and forest fringes to further explore the association between proximity to the forest environment and risk of malaria in Remansinho. Interestingly, malaria transmission appears to be relatively homogeneous across all settlements in the area, equally affecting the poorest and least poor people of both sexes, with no differences in risk according to main house characteristics. Whether the vectorial capacity of An. darlingi is spatially homogeneous is a key question to be answered by ongoing vector biology studies in this site.
Detection of gametocytes, through pvs25 gene transcripts, in nearly all qPCR-confirmed P. vivax infections tested is somewhat surprising, since recent studies have found much lower proportions of gametocyte-positive infections in Southeast Asia [47], [48] and Papua New Guinea [49]. Since gametocytes comprise only 2% of circulating blood stages [50], microscopists are likely to miss gametocytes in population-based studies where low-density infections are often sampled [23]. Furthermore, we argue that even molecular methods may be poorly sensitive if suboptimal techniques for sample storage and RNA extraction are used under field conditions. For RNA isolation, we cryopreserved venous blood samples at −70°C or in liquid nitrogen a few hours after collection, since our previous attempts to amplify pvs25 transcripts from RNA isolated from classic FTA microcards (Wathman), QIAcards (Qiagen), 903 protein saver cards (Whatman), and 3MM filter papers (Whatman) impregnated with P. vivax-infected blood and kept at ambient temperature had all failed [24]. Storing filter papers impregnated with blood in TRizol reagent (Qiagen) may improve RNA yield, but almost two thirds of the bloodspots from PCR-confirmed P. vivax infections tested by Wampfler and colleagues [49] were negative for pvs25 transcripts by TaqMan assays, despite previous TRizol reagent treatment.
Long-term asymptomatic carriage of P. falciparum has been suggested to protect against subsequent malaria-related disease in Africa [51], [52], possibly by reducing the risk of superinfection with more virulent strains. An explanation for this finding is premunition, originally defined by Sergent and Parrot (1935) as the protection against new infections resulting from immune responses to the existing infection [53]. Alternatively, ongoing blood-stage infection might arrest the development of subsequently inoculated sporozoites in the liver. Such an inhibition of superinfection appears to be mediated by the iron regulatory hormone hepcidin, produced in response to blood-stage parasitemia [54]. However, an opposite effect (i.e., increased risk of subsequent disease in asymptomatic P. falciparum carriers) has also been described, suggesting that a proportion of asymptomatic infections will eventually reach the host's pyrogenic threshold [55]. Here we found no significant association between asymptomatic carriage of low-density P. vivax infection and protection from subsequent malaria morbidity, suggesting that treating individuals with asymptomatic P. vivax infections would not render them more vulnerable to clinical malaria over the next few weeks or months.
Although we have identified challenges for malaria control that are not currently addressed by routine surveillance, malaria transmission in Remansinho has declined dramatically over 3 years of surveillance, and P. falciparum was found only during the first four surveys. Factors that may have contributed to this decline include drastic environmental changes resulting from logging and land clearing for farming, variation in climate, the widespread use of insecticide-treated bed-nets since August 2012, and the implementation of research activities in the area.
To address the first two hypotheses, we are now analyzing high-resolution satellite images to track environmental changes over time. Consistent with the third hypothesis, two studies have provided evidence that insecticide-treated bed-nets are effective for malaria control in Amazonia. The first was a case-control study in Colombia that showed more than 50% reduction in malaria, relative to no net use, although the advantage of impregnated over non-impregnated nets was not statistically significant [56]. The second study, a randomized trial of lambdacyhalothrin- versus placebo-treated nets in the Amazonas State of Venezuela, showed a protective efficacy of 55% [57]. Whether insecticide-tread bed-nets alone can reduce malaria incidence rates throughout the Amazon Basin remains uncertain, mostly due to the highly variable biting behavior of An. darlingi across the region [58], with strong evidence of significant blood-fed and exophilic host-seeking behavior [59]–[61]. In addition, the decline in transmission in Remansinho preceded the distribution of bednets. Finally, the presence of a research team continuously working in the area for over 3 years may affect positively both diagnostic and treatment practices. The external slide revision routinely carried out by our team provides an example of intervention that may have enhanced the diagnostic skills of local microscopists. Moreover, active case detection during 8 consecutive surveys allowed for the early diagnosis and prompt treatment of several slide-positive asymptomatic infections that would have been missed by routine passive surveillance.
Eliminating residual foci when malaria is nearly disappearing, but remains entrenched in a few hotspots, is the next major goal in Remansinho and many other similar endemic settings. Case detection strategies in areas approaching malaria elimination often target only subjects presenting with fever or with a history of recent fever, who are screened for malaria parasites by conventional microscopy or rapid diagnostic tests (RDT) and receive prompt antimalarial treatment if found to be infected [62]. These strategies overlook asymptomatic infections that might be detected by periodic cross-sectional surveys of the entire population at risk [63], as we did in Remansinho. Nevertheless, the cost-effectiveness of mass blood surveys for detecting and treating these residual infections decreases proportionally as malaria transmission declines, since: (a) large populations must be screened to diagnose relatively few asymptomatic carriers, and (b) diagnostic techniques available for large-scale use, such as microscopy and RDT, are not sensitive enough to detect low-grade infections that are typical of residual malaria settings [64]. As an alternative, we are currently testing a reactive case detection strategy that has been tailored for the relapsing parasite P. vivax to detect new infections in the neighborhood of malaria cases diagnosed by routine surveillance in frontier settlements similar to Remansinho. Evaluating this and other strategies of active surveillance to cope with asymptomatic infections in residual P. vivax foci is a top research priority in the context of current malaria elimination efforts worldwide.
|
10.1371/journal.pmed.1002146 | Circulating Apolipoprotein E Concentration and Cardiovascular Disease Risk: Meta-analysis of Results from Three Studies | The association of APOE genotype with circulating apolipoprotein E (ApoE) concentration and cardiovascular disease (CVD) risk is well established. However, the relationship of circulating ApoE concentration and CVD has received little attention.
To address this, we measured circulating ApoE concentration in 9,587 individuals (with 1,413 CVD events) from three studies with incident CVD events: two population-based studies, the English Longitudinal Study of Ageing (ELSA) and the men-only Northwick Park Heart Study II (NPHSII), and a nested sub-study of the Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT). We examined the association of circulating ApoE with cardiovascular risk factors in the two population-based studies (ELSA and NPHSII) and the relationship between ApoE concentration and coronary heart disease and stroke in all three studies. Analyses were carried out within study, and, where appropriate, pooled effect estimates were derived using meta-analysis. In the population-based samples, circulating ApoE was associated with systolic blood pressure (correlation coefficient 0.08, p < 0.001, in both ELSA and NPHSII), total cholesterol (correlation coefficient 0.46 and 0.34 in ELSA and NPHSII, respectively; both p < 0.001), low-density lipoprotein cholesterol (correlation coefficient 0.30 and 0.14, respectively; both p < 0.001), high-density lipoprotein (correlation coefficient 0.16 and −0.14, respectively; both p < 0.001), and triglycerides (correlation coefficient 0.43 and 0.46, respectivly; both p < 0.001). In NPHSII, ApoE concentration was additionally associated with apolipoprotein B (correlation coefficient 0.13, p = 0.001) and lipoprotein(a) (correlation coefficient −0.11, p < 0.001). In the pooled analysis of ASCOT, ELSA, and NPHSII, there was no association of ApoE with CVD events; the odds ratio (OR) for CVD events per 1-standard-deviation higher ApoE concentration was 1.02 (95% CI 0.96, 1.09). After adjustment for cardiovascular risk factors, the OR for CVD per 1-standard-deviation higher ApoE concentration was 0.97 (95% CI 0.82, 1.15). Limitations of these analyses include a polyclonal method of ApoE measurement, rather than isoform-specific measurement, a moderate sample size (although larger than any other study to our knowledge and with a long lag between ApoE measures), and CVD events that may attenuate an effect.
In the largest study to date on this question, we found no evidence of an association of circulating ApoE concentration with CVD events. The established association of APOE genotype with CVD events may be explained by isoform-specific functions as well as other mechanisms, rather than circulating concentrations of ApoE.
| Heart attacks and strokes, jointly referred to as cardiovascular disease (CVD), are the most common causes of death worldwide. Identifying risk factors that can be used to predict these events can be valuable for prevention, and identifying molecular pathways involved in risk can be useful for the development of preventive drugs.
ApoE is a circulating protein that binds to and may regulate circulating lipoproteins, which are proteins that combine with and transport lipids and fat in the bloodstream. ApoE has several genetic variants in the human population; in previous studies, certain ApoE variants (genotypes) have been shown to be associated with greater risk of CVD.
One hypothesis is that the amount of ApoE protein circulating in the body, which has been shown to vary by ApoE genotype, mediates the risk of CVD.
We measured ApoE concentrations directly in ~10,000 individuals, ~1,400 of whom had CVD events. We did not find any association of circulating ApoE with CVD, even after adjusting for other CVD-related factors that might obscure an association.
We also placed ApoE concentration in the context of clinical risk equations that are used to assess risk in patients, and found that it did not improve risk prediction.
This study demonstrates that ApoE genotypes that confer risk of CVD might do so via protein function rather than protein concentration in the blood.
ApoE may still be a drug target for prevention of CVD.
| Apolipoprotein E (ApoE) is a 34-kDa liver-derived multifunctional protein found associated with triglyceride-rich chylomicrons and very low density lipoproteins (VLDLs), their remnants, and a subset of high-density lipoprotein particles [1,2]. One of ApoE’s major physiological roles is in lipid metabolism; ApoE mediates high-affinity binding of ApoE-containing lipoproteins to the low-density lipoprotein receptor (LDL-R) and LDL receptor related protein 1 (LRP1), facilitating clearance of triglyceride-rich lipoproteins from the circulation. It is this mechanism that is thought to confer protection from atherogenesis. ApoE is a polymorphic protein with three major circulating isoforms, E2, E3, and E4 [3], and lipid metabolism is isoform dependent. Isoforms are determined by a combination of two common non-synonymous single nucleotide polymorphisms (SNPs rs7412 and rs429358) in exon 4 of the APOE gene on Chromosome 19 [4]. The most commonly observed, and the reference, isoform is E3. E3 has a cysteine residue at 112 and an arginine residue at 158, and is present in ~79% of the population. E4 (rs429358), the next most commonly encountered isoform (~14%), has an arginine residue substituting cysteine at 112. Finally, E2 (rs7412) is present at a frequency of ~7% and has a cysteine substituting arginine at residue 158. The resultant six common genotypes, in order of observed frequency, are ε3/ε3, ε3/ε4, ε2/ε3 ε2/ε4, ε4/ε4, and ε2/ε2. The most common genotype group is ε3/ε3, which serves as the reference category.
Very low circulating concentration of ApoE in humans is associated with early onset atherosclerosis [5]. Mice in which the APOE gene is deleted are prone to developing atherosclerosis [6]. Together, these findings have contributed to the view that circulating ApoE may have anti-atherogenic properties. Large population-based genetic studies in humans support this view. Carriage of ε2 is associated with higher circulating concentrations of ApoE [7], lower circulating concentrations of low-density lipoprotein cholesterol (LDL-C) [8], and a lower risk of cardiovascular disease (CVD) events [8]. In contrast, carriage of ε4 is associated with directionally opposite changes in these markers and a higher risk of CVD. It is therefore plausible that circulating concentrations of ApoE are causally associated with CVD, and, as such, ApoE could be a useful biological marker of atherosclerotic risk or a potential drug target.
To date, few studies have reported on the association of circulating ApoE concentration with CVD events (i.e., stroke and coronary heart disease [CHD]). In one study, higher circulating concentrations of ApoE were associated with a higher risk of incident CVD [9]; this contrasts with the inverse association that might be expected from animal and prior human studies. In the same population setting, a related study with a focus on stroke also found an increase in ApoE level associated with an increased risk of stroke [10]. A third study, the largest (total n = 2,951), also demonstrated an increase in CVD with ApoE level; however, this increase was limited to a subgroup of women with high high-density lipoprotein cholesterol (HDL-C) levels [11]. These findings require replication for a number of reasons. First, two of these studies were focussed, by design, on those aged 85 y and above. Results may not apply in younger individuals, particularly as there is evidence of an age-dependent decline in the prevalence of individuals with the ε4/ε4 genotype [12], who also have lower circulating concentrations of ApoE. Second, each study was small, with few CVD outcomes (68, 54 [strokes in stroke only study], and 156 in the three studies, respectively). Third, the ApoE-CVD associations found in these studies were adjusted for varying risk factors, but the adjustments were not uniform throughout. Lastly, initial reports of any association often yield an inflated effect estimate. This phenomenon, known as the Proteus effect or winner’s curse [13], prompts the need for replication. We therefore aimed to examine the association of ApoE concentration with CVD events in three new studies of ~10,000 middle-aged individuals in total, with a wider range of measures of cardiovascular risk factors and 1,413 CVD events.
Study details are described in detail elsewhere, but in brief; the Northwick Park Heart Study II (NPHSII) [14] is a prospective study that recruited 3,012 men from nine general practices across the UK with 15 y follow-up. All participants were free of CVD at the time of recruitment. Median follow-up from time of ApoE measurement was 9.9 y. There were 205 CHD events, including 136 acute myocardial infarctions (MIs), of which 50 were fatal, 65 involved coronary revascularisations, and four were silent MIs. There were 79 strokes (35 fatal). Sixteen individuals had both CHD and stroke events. In total there were 268 CVD events.
The English Longitudinal Study of Ageing (ELSA) [15] is a prospective study of household participants aged 50 y and over and resident in England. ELSA participants were recruited from respondents of the annual Health Survey for England in 1998, 1999, and 2001. In ELSA, ApoE measurement was in wave 2 (2004–2005), and participants were asked if a doctor had diagnosed MI and stroke at waves 3 (2006–2007), 4 (2008–2009), and 5 (2010–2011), a mean of 5.4 y from wave 2. Mortality and follow-up was available through the National Health Service Central Registry until 31 January 2010. Registration of death within 5 d is a legal requirement in England, so participants not registered can be assumed to be alive. Death certificates were coded using the tenth revision of the International Classification of Disease, and those categorised as CVD were extracted. During this period there were 165 MIs, of which 65 were fatal; 170 strokes, of which 40 were fatal; and 12 individuals with both diagnoses. In total, there were 323 CVD events.
The Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT) [16] is a randomised clinical trial of blood-pressure- and lipid-lowering treatment in the prevention of CVD in individuals at high cardiovascular risk. A 1:1 nested case-control sample of 1,666 individuals matched for age, sex, and country of origin was used for this study. The primary endpoint of the study was combined non-fatal MI (including silent MI) and fatal CHD. Secondary endpoints included all-cause mortality, total stroke, all coronary events, total cardiovascular events and procedures, cardiovascular mortality, and non-fatal and fatal heart failure. Here, only CHD cases were included, where ascertainment was at clinical follow-up. Certified causes of death were sought, and, when available, national registries were used to find information on patients who did not return for their final visits. Endpoints were submitted and adjudicated by the ASCOT endpoints committee, the members of which were unaware of treatment assignment.
The analyses in all studies were in accordance with approval from relevant research ethics committee, and all participants provided written informed consent. See STROBE statement (S1 Text).
Circulating ApoE was measured using a nephelometric method on a BN II nephelometer (Siemens), using a non-isoform-specific polyclonal antibody. ApoE measures from NPHSII were made in citrated plasma taken at the fourth annual visit and were taken after a light breakfast. ELSA and ASCOT measures were made in serum. Samples used in ELSA were taken at wave 2. Most participants were fasted; those over age of 70 y (35%) and those with diabetes (7.2%) were not asked to fast (total 35.8% of ELSA participants), although most were seen prior to eating breakfast. In ASCOT, samples were taken at randomisation in the fasted state. Samples used for all measures were taken at a time prior to any CVD events. Differences in concentration dependent on whether plasma or serum was used for ApoE measurement were overcome at the analysis stage by standardising measures of ApoE.
The ELSA and NPHSII studies provided information on the APOE SNPs rs7412 and rs429358, which were used to reconstruct the ε2/ε2, ε2/ε3 ε2/ε4, ε3/ε3, ε3/ε4, and ε4/ε4 haplotypes. Genotyping in NPHSII was carried out using Taqman (Applied Biosciences), and genotyping in ELSA was based on KASP chemistry at LGC Genomics.
CHD events were defined as fatal or non-fatal MI, angiographic evidence of coronary atherosclerosis, or CHD events that required intervention or where individuals underwent coronary artery bypass grafting. Stroke was defined as fatal or non-fatal; the definition encompassed both ischaemic and haemorrhagic stroke. Total CVD was a composite of CHD and stroke cases and was defined as fatal or non-fatal. Specific definitions are given in primary study reports. Comparison between definitions was carried out in order to ensure that these could be pooled for meta-analysis.
Statistical analysis was carried out according to a prespecified analysis plan (S2 Text). Non-normally distributed continuous variables were logarithmically transformed. For these variables we report geometric means with approximate standard deviations (SDs). Differences between case-control groups were tested by unpaired t-test or ANOVA, and differences in the distributions of categorical variables were assessed by the χ2 test. Patients with missing data were not included in the analysis. ApoE was in addition standardised, that is, rescaled to have a mean of 0 and SD of 1, to overcome any difference between sample types (e.g., serum versus plasma) across studies.
Studies contributed to the analyses in different ways, summarised in S1 Table.
A total of 9,587 individuals (1,642 from ASCOT, 5,389 from ELSA, and 2,556 from NPHSII) for whom measures of ApoE were available were included from all three studies. Baseline characteristics are shown in Table 1. Variables that were loge transformed included ApoE, triglycerides, ferritin, fibrinogen, C-reactive protein (CRP), glucose, and body mass index. Of the 9,587 individuals with ApoE measures included in the study, 1,152 had missing data on one or more of the Framingham variables included in model 2. The proportion of missing data was 11.8% for ELSA, 1.6% for ASCOT, and 19.2% for NPHSII.
Bivariate associations of log-transformed ApoE with other risk factors were assessed in the two population-based studies ELSA and NPHSII (Table 2). Significant associations were observed with a number of cardiovascular risk factors, including systolic and diastolic blood pressure; total cholesterol, LDL-C, HDL-C, and triglycerides; and body mass index (Figs 1–3; Table 2). ApoE was also significantly associated with ApoB (correlation coefficient 0.174, p < 0.001) and lipoprotein(a) (correlation coefficient −0.12, p < 0.001) in NPHSII, where these measures were available, but not with ApoA1 (correlation coefficient 0.02, p = 0.46). In addition, circulating ApoE was significantly associated with markers of inflammation including CRP (measured in ELSA and NPHSII, correlation coefficient 0.11 and 0.13, respectively, p < 0.001 for both), ferritin (ELSA only, correlation coefficient 0.07, p < 0.001), and, to a lesser extent, fibrinogen (correlation coefficient 0.05, p < 0.001, in ELSA, and 0.01, p = 0.52, in NPHSII) (Figs 4 and 5; Table 2). There were no significant differences following adjustment for age or gender.
There were no statistically significant differences in circulating concentrations of ApoE between CVD cases and controls for all three studies (Table 3). There was no association of ApoE concentration either per SD increase or by tertile with CVD, with no additional evidence of a trend of association across tertiles (Table 4). There was no significant association between ApoE concentration and CVD in individual studies (Table 4) or when estimates were pooled using random effects meta-analysis (OR 1.02, 95% CI 0.96, 1.09) or after adjustment (Framingham-adjusted OR 0.97, 95% CI 0.82, 1.15) (Fig 6). Adjustment for LDL-C alone also did not impact the overall association (OR 1.04, 95% CI 0.96, 1.13). When CHD (all CHD, fatal and non-fatal) and stroke (all stroke, fatal and non-fatal) endpoints were assessed separately across all three studies, there was no significant association of circulating concentration of ApoE with either outcome (Fig 6; S2 Table). For the crude associations, analysis of ApoE quintiles and CVD risk was carried out and a quadratic model was fitted following peer review to confirm the absence of a U-shaped association as an explanation of a null effect (S3 Table).
In an exploratory analysis restricted to the sub-sample of individuals from NPHSII with the ε3ε3 genotype, we conducted a time-to-event analysis by tertile of ApoE concentration to overcome any isoform-specific effect. The Kaplan-Meier curves showed no major differences in incidence of CVD events according to ApoE concentration (Fig 7).
Levels of circulating lipids—including total, LDL-, and HDL-C and triglycerides—and ApoE in the APOE genotype groups were as predicted from previous studies (S4 Table). CVD events have previously been shown to be associated with APOE genotype. Genotype was available in NPHSII and ELSA, and this genotypic effect is replicated on a smaller scale (S5 Table).
We have described the molecular epidemiology of circulating ApoE in a large study of middle-aged individuals that was population-based, in contrast to previous studies, which have focused on ApoE concentrations in related individuals [7]. We have confirmed here that circulating ApoE concentration is associated with the levels of major circulating lipids and lipoproteins including total cholesterol, LDL-C, and HDL-C as well as triglycerides; this finding is consistent with published findings [7]. We further investigated the association of circulating ApoE concentration and CVD risk in ~10,000 individuals including 1,413 CVD cases. In contrast to prior reports [9–11], we found no association between ApoE concentration and CVD events, either when expressed as an OR per SD change of ApoE concentration or when examined by tertile of ApoE concentration or by subtype of CVD.
The lack of an association between circulating ApoE and CVD contrasts with other studies examining this association [9–11] and with an anticipated CVD protective effect of higher circulating ApoE. In explaining the discrepancy of findings between the current and previous reports, the following points are important. First, the study presented here includes individuals who are middle-aged and are more likely to represent the major group targeted for primary CVD prevention. Two of the previous reports included only individuals ≥85 y old, a population that may display a survivor bias. Moreover, it is known that the relative frequency of the ε4ε4 genotype group, which is strongly associated with lower circulating ApoE concentrations, declines with age [12]. Therefore, older populations are predicted to have higher circulating ApoE concentration (associated with enrichment for the ε2 and ε3 alleles). We did observe an interaction between age and sex (ELSA: p for interaction < 0.001; ASCOT: p for interaction = 0.02). This interaction has also been observed in other studies that measured ApoE at scale [17]. This interaction could be due to genotypic effect or could be a result of the effect of other unmeasured lipid intermediates that could also alter circulating levels of ApoE in men and women. The association of ApoE with CVD risk was also replicated in PREVEND [11], a more representative sample; however, in this study a borderline association was observed between circulating ApoE and CVD risk in women only. Second, all CVD cases in the current analysis are incident events, and this should largely overcome any issues of reverse causality that can affect conclusions drawn from the retrospective case-control or cross-sectional designs of some, but not all, previous reports [9,10]. Lastly, the size of the current analysis (1,413 cases, five times more than the largest published study), and the replication of findings across three different datasets, overcomes some of the limitations of prior reports on the association of circulating ApoE and CVD. During the review process of this manuscript, a large (~92,000 individuals) population-based study examining the association of ApoE and CHD was published [18] demonstrating that circulating ApoE increases CHD risk. It is interesting to note that our findings from all CHD (OR 1.05, 95% CI 0.98, 1.13) approach their findings (OR for ischaemic heart disease of 1.15, 95% CI 1.04, 1.27; OR for MI of 1.16, 95% CI 1.00, 1.36), however do not reach significance. Moreover, the association reported by Rasmussen et al. is limited to men and attenuates with the adjustment of triglycerides. Like the other published studies, Rasmussen et al. did not demonstrate a convincing association with ApoE, despite a greater power to detect such an association. The findings that they report are also contrary to the biological prior from genetic and animal studies, that higher ApoE concentration should be protective against atherosclerotic disease.
The observation that individuals with very low or completely absent circulating ApoE develop early atherosclerosis, and that this has a genetic basis [5], has been influential in the development of hypotheses on the role of ApoE in atherosclerosis. The largest genetic study investigating the association of APOE mutations with CVD support this hypothesis [8]. For the ε2ε2 (associated with higher circulating ApoE) versus ε3ε3 comparison, Bennet et al. [8] demonstrated an OR of 0.80 (95% CI 0.70, 0.90) for CVD. Our data do not reach significance; for the same comparison, we have 30% power to detect an OR of this size at a 5% significance level using 377 cases and 4,709 controls, but the trend of the effect is similar (S5 Table). Human studies have been dominated by evaluation of the relationship of APOE genotype with a range of diseases including CVD, Alzheimer disease [19], and age-related macular degeneration [20]. Together with animal studies, these human studies have contributed to the prevailing view that circulating ApoE is anti-atherogenic. For example, carriage of ε2 is associated with higher circulating concentration of ApoE, lower LDL-C concentration, and lower risk of CVD [8]. Conversely, carriage of ε4 is associated with directionally opposite changes in ApoE and LDL-C concentration and a higher risk of CVD. These findings have been interpreted as a consequence of the functional differences between isoforms of ApoE, namely, lipid and receptor binding, rather than circulating concentrations of the protein. For example, the E2 isoform binds LDL-R and LRP1 with a reduced efficiency compared to the other isoforms (≤2% compared to E3; E4 has affinities comparable to E3) [21,22]. This reduced binding efficiency has been thought to result in up-regulation of expression of LDL-R and LRP1, with the consequent effect of increasing clearance and lowering circulating concentrations of pro-atherogenic VLDLs, chylomicrons, and their remnants. E2 as well as E3 preferentially bind to small phospholipid-rich HDL-C particles, in contrast to E4, which binds larger pro-atherogenic triglyceride-rich particles [23]. Binding with HDL-C supports an additional role of E2 in reverse cholesterol transport, as ApoE acts as a ligand for high-density-lipoprotein-mediated cholesterol delivery to the liver. However, the present study indicates a null association of ApoE concentration with CVD risk overall and in analyses restricted to individuals with the ε3/ε3 genotype.
One of the functional consequences of the reduced binding to LDL-R and LRP1 of E2 could be higher detectable concentrations of ApoE. This has been interpreted as higher levels of circulating ApoE being protective. However, a higher detectable concentration of ApoE could be a paradoxical increase as a consequence of the genetic point mutation rather than being causal in itself. One way in which ApoE concentration might be useful in CVD prediction is with isoform-specific measures. The predictive utility of isoforms could then be tested in large populations. However, it should be borne in mind that, given the distribution of cases and controls amongst genotype groups, most cases will occur in those with circulating ApoE3, as these individuals make up the majority of the population (S1 Fig). Most cases of CVD will therefore have circulating concentrations of ApoE within the normal range (i.e., within 2 SDs of the population mean). However, that does not preclude ApoE being a useful target for CVD; rather, any drug would need to recapitulate the pattern of lipids seen with circulating E2. Such a drug could be useful in individuals with any genotype.
Two other courses of investigation to more precisely defining the role of ApoE in health and disease are now necessary. First, the view that APOE confers disease risk (and/or protection) mediated through LDL-C may be over simplistic. Functional studies of all isoforms indicate the importance of ApoE in the clearance of other potentially atherogenic lipid particles including VLDLs, chylomicrons, and their remnants. To date, studies in small samples (~200 individuals) [24] have indicated that there is indeed a differential effect of APOE genotype on circulating lipids other than the “traditionally” measured LDL-C, HDL-C, and triglycerides. Investigation of other lipid intermediates and their association with APOE now requires a population-level effort and may prove to be more fruitful than ApoE with respect to predictive utility for CVD. Measurement of these lipid intermediates is now possible in a time- and cost-efficient way using high-throughput mass spectrometry and nuclear magnetic resonance platforms. Second, the view that the genetic variation in the APOE locus that influences CVD risk is limited to the ε2/ε3/ε4 haplotype is also being challenged. Genetic studies using genome-wide [25] and gene-centric SNP arrays [26] implicate genes flanking APOE, including BCL3, PVRL2, TOMM40, and APOC1-C4-C2, rather than APOE itself in regulation of circulating lipid concentrations, in particular LDL-C, as well as additional variants within the APOE gene. The strength of these associations is greater than that with SNPs in APOE alone. Limitations of some of these studies include the fact that both SNPs that determine the ε2/ε 3/ε 4 haplotype (rs429358 and rs7412) are not included on all SNP arrays or are not well imputed and therefore have not been identified as signals in such analyses. However, even where these SNPs are directly typed and included in analyses, the strength of the association with circulating lipids is greater from the flanking regions [26]. Results from large genome-wide association studies and consortia data can now be used to inform further functional and molecular analyses, in order to truly assess the effect of APOE and circulating ApoE on disease risk. This is not only of interest in CVD, but also in other diseases of ageing including Alzheimer disease, where the risk conferred by APOE ε4 is >5-fold greater than that seen in CVD, and age-related macular degeneration, where, in contrast to Alzheimer disease and CVD, ε2 confers an increase in risk of disease and ε4 confers a protective effect.
It is important to address some limitations of this study. First, the assay used here is commercial, using a polyclonal antibody. It does not discriminate between isoforms of circulating ApoE. However, the pattern of observing higher circulating concentrations with carriage of E2 and lower concentrations with carriage of E4 is preserved and consistent with other studies. We are therefore confident that the concentrations measured are a reflection of the underlying genotype. Second, although the sample size is larger than that of previously reported studies, compared to studies that reliably estimate the risk of a marker for a disease, it is still moderate. These studies require further replication in order to more precisely delineate the direction and magnitude of the effect. Third, we are unable to differentiate between ischaemic and haemorrhagic stroke in the presented results, but, given that haemorrhagic stroke represents 10%–15% of stroke cases in the general population (https://www.strokeaudit.org) and a total of 249 strokes are included, it is doubtful that this would bias the results. Finally, samples that were used for measurement were taken before incident events, and the median time before an event across all included studies was 5.5 y. This time between sampling and event could be one explanation of an attenuated effect. However, the follow-up time is comparable to that of other published studies that do show a relationship of circulating ApoE with CVD, even with small sample sizes. One way in which this variability of ApoE could be overcome is to construct and then adjust for regression dilution ratios. Ratios are based on regression of serial measures of ApoE on baseline levels with any additional and necessary adjustments. However, serial measures were available neither in the studies presented here nor in the current ApoE literature, and these analyses will be possible only once these data emerge. Exploratory analysis applying to ApoE regression dilution ratios calculated by the CRP-CHD Genetics Consortium for another blood-based biomarker, CRP (where the regression dilution ratio adjusted for age and sex was 0.57, 95% CI 0.51, 0.64) [27], indicated that the effect of regression to the mean is unlikely to have a major effect on the observed association, with the odds of CVD for a 1-SD increase in ApoE increasing from 1.02 to 1.035. These analyses were carried out following editorial comments during the review process.
In order to more precisely link genetic variation in the APOE gene locus to disease and biomarkers, finer resolution of the genetic variation in this region is required. This can be achieved through next-generation sequencing or imputation against broad (1000 Genomes) and deep marker panels. Integrating this information with circulating concentrations of ApoE and a range of other biomarkers, and specifically lipid intermediates, may clarify the link between genotype, circulating ApoE, and CVD and other diseases. This will inform the clinical utility of tests based on measurement of either genotype or circulating ApoE, or of indeed targeting ApoE with new therapeutic agents.
|
10.1371/journal.pntd.0002588 | The Intestinal Expulsion of the Roundworm Ascaris suum Is Associated with Eosinophils, Intra-Epithelial T Cells and Decreased Intestinal Transit Time | Ascaris lumbricoides remains the most common endoparasite in humans, yet there is still very little information available about the immunological principles of protection, especially those directed against larval stages. Due to the natural host-parasite relationship, pigs infected with A. suum make an excellent model to study the mechanisms of protection against this nematode. In pigs, a self-cure reaction eliminates most larvae from the small intestine between 14 and 21 days post infection. In this study, we investigated the mucosal immune response leading to the expulsion of A. suum and the contribution of the hepato-tracheal migration. Self-cure was independent of previous passage through the liver or lungs, as infection with lung stage larvae did not impair self-cure. When animals were infected with 14-day-old intestinal larvae, the larvae were being driven distally in the small intestine around 7 days post infection but by 18 days post infection they re-inhabited the proximal part of the small intestine, indicating that more developed larvae can counter the expulsion mechanism. Self-cure was consistently associated with eosinophilia and intra-epithelial T cells in the jejunum. Furthermore, we identified increased gut movement as a possible mechanism of self-cure as the small intestinal transit time was markedly decreased at the time of expulsion of the worms. Taken together, these results shed new light on the mechanisms of self-cure that occur during A. suum infections.
| Ascaris lumbricoides is the most common intestinal parasite in humans. A. suum is closely related to A. lumbricoides but infects pigs and can be used to study the immune response against larval stages. Most larvae are eliminated from the small intestine between 14 and 21 days after infection in what is called a self-cure reaction. The remaining larvae after this point will be able to grow into adults and reproduce. We show here that the intestinal self-cure of A. suum is locally triggered as part of an innate immune defense mechanism. When pigs received lung stage larvae, they were still able to eliminate the parasite, indicating that passage through the liver or lungs is not essential to eliminate the larvae upon their return in the small intestine. We could identify a decrease in the intestinal transit time at 17 days post infection, indicating an increase in gut movement, which could explain why the worms were being driven out at this time.
| In (sub)tropical countries, Ascaris lumbricoides is an important soil transmitted helminth, infecting around 1 billion people worldwide. Although most cases are sub-clinical, Ascaris infections lead to malabsorption and malnutrition and in rare cases obstruction or puncture of the intestinal wall and penetration of the bile and pancreatic ducts occur [1]. The closely related roundworm A. suum is one of the most common parasites in pigs causing economic losses in agriculture due to increased feed conversion rate and liver condemnation [2]. Because of the identical life cycle, the high genetic similarity between these parasites [3], and because A. suum is a zoonosis [4], [5], A. suum infections in pigs make an ideal model for A. lumbricoides infections in humans. Cross infections and gene flow between the 2 species also occurs [6], [7], which led to the debate whether or not they belong to the same species [8], [9].
After ingestion, the A. suum eggs hatch and release third stage larvae (L3) in the intestine. The larvae will penetrate the caecal or colonic wall, reach the lungs via the liver, after which they will be coughed up and swallowed back in. Once back in the small intestine, they will develop first into L4 and L5 stage larvae and eventually into adults, preferentially inhabiting the proximal half of the small intestine [10]. Immunity against invading third stage larvae takes several weeks of exposure to infective eggs to develop [11], [12]. In contrast, even in primary infections an expulsion mechanism, termed self-cure, causes the elimination of most of the fourth stage larvae (L4) from the small intestine between 14 and 21 DPI, and this self-cure is independent of the inoculation dose [10]. The effector mechanisms driving this elimination are largely unknown.
To date, it is not known if humans infected with A. lumbricoides also undergo spontaneous cure. However, in pigs, before self-cure the number of larvae in the small intestine is roughly 30–50% of the infection dose. After 21 DPI, however, the number of larvae is greatly aggregated, with the majority harboring low numbers of worms and a small proportion having the majority of worms [10]. This overdispersion is also seen in humans infected with A. lumbricoides [13] and is likely caused by a similar reaction. Although predisposition is a multifactorial phenomenon that includes external factors such as exposure, understanding the mechanism of aggregation might also help to explain the predisposition to high or low worm burdens observed in humans [13].
The aim of this study was to investigate in more detail the gastro-intestinal immune response leading to the elimination of L4 A. suum larvae from the small intestine and the contribution of the hepato-tracheal migration to the expulsion of the parasite.
All animal experiments were conducted in accordance with the E.U. Animal Welfare Directives and VICH Guidelines for Good Clinical Practice, and ethical approval to conduct the studies were obtained from the Ethical Committee of the Faculty of Veterinary Medicine, Ghent University (EC2011/086, EC2009/145 and EC 2013/51) who have also approved the document.
Helminth naive Rattlerow Seghers hybrid piglets of 10 weeks old were used. The animals were routinely checked for A. suum by coprological examination and at the start of the experiment 2 sentinel animals from the same pen were euthanized to confirm absence of larval stages. The animals had access to feed and water ad libitum.
A. suum eggs were obtained from gravid females collected at the local abattoir from pigs that were being processed as part of the normal work of the abattoir. After incubation in 0,1% KCr2 for 2 months, embryonation was confirmed by way of light microscopy.
All animals were fasted before necropsy and then killed with a captive bolt pistol, exsanguinated and the intestines were removed. Samples for RNA extraction and histological analysis of the jejunum were taken 3 meter caudal of the pylorus. The small intestine was further divided in duodenum, jejunum and ileum. The contents of the 3 parts of the small intestine were collected separately and the intestines were rinsed twice with tap water to collect any remaining larvae. The washing was added to the corresponding content and sieved with a 122 µm sieve. A. suum larvae were subsequently counted under a microscope.
Jejunal tissue was immediately snap frozen in liquid nitrogen and stored at −80 until RNA extraction. RNA extraction was performed using Trizol reagent (Invitrogen), combined with an RNeasy mini kit (Qiagen). A DNase treatment was included to prevent genomic contamination. RNA integrity was assessed using a Biorad Experion with a standard sensitivity chip. cDNA was synthesized with a Biorad cDNA synthesis kit, starting from 1 µg of RNA.
Primers for the real time PCR reactions were designed with the Primer3 software [14] and are listed in Table S1. PCRs were run using Fast SYBR Green Master Mix (Applied Biosystems) on an AB StepOnePlus Real-Time PCR System. Primer specificity was confirmed by observing the melting curve and by sequencing PCR products. Gene expression levels were normalized based on housekeeping genes selected using Genorm [15]. Housekeeping genes tested were: b2m, gapdh, hmbs, rpl4, tbp1 and ywhaz. The genes selected for normalization were hmbs and tbp1. Gene transcription levels are expressed as fold change compared to uninfected controls.
Tissue samples were washed in PBS, processed with the Swiss roll technique in order to obtain a large surface for histological examination [16] and fixed in either 10% formaldehyde or Carnoy's fixative for 24 h at room temperature. After fixation, the tissues were dehydrated by passage through a series of graded alcohol dilutions, followed by embedment in paraffin. Tissue samples were cut in 4 µm sections. To assess general histopathological damage and the accumulation of eosinophils, formaldehyde fixed samples were routinely stained with haematoxylin-eosin. The length of the villi and depth of the crypts in the jejunum were measured for 20 villi and their corresponding crypts under a microscope using a calibrated micrometer at 100× magnification. Mucosal eosinophils were counted at 400× magnification on 10 fields corresponding to 0,162 mm2. Mast cells were counted on toluidine blue stained slides at 200× magnification using a weibel2 graticule [17]. For immunohistochemistry: formaldehyde fixed, paraffin embedded sections were rehydrated and an antigen retrieval step with citrate buffer was included. Endogenous peroxidase activity was blocked using 1% hydrogen peroxide. Sections were stained with rabbit anti-human CD3 (Dakocytomation A/S) to detect intra-epithelial lymphocytes (IELs) or mouse anti-human MAC387 (Serotec) to stain macrophages. Biotinylated secondary antibodies (Dakocytomation A/S) were added and staining was performed using the peroxidase streptavidine complex (Dakocytomation A/S), diaminobenzidine tetrahydrochloride (DAB, Sigma–Aldrich) and H2O2. Sections were counterstained with haematoxylin. Macrophages were counted at 200× magnification using a weibel2 graticule [17] while IELs were counted for 5 villi randomly and expressed as number of IELs per 100 µm villus epithelium.
The Ascaris-specific IgA, IgG, IgE and IgM levels in the serum against the L4 larvae were determined using an indirect in-house ELISA. L4 larvae were collected from the small intestines of animals at 14 DPI. The larvae were ground using a mortar and pestle in liquid nitrogen to a fine powder and subsequently dissolved in PBS to which a 1∶1000 dilution of protease inhibitor cocktail (Sigma-Aldrich) was added. After incubating for 2 hours at 4°C, the extract was centrifuged at 10.000 g for 10 minutes. The supernatant was passed through a 0,22 µm filter and stored at −70°C until use. This extract is being referred to as AsL4.
Plates were coated overnight at 4°C with 100 µl of 5 µg/ml AsL4 in 0,05M sodium bicarbonate buffer (pH 9,6). Serum was added at a concentration of 1/100 and HRP-conjugated goat anti-pig IgM (Thermo Scientific), IgG and IgA (Bethyl laboratories) were used as conjugate at a dilution of 1∶50000, 1∶10000 and 1∶5000, respectively. For the detection of pig IgE antibodies, a pig IgE cross-reacting mouse anti-human IgE antibody (Sigma-Aldrich) at 1∶5000 was used in combination with a HRP-conjugated rabbit anti-mouse IgG at 1: 10000. Finally, O-phenylenediamine 0.1% in citrate buffer (pH 5.0) served as substrate and optical density (OD) was measured at 492 nm. All measurements were performed in duplicate.
The purification of circulating eosinophils and the degranulation assay were performed as previously described [18]. Reactive oxygen species production was measured using a chemiluminescence assay with PMA 5 µg/ml as positive control, HBSS with Ca2+/Mg+ as negative control or 1 mM SIN-1 as a ROS donor. Eosinophils from 1 pig were seeded in a 96-well plate at 2×105 cells/well in 100 µl luminol (1 mM) in HBSS with Ca2+/Mg+. After 5 min of background measurement at 37°C, 10,20 or 50 A. suum L4 larvae collected from infected pigs at 14 DPI were added in 100 µl HBSS, as well as the control agents. To test if there was antibody or complement dependent degranulation, serum from 5 uninfected and 5 animals at 17 DPI was pooled and added at 1/100 dilution. Heat inactivation of serum was done at 58°C for 30 minutes. ROS-production was measured during 120 min in the integration mode. Each condition was performed in triplicate and ROS-production was expressed as the fold change in relative light units (RLU) compared to negative controls (HBSS). The experiment was performed 3 times independent from each other.
Eleven pigs were infected with 3000 A. suum eggs. Ten days after infection, 2 animals were euthanized to confirm batch infectivity. The small intestinal worm counts in these two pigs were 2019 and 2315. Small intestinal transit time was measured in the remaining 9 pigs at 5 days before infection and at 9, 17 and 35 days after A. suum infection. The pigs were starved for 12 hours before barium sulfate was given through gastric intubation at a dosis of 4 ml/kg bodyweight. Lateral and dorso-ventral radiographs were taken every half hour until barium sulfate was located in the colon. If a radiograph was inconclusive about the presence of contrast material in the colon, it was repeated after 10 minutes. The time it took for the barium to reach the colon was recorded as the small intestinal transit time. After the last transit time measurement, the animals were euthanized and worms were collected.
For statistical analysis, GraphPad Prism software (v5.0c) was used. Since a non-Gaussion distribution could be expected, differences between the infected groups and uninfected animals were tested using a nonparametric Kruskal-Wallis test with Dunn's multiple comparison post hoc tests. Finally, a repeated measures Friedman test with Dunn's multiple comparisons post hoc test to find differences in transit time between the different time points.
Animals were either orally infected with 2000 A. suum eggs or with 1000 9-day-old L3 that were collected from the lungs of donor animals. The worm counts are summarized in Table 1. For egg infected pigs, the average total worm count at 10 DPI was 312, with 19% of larvae present in the duodenum, 73% in the jejunum and 9% in the ileum. At 17 DPI, the average total number of larvae present was reduced to 19, most of which were present in the ileum. By 28 DPI, all animals were negative for A. suum.
When animals were orally infected with lung L3's obtained from donor animals, they were still able to eliminate the larvae. At 2 DPT, 38% of transferred larvae were recovered, almost exclusively from the jejunum, indicating a successful transfer. At 7 DPT, although the total number of worms was similar to that of 2 DPT, 50 % of the larvae were now present in the ileum. At 18 DPT, no larvae could be recovered from the animals.
Ascaris L4 specific IgA, IgE, IgG and IgM antibody levels in serum of A. suum egg or lungs stage infected animals were measured using an indirect ELISA (Figure 1). During infections with eggs, AsL4 specific IgA, IgM and IgG levels were increased from 10 DPI onwards, whereas AsL4 specific IgE levels were only detectable in serum at 17 DPI. Although the self-cure reaction occurs 7 days after lung L3's are transferred, no statistically significantly increases of AsL4 specific IgA, IgM, IgG and IgE antibodies could be detected at this time. IgM, IgG and IgE levels were significantly increased only at 18 DPT whereas no change in serum IgA levels was observed.
We examined whether the release of antigens during the molt from L3 to L4 that occurs around D12 is necessary to trigger the expulsion of the larvae. Therefor we collected 14-day-old L4 intestinal larvae from donor animals and transferred 1000 larvae orally into naïve animals. The number of larvae in each section of the small intestine was counted at 2, 7 and 18 days post transfer. The larvae counts are summarized in Table 2. At 2 DPT around 60% of the transferred larvae could be recovered and 87% of the recovered larvae are present in the jejunum. Five days later the total number of larvae in the small intestine is similar to that at 2 DPT, but most larvae are present in the terminal part of the small intestine. At 18 DPT the total number of larvae has not decreased compared to 7 DPT, but 90% of larvae are now present again in the jejunum, indicating that they could counteract the peristaltic movement to inhabit the proximal region of the small intestine.
The results of the histological parameters investigated are shown in Figure 2. To assess general histopathological changes, villus length and crypt depth were measured. Villus/crypt ratios decreased shortly after contact with A. suum larvae, due to a blunting of the villi. Although this effect was observed in both infections with eggs and L3 and L4, it was only temporary, as the villi recovered by 17 DPI/7 DPT. At 17 DPI, coinciding with the expulsion of the parasite, there was a significant increase in mucosal eosinophils. After elimination of the larvae, i.e. 28 DPI, the number of eosinophils decreased to a level similar to that before the infection. A similar pattern was observed following transfer of L3's, with a peak in eosinophil counts at 7 DPT. The transfer of L4 larvae resulted in high eosinophil numbers at 7 DPT and 18 DPT.
Mucosal macrophages followed a similar pattern as eosinophils in A. suum egg infected pigs, with a 9-fold increase in the number of macrophages per mm2 mucosa at 17 DPI that returned to baseline level at 28 DPI. In contrast to normal infections, in both L3 and L4 transfer infections, no increase in the number of macrophages was observed at any of the time points investigated. No statistically significant changes were observed in the number of intestinal mast cells in any of the infection experiments. Finally, intra-epithelial T cells were significantly elevated in all infection experiments at the time when larvae were being driven towards the distal end of the small intestine, i.e. at 17 DPI/7DPT. In the A. suum egg infections and in the L3 transfer experiment, IELs were still elevated even after the worms were eliminated, while in the L4 transfer experiment the IELs returned to normal levels at 18 DPT.
The results of the quantitative PCR analysis on a set of 25 genes for egg infected, L3 and L4 transferred animals are summarized in Table 3. With egg infections, the gene expression pattern was polarized towards a Th1-like response, with significant upregulations observed for ifng, il12a, il12b, stat4 and nos2a. In contrast, none of the Th2 related genes were significantly impacted during infection with A. suum eggs. In the L3 and L4 transfer experiments, more mixed responses were measured. In addition to some Th1 markers, an increase in the typical Th2 transcripts il4 and il13, together with increases in regulatory transcripts, such as foxp3, and tgfb were observed.
For all infection experiments there was an upregulation of genes associated with cytotoxic cells, mainly granzyme A and B, perforin 1 and NKG2D. Additionally, several eosinophil-associated genes were induced, such as those encoding for eosinophil peroxidase, eotaxin 1, eotaxin receptor and IL-5 receptor alpha.
Results of the eosinophil degranulation assay are shown in Figure 3. Measurement of the reactive oxygen species (ROS) indicated that the eosinophils did not degranulate after incubation with A. suum L4 larvae, even in the presence of serum from infected animals. To exclude the possibility that L4 larvae would capture ROS released in the medium, A. suum L4 larvae were cultured together with SIN-1, a molecule that releases NO and ROS. A. suum L4 larvae together with SIN-1 in medium gave no significant differences in measured ROS compared to SIN-1 without L4 larvae (1636±704 RLU versus 977±344 RLU, respectively).
The small intestinal transit time was measured by following barium sulfate passage through the small intestine before infection and at 9, 17 and 31 days post infection with 3000 A. suum eggs (see Figure 4). There was a small, non-significant increase in the small intestinal transit time at 9 DPI compared to their pre-infection transit time. At 17 DPI the small intestinal transit time was significantly lower than before the infection. By 35 DPI, 8 out of 9 animals were A. suum negative and one pig had 29 A. suum worms. At this time, the intestinal transit time was still somewhat lower than before infection, but not significantly.
Here we investigated the immunological basis of the self-cure reaction during primary A. suum infections. In addition, we studied the influence of the migration of the larvae through the body on the self-cure reaction. By transferring lung stage A. suum larvae from one animal to another, we have a simple model to study the effect of tissue migration on the initiation of the self-cure response. In animals bypassing the passage through the liver and lungs, the self-cure reaction occurred with the same kinetics as animals receiving infectious eggs, i.e. around 7 days after contact with the small intestine. Furthermore, both in infections with eggs and with lung stage larvae all larvae were expelled by 18 days of exposure to the small intestine. A previous study by Jungersen et al. led to the speculation that the expulsion of A. suum might be affected when the liver is bypassed. They injected in vitro hatched L3's intravenously in pigs and found a higher percentage of animals harboring adult A. suum at 70 DPI than what is usually observed, even though at 14 DPI there were comparable numbers of L4 larvae between intravenously and orally infected animals [19]. Unfortunately, not enough time points and control groups were included to confirm if previous priming in the liver was indeed required to eliminate the larvae from the small intestine. The results presented here now unequivocally show that the self-cure mechanism is a locally triggered phenomenon, independent of previous passage through the liver or lungs.
Additionally we sought to determine whether antibodies play and important role during self-cure. Since in normal infections there are already A. suum specific antibodies present at 10 DPI, it was previously suggested that antibodies played an important role in the expulsion of the parasite [20]. Although we confirm the presence of antibodies during self-cure in egg infected pigs, the absence of A. suum specific antibodies when larvae were being expelled in animals that received lung stage larvae would indicate that A. suum specific antibodies do not have a major role in the early self-cure against A. suum. This is further supported by the observation that when L4 larvae are transferred, most larvae were being driven to the distal end of the small intestine around 7 DPT. However, it still remains possible that both specific and non-specific antibodies present in the mucosa itself could contribute to the A. suum expulsion [21]. In addition, although it is not clear to what extent maternal antibodies would still be present in pigs of 12 weeks age, the passive transfer of antibodies has also been shown to contribute to parasite expulsion [22]. Therefore, it would be interesting in future studies to also analyse the mucosal antibody responses.
Remarkably, and in contrast to the transfer of lung stage larvae, the L4 transferred larvae were able to return to the jejunum by 18 DPT. By this time, the larvae are already 32 days old, i.e. an age at which in natural infections they are also not affected by the self-cure response anymore. It appears that these larvae, once they have developed to a certain stage, are able to counteract the self-cure response. This is in agreement with a microarray study on larvae in the jejunum and ileum during self-cure, where they found that only the more metabolically active larvae could remain in the jejunum [23]. However, from our results it is clear that the larvae present in the ileum are still alive and can return to the jejunum if they are active enough. This also contradicts the suggestion that self-cure is a parasite-driven suicide phenomenon based on the density of the parasites [24]. It indicates that there is a fine balance between the host that is trying to drive the parasite out and the parasite's ability to counteract this effort. This also explains why adults can remain in the small intestine for months or years without being driven out.
The histological and RNA transcription analysis showed some common characteristics associated with the expulsion of larvae in all the experiments performed here. The peak of expulsion coincided with a peak in mucosal eosinophils and IELs, suggesting an important role for these cells in the innate defense against A. suum. Eosinophils can directly respond to a broad spectrum of pathogens through signaling via Toll like receptors, complement receptors and immunoglobulin receptors. In order to investigate whether eosinophils responded directly to A. suum L4 larvae, we monitored the release of reactive oxygen species from the eosinophilic granules after co-incubating the cells with the larvae. In contrast to results obtained with freshly hatched L3's, where eosinophil degranulation occurred quickly after contact with L3's in the presence of serum of either infected or uninfected animals [18], eosinophils did not respond directly to L4 Ascaris, even in the presence of serum. In addition, the larvae in the L4 transferred animals at 18 DPT were seemingly unharmed, even though eosinophil numbers remained high. These results may indicate that the L4 larvae are expressing inhibitory factors that prevent eosinophil degranulation and that eosinophils are better equipped to deal with tissue-residing larvae, rather than lumen dwelling ones. This seems indeed the case for many helminth infections [25]. The function of eosinophils in the defense against L4 larvae might also be of an indirect nature. Since the eosinophils were located deep in the mucosa, this assumption seems indeed likely. Through the release of preformed cytokines, chemokines, lipid mediators and cytotoxic molecules, eosinophils could quickly initiate a potent immune response after recognition of pathogen-associated molecular patterns, which in turn may lead to the initiation of the expulsion of A. suum.
Another important finding was a clear increase in the number of intra-epithelial T cells during the course of the infection. Although the IELs were not phenotyped, RNA transcription data would suggest that it was the cytotoxic T cell subset that was the most impacted, as there was an overall induction of molecules associated with cytotoxicity such as granzymes, perforin and NKG2D, all of which have been found to be expressed by IELs [26]. One of the functions of IELs is epithelial repair [26]. IELs may be activated in response to damage caused by the larvae. For example, Granzyme B has been found to be correlated with villus damage in helminth infections [27]. Our findings support this, as in all our experiments villous blunting and granzyme B upregulation were observed shortly after contact with A. suum larvae. The negative effect of A. suum on the intestinal structure might have important consequences for humans suffering from A. lumbricoides as well, as it might help to explain the malabsorption often associated with these infections [28]. Whether there is a direct effect of the IELs on the expulsion of the parasite deserves further attention, since resistance against helminth infections in sheep has been associated with genes involved in cytotoxicity [29]. Increased epithelial turnover and shedding caused by cytotoxic cells might make it harder for the small L4's to stick to the mucosa. Interestingly, IELs were lower in the L4 transferred group at 18 DPT, which may indicate an active regulation of the immune response by these larvae.
Mast cells and basophils have previously been associated with A. suum infections [18], [30], [31]. Repeated infections induced blood basophilia and intestinal mastocytosis, and these cells responded to stimulation with L3 or L4 secretory antigens by releasing histamine [30], [31]. The maximum histamine release occurred between 14 and 21 days after daily exposure, therefor it has been suggested that these cells played an important role during self-cure [32]. However, only basophils or mast cells that had previously been exposed to Ascaris released histamine following contact with L3 or L4 secretory antigens [30], [31]. We also show here that in contrast to experiments with repeated infections, mast cells were not induced in the small intestine after primary infections, suggesting that basophils or mast cells may only play a role in protection against secondary infections.
Interestingly, the local cytokine response in the jejunum seemed to be greatly impacted by the initial migration through the body. Naturally infected animals were more biased towards a Th1 type response with macrophages, while in both the L3 and L4 transfer experiments there was a much more mixed Th1/Th2 response and no recruitment of macrophages. Especially the animals infected with L4 larvae showed high il13 transcription at 18 DPT, which may indicate that the initial Th1 bias shifts towards a Th2 response as the infection progresses.
Together, these results suggest that the expulsion mechanism does not target the A. suum larvae directly. One possible mechanism by which larvae could be eliminated from the small intestine is increased gut movement. We show here that animals infected with A. suum indeed have decreased transit time around 17 DPI. This decrease is in agreement with a previous study showing an increase in smooth muscle contractility from 14 to 21 DPI and an increase in fluid secretion ex vivo [33]. Any increase in gut movement would indeed make it more difficult for the relatively small larvae to remain in the small intestine and may in fact be a universal mechanism of expulsion of intestinal lumen dwelling nematodes, as changes in intestinal smooth muscle contractility have been identified in Cooperia oncophora infected calves and Trichinella spiralis and N. brasiliensis infected mice [34]–[36]. Studies in mice have shown that the helminth induced increase in smooth muscle contractility is signaled through IL4 or IL-13 [36]–[38], which could explain why it is a common observation with helminth infections. Of particular interest is the contribution of alternatively activated macrophages on the regulation of smooth muscle contractility [36]. While we could only detect an increase in macrophages in the A. suum egg infected animals, it remains possible that changes in the activation state of macrophages contribute to the change in smooth muscle contractility.
Taken together, this study indicates that the self-cure is a locally initiated mechanism., Faster gut movement will make it harder for the larvae to remain in the small intestine. It is also part of a weep and sweep response that is often associated with helminth infections and which consists of increased luminal secretion (weep) and increased gut movement (sweep) [39]. This effect can probably be overcome once A. suum larvae have developed to a point where they are large and active enough to counteract the increased peristaltic movements. Eosinophils and intra-epithelial T cells appear to play a pivotal role since they are consistently associated with self-cure, but further research is needed to elucidate how these cells operate in order to induce the weep and sweep response.
|
10.1371/journal.pcbi.1003655 | An Expanded Notch-Delta Model Exhibiting Long-Range Patterning and Incorporating MicroRNA Regulation | Notch-Delta signaling is a fundamental cell-cell communication mechanism that governs the differentiation of many cell types. Most existing mathematical models of Notch-Delta signaling are based on a feedback loop between Notch and Delta leading to lateral inhibition of neighboring cells. These models result in a checkerboard spatial pattern whereby adjacent cells express opposing levels of Notch and Delta, leading to alternate cell fates. However, a growing body of biological evidence suggests that Notch-Delta signaling produces other patterns that are not checkerboard, and therefore a new model is needed. Here, we present an expanded Notch-Delta model that builds upon previous models, adding a local Notch activity gradient, which affects long-range patterning, and the activity of a regulatory microRNA. This model is motivated by our experiments in the ascidian Ciona intestinalis showing that the peripheral sensory neurons, whose specification is in part regulated by the coordinate activity of Notch-Delta signaling and the microRNA miR-124, exhibit a sparse spatial pattern whereby consecutive neurons may be spaced over a dozen cells apart. We perform rigorous stability and bifurcation analyses, and demonstrate that our model is able to accurately explain and reproduce the neuronal pattern in Ciona. Using Monte Carlo simulations of our model along with miR-124 transgene over-expression assays, we demonstrate that the activity of miR-124 can be incorporated into the Notch decay rate parameter of our model. Finally, we motivate the general applicability of our model to Notch-Delta signaling in other animals by providing evidence that microRNAs regulate Notch-Delta signaling in analogous cell types in other organisms, and by discussing evidence in other organisms of sparse spatial patterns in tissues where Notch-Delta signaling is active.
| The nervous system of many animals, including the marine invertebrate Ciona intestinalis in our study, develops through a cell-to-cell communication mechanism called Notch-Delta signaling. Mathematical models for Notch-Delta signaling have been developed that can explain the development of animal nervous systems with a dense arrangement of neurons. However, there are several cases where the spatial arrangement is much more sparse; we found that the peripheral nervous system of Ciona is one such example. Here, we develop an expanded mathematical model that is able to account for this sparser spacing, and furthermore demonstrate that the spacing can be widened or shortened through changing a single parameter that is influenced by the concentration of a regulatory microRNA called miR-124. The underlying differential equations contain only two variables representing the activity levels of Notch and Delta, and are thus general enough to be applicable to a wide variety of physical and biological systems that exhibit a similar sparse patterning.
| Differentiation of tissues during early animal development as well as tissue homeostasis during adulthood requires constant communication between cells. One of the most common ways by which cells communicate with each other is through the Notch-Delta signaling pathway [1]–[4]. Notch-Delta signaling is a fundamental cell-to-cell communication mechanism whereby a membrane-bound Delta ligand in one cell binds to a membrane-bound Notch receptor in a neighboring cell, generating a particular downstream response that depends on cellular context [1], [5]. Studies in several animals have shown that Notch expression is both temporally and spatially widespread [2]–[4], [6], [7]. It is not surprising, then, that Notch-Delta signaling is involved in the development and homeostasis of many tissues, most notably those of the nervous system [7], but also within the heart, kidney, liver, pancreas, breast, inner ear, prostate, thyroid, respiratory system, immune system, and many other cell types (reviewed in [1]).
Although the specific molecular factors and interactions are remarkably complex and vary among different organisms and cell types, the core Notch signaling pathway is relatively simple and is conserved across all bilaterian animals [1], [3]. The core pathway consists of five main components: a Notch receptor, a CSL family transcription factor (TF), the Hairy and Enhancer-of-split (Hes) family of TFs, the basic helix-loop-helix (bHLH) proneural TFs, and a Delta ligand (Figure 1). In most animals there are multiple genes that encode each component. For example, mammals have four Notch receptor genes and at least seven genes for Hes family members that mediate Notch-Delta signaling in different tissues [8], [9].
Most importantly, experimental studies have shown that neighboring cells, which communicate via Notch-Delta signaling have opposing expression patterns of these five core components [1], [5], [10]. In the signal-sending or Notch-suppressed cell, only the bHLH proneural TFs and Delta are constitutively active, while Notch and Hes expression are suppressed. This suppression is thought to be mediated in part through cis-inhibition of Notch by Delta within the same cell [2], [11], [12], and through loss of signaling feedback because Delta is downregulated in the neighboring cell [13], [14]. Conversely in the signal-receiving or Notch-activated cell, Notch and Hes are active, while Delta and bHLH proneural gene expression, even if initially active, are eventually suppressed by a Hes family member [5], [10]. Notch-Delta signaling is often used in a process called lateral inhibition, where the signal-sending cell eventually differentiates into one cell type while inhibiting the signal-receiving cell from adopting the same developmental fate [15]–[17]. Finally, the transcription factor CSL functions as a repressor of Hes family members in the signal-sending cell but becomes an activator of Hes genes in the signal-receiving cell [18], [19]. This functional switch of CSL from repressor to activator occurs when the intracellular domain (ICD) of Notch translocates to the nucleus where it displaces a co-repressor complexed with CSL [2].
With this biological background in hand, several mathematical and computational models have been developed over the years to try and quantitatively explain the dynamics of Notch-Delta signaling [12], [20]–[24]. These Notch-Delta models usually fall into one of two categories: comprehensive models and minimal models. In comprehensive models, all of the experimentally validated (and sometimes solely computationally predicted) molecular components are represented as separate variables, and all of the known or predicted interactions are represented as separate equations in the model [23], [24]. Although complex, these models have led to some key insights into the specific dynamics of particular Notch-Delta pathway genes. For example, one model that incorporated extensive feedback between Notch, CSL, and Hes resolved the long-standing issue that Hes can act both as a bistable switch and as an oscillator by showing that the transition between these two states can occur by tuning a single parameter, the Hes1 repression constant [23]. Another model incorporating Goodwin-modified biochemical kinetic equations for transcription, nuclear export, translation, and DNA-binding and dimerization of each factor showed the importance of the decay rate of Hes1 [24]. However, one drawback of comprehensive models is that they are usually based on experimental data from one particular cell type and, therefore, are not generalizable to other systems.
By contrast, in minimal models only the core molecular components and interactions, which capture the overall, essential Notch-Delta signaling dynamics, are represented in the differential equations. Unlike comprehensive models, minimal models have the advantage of being applicable to many biological contexts and are also more amenable to parameter sensitivity and stability analyses, which can shed important insight into the dynamics of the system. The first minimal Notch-Delta model was published by Monk and colleagues [20], which at its core is a simple two-cell model with a feedback loop involving just two variables: Notch and Delta. Because the core cascade is essentially linear, they postulated that the Notch variable could represent the quantity of activated Notch protein (i.e., Notch ICD) in the cell or the quantity of downstream Hes TF [20]. The production functions representing Notch-Delta interactions could be modeled using Hill functions, which are commonly used to model protein-protein as well as protein-DNA interactions [12], [20], [25] and for which we now have extensive experimental confirmation through biochemical studies [12], [26]. Through their model, Monk and colleagues demonstrated that such a feedback model results in a checkerboard spatial expression pattern of Notch and Delta, which mimics the Notch-Delta pattern found in several biological contexts for which lateral inhibition occurs [20], [21], [27]. With lower cooperativity (i.e., a lower Hill coefficient), occasionally a spacing of two or three cells can occur [20]. Subsequent models over the next several years were for the most part variations of the original Monk model (e.g., [21], [22]). Eventually, growing experimental evidence of cis-inhibition of Notch by Delta led to an updated model by Elowitz and colleagues that incorporated this interaction [12]. Such cis-inhibition was thought to facilitate Notch-Delta lateral inhibition, and indeed the expanded model resulted in faster dynamics, sharper checkerboard patterning and greater robustness to noise [12].
While the Monk and Elowitz models can explain the patterning in some biological systems such as ciliated cells in the early Xenopus ectoderm [21], there are cases in both invertebrates [28]–[34] and vertebrates [7], [35]–[37], where Notch-Delta signaling is clearly active but the pattern is not checkerboard. In many cases, the pattern is much more random and sparse, where the spacing between signal-sending cells can range from a single cell to dozens of cells in between [30], [31], [33]. For example, studies in zebrafish and chick neuroepithelial tissues have demonstrated a gradient of expression for Notch and/or Delta [7], [36], [37]. Also, the sensory organ precursor (SOP) cells of the Drosophila thorax that give rise to microchaetes are spaced about five cells apart when fully developed [5], [28]–[30], [38]. A pair of studies demonstrated that SOPs in wild-type Drosophila extend dynamic projections called filopodia, and that these filopodia express graded amounts of Delta along the filopoidia and allow the SOPs to reach out and activate Notch signaling in non-neighboring cells [30], [31]. Another form of extended communcation in Notch signaling can occur through a process called lateral induction, in which a Delta-bound Notch receptor in the signal-receiving cell can induce the expression of other ligands, which signal Notch in downstream cells [39]–[41]. Several authors analyzed more generalized models[42]–[44] with nearest neighbor or juxtacrine inhibition and induction and found these systems could generate Turing solutions[45] from a homogeneous steady-state with various wavelengths. Thus, a model for a juxtacrine system can produce stable periodic patterns with larger spacing between peaks of Delta activity. Hence, in addition to neighboring-cell lateral inhibition, a form of communication leading to long-range patterning can also operate in the context of Notch-Delta signaling. Since these filopodia are wide at the base but gradually thin out towards the tip, this suggests a concentration gradient where cells touching near the base of filopodia receive stronger Notch activation compared to cells in contact with the tips.
In this report, we present a minimal Notch-Delta model, which expands upon the previous Monk and Elowitz models [12], [20] by adding a simple nearest-neighbor Notch gradient term that makes it possible for the system to exhibit long-range effects on cell morphogenesis. We show that incorporation of a Notch activity gradient term is able to produce a sparse pattern of Delta expression whereby Delta-expressing cells can be spaced many cells apart. In our studies, we focus on the patterning of larval tail epidermal sensory neurons (ESNs) within the peripheral nervous system (PNS) of the ascidian Ciona intestinalis. We quantify the number and spacing of ESNs in wild-type larvae, and show that our expanded Notch-Delta model accurately reproduces the experimentally observed ESN pattern [33], [34], [46]. Ascidians are invertebrate chordates and are the closest invertebrate relatives of vertebrates [47]. As such, they occupy an important phylogenetic position for understanding how molecular developmental pathways evolved when invertebrates and vertebrates diverged from their last common ancestor [34], [48]. Sensory neurons, like those in the Ciona intestinalis PNS, the mechanosensory bristles found in Drosophila, and the hair cells of the mammalian inner ear, are thought to have evolved from a common ciliated sensory-neuron precursor [34], [49]. Since Notch-Delta regulated tissues in flies, ascidians, zebrafish, chick and mice have all been shown to exhibit sparse spatial patterning [7], [30], [31], [36], [37], our model suggests that Notch-Delta-mediated long-range inhibition may be broadly conserved in bilaterians.
We also demonstrate that regulation of Notch-Delta signaling by microRNAs (miRNAs) is conserved across bilaterians. The miRNAs are a class of conserved small RNAs that regulate expression of target genes through transcript destabilization, deanylation and/or translational inhibition, leading to downregulation of the protein product [33], [50]. Previously we demonstrated that in Ciona the miRNA miR-124 downregulates Notch and all three Hes factors, and that these operate in a negative feedback loop [33]. Here, we show that miRNA-mediated regulation of Notch signaling can be incorporated into the parameter representing the decay rate of the Notch variable, and that modulation of the Notch decay rate in the model accurately mimics the ESN pattern observed in wild type larva and in miR-124 overexpressing transgenic larvae that have altered ESN spacing patterns. Finally, through a bioinformatics analysis we demonstrate that the majority of miRNAs expressed in sensory cell types of other animals are predicted to target Notch pathway genes in their representative systems, suggesting that miRNA interactions with the Notch signaling pathway may be functionally conserved.
In Ciona intestinalis, the tail epidermal sensory neurons (ESNs) differentiate from epidermal precursor cells within the dorsal and ventral midlines. Previous work in our lab and others [33], [34], [46], [51] has qualitatively shown that the midline ESN pattern is very irregular, although a quantitative investigation of the number, spacing and distribution of ESNs has not been done. Thus, we began by quantifying ESN numbers and ESN spacings in wild-type embryos by immunohistochemically-labeling the associated cilia with an anti-acetylated tubulin antibody. We focused on an older developmental stage (22 hours post-fertilization at ), when the larvae have extended their tails and when the final midline ESN pattern has emerged [32]–[34]. To identify the midlines, we generated transgenic embryos expressing either an Acete-Scute homolog(ASH) RFP reporter or a Delta RFP reporter (see Materials and Methods) [34]. To identify the ESNs, we used fluorescent microscopy to image cilia in embryos immunohistochemically detected with an antibody against acetylated-tubulin. ESN cell nuclei are smaller than those found in the surrounding epidermal cells, and could be visualized with DAPI staining [32].
Figure 2 shows a representative embryo used for quantitation. In agreement with previous qualitative observations [33], [34], [51], we found that the number, distribution, and spacing of ESNs varied considerably from embryo to embryo ( embryos quantitated across three independent biological replicates). Overall, we found no obvious differences between the number of midline cells, number of ESNs or the spacing between ESNs along the dorsal versus ventral midline at 22 hours post-fertilization (see Figure S1). Therefore, we only considered statistical averages per midline without distinction between dorsal and ventral counts. No larvae had fewer than six ESNs per midline, consistent with previous observations that six dorsal midline precursor cells express Delta early in embryogenesis prior to midline formation [32]. We observed as many as eleven ESNs along a single midline in 22 hr larvae. We never observed more than eight or nine ESNs in earlier embryos ( hours post-fertilization) [34], suggesting that ESNs continue to be specified as the larval midline develops. We observed a variable pattern in ESN spacing with as few as one and as many as thirteen epidermal (non-ESN) cells separating consecutive ESNs. We never observed two ESNs next to each other, consistent with the hypothesis that Notch-Delta-mediated lateral inhibition is active between neighboring ESN-epidermal cells [32], [33]. These results are summarized in Figure 3A–B. Regarding the distribution of ESNs, we found no apparent bias of ESN position along the anterior/posterior axis. However, we did observe that consecutive ESNs spaced at least ten cells apart were almost invariably flanked on at least one side by two or three ESNs spaced very closely (Figure S2).
With this quantitative experimental data in hand, we began drafting a Notch-Delta mathematical model that could adequately explain the patterning of midline ESNs in Ciona. We began with a linear array of cells representing a single midline at a fixed time point. As mentioned, we did not notice any obvious differences between the dorsal and ventral midlines at the larval stage (see Figure S1), so our model is appropriate for modeling either midline. Future models will modify this static array into a dynamic array that includes cell division. This 1-D model could also be easily expanded to a 2-D array for modeling planar systems such as the proneural clusters in Drosophila [5], [12], [20], [30].
Consistent with previous minimal models, each cell tracks the activity of just two biochemical species, Delta () and Notch () or some closely affiliated biochemical species, such as a transcription factor directly linked to these primary proteins. Note that because our model can be applied to other biochemical and physical systems, when we present the differential equations of our model below, we will denote the Delta and Notch species more generally as and , respectively. As discussed in the original Monk model [20], could be taken to represent the quantity of activated Notch (i.e., Notch ICD) in the cell; or it could be taken to stand for the quantity of downstream Hes TF in the cell. In addition, since the Notch-SuH-Hes cascade is linear and exhibits bistability (i.e., there are only one of two stable states for each node - either all "ON'' or all "OFF''), we can regard the states of Notch, SuH and Hes as equivalent, and can therefore consider any of these or all of these lumped together as the variable [52]. Analogously, since we know that the bHLH proneural genes are expressed in a linear cascade and are upstream of Delta [34], could represent the quantity of membrane-bound Delta in the cell or could incorporate the activity of the upstream proneural TFs [52].
Figure 3C shows a schematic of our model for the interaction between neighboring cells. All the cells in the linear array interact with their nearest neighbors with the exception of the end cells. The model localizes inside the cell or expressed on the cell surface to signal only the neighboring cells. It is repressed internally by and activates neighboring cells to stimulate production of . The species also catalyzes the cis-inhibition of inside the same cell. The production of depends on the activity of in the neighboring cells. Both species have linear decay terms based on the half-lives of Notch, , and Delta, . Finally, we include a communication term for to neighboring cells based on the gradient in activity of active Notch or a related biochemical species between the cells. The addition of this gradient term is the primary distinction of our model from previous Notch-Delta models. In earlier models, interactions are exclusively with neighboring cells, which restricts the patterning to primarily alternating on and off states, while our model by including a Notch activity gradient can simulate larger cell spacings, which match that found in Ciona and in other analogous Notch-Delta systems [7], [30], [37]. Although the exact mechanism of long-range communication is currently unknown in Ciona, we favor a nearest-neighbor Notch gradient term versus other possibilities based on our current biological knowledge of Notch-Delta signaling in the Ciona PNS (see Discussion).
All of the above interactions represent the core conserved interactions of Notch-Delta signaling and are supported by extensive experimental evidence [4], [5], [10], [53]. Let and be the activity levels of Delta and Notch in cell , respectively, then the dynamics for the model described above is given by the following system of differential equations: (1)
In the system above, we let the boundaries satisfy: where and are the average activity levels of Delta and Notch over the entire array of cells. Clearly alternate boundary conditions could be considered, although other common boundary conditions such as zero or periodic boundary conditions are not appropriate for modeling the Ciona midline.
The functions and the parameters in the model are common in biochemical control models [12], [20], [25], [54]. The essential form of each function is the same as those found for earlier minimal Notch-Delta models [12], [20] (Figure 3D). A full explanation of each of these functions and parameters can be found in Materials and Methods, but here we briefly mention the functions and parameters that are immediately relevant for our analysis. The first term on the RHS of the equation represents cis-inhibition by . The parameters and are the linear decay rates of Delta and Notch or a related biochemical species, respectively. Because our biochemical species do not distinguish between mRNA and protein levels, we may take them as representing mRNA and/or protein decay rates. The last term in the equation is the linear gradient term representing long-range communication. This cell-to-cell gradient term could result from bound Notch molecules self-signaling to create a gradient-like pattern of activity. It could be the result of another signaling biochemical closely aligned with Notch, but not necessarily bound so strongly to the membrane. From a modeling perspective this gradient form of nearest neighbor communication is the simplest mechanism of long-range patterning and makes a good first order approximation to the kinetic interactions of this signaling pathway. For the remainder of the article, we will refer to as and as to associate the model state variables with the Delta () and Notch () pathways.
We wrote programs to simulate our Notch-Delta model using the Matlab solver. We began our simulations with random low activity levels of and in all cells and first observed the qualitative behavior of our system over time. After some time passed, a few cells developed a high level of . The high level of in Cell suppressed in the same cell (cis-inhibition) and led to above average levels of in Cells and (lateral inhibition). Via the linear gradient term, subsequent neighboring cells had decreasing levels of , until some critical threshold was reached with sufficiently low that another cell could once again produce a high level of , then the pattern repeated. The dynamical system exhibited very stable behavior for the levels of and in the immediate region near Cell . However, we observed decreasing stability of the activity levels as levels of decrease.
When there was sufficient spacing between cells with high levels of , then we observed later development of cells with high levels of in the intervening area of cells. These later developing cells arose from two distinct dynamical behaviors. In one case there were sufficiently low levels of far from the ones with high levels of , resulting in the smooth development of an intervening cell with a high level of . This case was most common early in the simulation. In the second case, the levels of and oscillated in the regions between stable areas of high , with the amplitude of the oscillations appearing to increase with increased ESN spacing. With enough spacing, the oscillations increased until a threshold was crossed, allowing the development of another cell with a high level of .
Because of the random initial conditions, different patterns of cells with high levels arose. The spacings in these patterns depended strongly on the parameter values; however, after sufficient time a stable pattern emerged. A representative example is shown in Figure 3F. Note that spacings of more than two cells cannot be achieved with either the original Monk model [20] nor the model incorporating cis-inhibition [12] (Figure 3E).
To determine if our model could explain the ESN pattern along the Ciona midline, we ran a Monte Carlo simulation with = 1000 runs over = 4000 time steps for each run, and compared the number, spacing, and distribution of high Delta-expressing cells with that of the ESNs from wild-type embryos. Our simulations used the parameter values listed in Table 1.
The parameters were chosen for the following properties. The value for , the number of cells, was chosen to match the average number of midline cells from our experiments. The parameters , , , , and were fairly arbitrary, although they were chosen based on our knowledge of similar biochemical control models from previous work [12], [20], [25], [54]. As off-diagonal elements, these parameters should not be as significant to the behavior of the system as the other parameters (though the -mediated decay could be an important parameter when considering the effect of modulating cis-inhibition, as in a previous study [12]). The most significant parameters for the switching behavior are the parameters and , the Hill coefficients. These are chosen be be greater than one, but not too large to be biologically relevant. The decay rates and along with the gradient parameter are very significant as we will see in the bifurcation analysis. In particular, will be important when we consider the effect of microRNA-mediated regulation of Notch signaling. For these simulations, was adjusted so that the average number of high-Delta cells over the 1000 runs closely matched the number of ESNs from wild-type experiments. Since Delta is an epidermal sensory neuron marker [34], throughout the text we will refer to high-Delta cells and ESNs interchangeably.
Figure 3F shows the end results of a typical run, with Movies S1 and S2 showing the dynamics of two separate runs starting with random low initial conditions for both Delta and Notch. Both movies show the appearance of new ESNs in regions where the spacing between existing ESNs is large. In movie S1, the levels of Notch and Delta settle into a very stable equilibrium; while in movie S2, the levels of Notch in the cells between the ESNs at Cells 27 and 39 show distinct stable oscillations. Figure 3A–B shows the statistics for the number and distribution of ESNs and inter-ESN spacing from 1000 runs. While agreement between the average number of ESNs predicted by the model and experimentally observed in larvae is expected, surprisingly the distribution of ESNs and the average ESN spacing matched very well with experimental observations. The majority of runs in our Monte Carlo simulations produced between 6 and 11 ESNs, with a peak of 9 ESNs, matching experimental observations. There were some instances of outliers on either side in our simulations, although if we were able to quantify an equivalent number of embryos (), we might expect some experimental outliers as well. Similarly, the ESN spacing in our simulations matched experimental observations, with the frequency histograms following an identical gamma distribution with a peak at 4 cells and dropping off after 13 cells. There were a few rare outliers where ESN spacing exceeded 13 cells. When we analyzed these outliers more closely, we noticed that these large spacings were flanked on at least one side by two closely ESNs (Figure S2). These closely spaced ESNs likely stabilize the cells within the large-spacing valley. This is in agreement with our experiments showing that cases of high inter-ESN spacing were flanked on at least one side by consecutive ESNs with tight spacing (Figure S2). Finally, we note that our model has a disproportionate number of one-cell spacings compared with experimental observations. This is likely due to the intense stability of the high-Delta cells and the strong effect of lateral inhibition in our model.
We chose our Hill coefficients and based on our knowledge of previous biochemical control models [12], [20], which produced the reasonable fits seen in Figure 3A–B. However, we know that changing the coefficients, and , affects the lateral inhibition and induction of immediately neighboring cells and results in differing distributions of cell spacing. Simulations with and produced significantly broader distributions (similar means, but a much larger variance), while and produced a much narrower distribution (similar mean with a smaller variance). Our modeling experiments suggest that increases, especially in , would produce more two-cell spacings at the expense of one-cell spacings as suggested in the experiments. However, since Figure 3A–B shows our model adequately represents the experiments, we chose to center our studies around the case and .
A stability analysis is used to determine equilibrium states of a system and the change in behavior of a system as the parameter values vary. This analysis is important because it allows us to determine the possible ESN patterns that can be produced from our model, and to rigorously determine if our model can really explain the biology. We therefore designed programs to help numerically find equilibria and allow the stability analysis of the equilibria. The stability analysis uses the Jacobian matrix analytically derived from linearizing the system (1) (see Materials and Methods).
There is a unique homogeneous equilibrium for system (1). Related systems [20], [42]–[44] have been analyzed in terms of the stability of the homogeneous equilibrium, showing the existence of Turing solutions. For system (1) with the parameters in Table 1, there is a homogeneous equilibrium with and , which is unstable with multiple positive eigenvalues. Since the experimental studies do not suggest a periodic pattern, we did not explore Turing solutions. Our primary interest was the behavior of the many inhomogeneous equilibria.
The Monte Carlo simulations showed the variety and large number of possible stable equilibria for model (1). This model can easily reproduce the stable alternating pattern of the previous Monk [20] and Elowitz [12] models. These models are very similar to (1) with = 0 and = 0, respectively; however, non-zero values of and allow the richer stable patterns shown in the Monte Carlo simulations. From the many equilibria for this system we chose to systematically explore the stability of the system with different spacings of high levels. The numerical observations showed decreased stability of the cells some distance from the cells with high levels, so we wanted to explore the nature of any bifurcations leading to limits on the spacing of the cells. Below we present the stability analysis for different ESN spacings, giving information about the dominant eigenvalues and commenting more about the observed eigenvalue structure. The parameters we use in this analysis come from Table 1. In biological terms, the eigenvalues and eigenvectors tell us the differentiation state of each of the midline cells. Roughly speaking, if a cell aligns with an eigenvector associated with the most negative eigenvalues, then it is stable and has fully differentiated into an ESN. The cells that align with the largest components of the eigenvectors associated with eigenvalues with positive real part are unstable and remain bipotent.
To help minimize the effects of the boundary, we varied the number of cells in our simulations to be as close as possible to (which is the average number of midline cells found in all of our experiments), while maintaining symmetry at the boundaries. Suppose two consecutive ESNs are Cell and Cell , then define (1 ESN and epidermal cells). We numerically find the equilibrium of (1) for each value of . From the linearized form computed in Materials and Methods, we can readily find the eigenvalues and eigenvectors for this system. Table 2 summarizes the results of different spacings using the parameters from Table 1 and shows the dominant eigenvalues of the system.
The linear stability analysis of (1) with the parameters from Table 1 and the spacings and numbers of cells from Table 2 gives a better understanding of this system. The overall stability of system (1) is determined by the real part of the dominant eigenvalue, , with this system being asymptotically stable if and only if . However, this is a high-dimensional system, and different components of the model behave differently near an equilibrium based on its structure. The time-series local behavior of different components vary more or less depending on their location, and their fate can be understood by careful examination of the eigenvector associated with specific eigenvalues.
With MatLab we computed all eigenvalues and eigenvectors for each of the cases in Table 2. In every case we had the smallest eigenvalue with a multiplicity matching the number of cells with high levels of . By examining the corresponding eigenvectors, we found the largest components centered on the highest (lowest ) values. (Note that because of the scaling, the components of the eigenvectors are much smaller than the components, so we compared only relative size within or components.) Each of the eigenvectors associated with one of the eigenvalues, , had a large component and a large component at one of the ESN positions with all other components at least four magnitudes of order smaller. This agrees with our observation that the model produces extremely stable regions near cells with high levels of , i.e., differentiated ESNs.
The real part of the dominant eigenvalue, , becomes larger as the spacing, , increases. This correlates to the decreasing stability of the levels of and as the spacing increases. The multiplicity of matches the number of interspacings between cells with high . When examining the particular components of the corresponding eigenvectors, the patterns were more complex, spreading across several interspacings. However, the maximum -component occurred near the center of the interspacings with the maximum -components flanking either side of the maximum . This is in line with the observation that the next highest -component always occurs near the middle of our cells with high levels of , while the flanking cells show the highest responses in agreement with Notch being highest in cells neighboring a cell with high Delta.
As increases, the real part of changes signs between and , giving a Hopf bifurcation. Figure 4A–B shows the equilibrium state of the system at and 13, and the simulations show distinct oscillations. From Table 2, any simulation with would show damped oscillations with the solution settling to the equilibrium. The eigenvalue for has a frequency of 0.1877, which implies a period, . Figure 4C–D shows the oscillatory solutions from a simulation with , and the period of oscillation agrees with the frequency of . The eigenvectors of with show a structure very similar to the graph in Figure 4D, where variation for each cell from its equilibrium is displayed. The variation in is very small (about 1%) compared to the size of the high Delta cells, while the oscillations in are quite substantial relative to the equilibrium Notch levels, especially in the cells flanking the cell, which has the greatest variation in near the middle of the interspacing region. This example with has an unstable equilibrium, but its oscillations are insufficient in magnitude to cross a threshold and pass to a different equilibrium with high Delta cells between the ones shown in Figure 4B. We note that slightly different initial conditions away from the equilibrium will cause new ESNs to arise, indicating that the basin of attraction for the equilibrium shown is quite small.
Once , our numerical algorithms cannot find an equilibrium solution to linearize around and any simulation results in new ESNs appearing, indicating the spacings are too unstable when evenly spaced. Thus, our model suggests that when the number of cells between ESNs becomes too large, then new ESNs appear in between. Importantly, in agreement with this bifurcation analysis on , our wild type experiments show a maximum spacing of 13 cells between ESNs. This suggests that if the midline cells divide and the spacing becomes greater than 13, the instability of such a state will cause a new ESN to appear. Also recall that with our parameter values the spacing mean and distribution matched the wild type experiments. Thus, our experimental results are in harmony with our numerical analysis of the spacing, .
The analysis above examines discrete changes in the spacing, . We next chose to explore continuous changes with the gradient parameter . For these studies we set , , and all other parameters from Table 1 except for . From the analysis above we know that instabilities should cause an ESN to appear midway between and create a pattern. Our interest is to determine something about the dynamics of change from a larger spacing, , to a smaller spacing, .
Decreasing in essence shortens the effective distance of Notch signaling. As noted before, when , the Monk model only produces an alternating pattern of high and with no spacings larger than two and most being one. Thus, we expect the stability of the pattern to be lost as decreases. We studied the linear stability of the pattern as ranged from 0.2 to 0.08845. At the ESNs, where is high, , the smallest eigenvalue is , making this region of the cellular array extremely stable. The maximum eigenvalue, has its eigenvector centered between the cells with high . Figure 5A shows the variation in the real part of as varies. When we decrease to , there is a Hopf bifurcation (verified with Auto in XPPAUT), introducing oscillations in cellular activity levels, and . The maximal oscillations in occur in the middle cells, , while the maximal oscillations in occur in the adjacent cells, e.g., Cells 9 and 11. Figure 5B–C shows the equilibrium levels for and , and the maximum and minimum of the oscillating levels after the Hopf bifurcation. As is typical of a Hopf bifurcation, these oscillations increase in amplitude away from the Hopf point.
As decreases further to approximately 0.0885, the instabilities are sufficient that the solution leaves the basin of attraction for the equilibrium. The result is that the solution converges to the very stable pattern where is high at , resembling the equilibrium. The maximum eigenvalue for this solution is , producing a very stable equilibrium. We note that the basin of attraction for this solution is significantly larger than the basin of attraction for the case. Figure 5B–C shows the increase of both and as decreases. It appears as though some threshold is reached, which results in approaching 100 and going to very low levels quickly. It is not clear if this transition is smooth and very rapid or if some saddle node bifurcation is occurring. At this time the specific type of bifurcation moving from the to the spacing has not been determined and needs further analysis.
Finally, we analyzed the change in behavior of the system as we increased the Notch decay rate parameter, . We began with a constant spacing of cells and the corresponding value of from Table 2, with all other parameters from Table 1. Starting with a low value of , we increased the value of with a step size initially of 0.01. As we stepped from to , a significant change in the system occurred whereby new ESNs appeared halfway between existing ESNs, similar to what occurs when we decrease . Through repeating this stepping process with decreasing step sizes, we determined the exact value of this critical value of to be . With every iteration of this process, we kept track of the minimum and maximum eigenvalues and associated eigenvectors (Figure 6A), as well as the equilibrium values of and (Figure 6B).
As in the case of , analysis of the min/max eigenvalues and associated eigenvectors revealed that the existing ESNs (e.g., Cell 5) are highly stable, while the middle intervening cells (e.g. Cell 10) are in regions of lower stability. However, unlike with , the levels of and do not exhibit oscillations as we approach the critical value (Figure 6B). The real part of the maximum eigenvalue remains negative as we vary , indicating that there is no Hopf bifurcation (Figure 6A). At , the system moves out of the basin of attraction for and converges to a new stable pattern with smaller spacings resembling the equilibrium (Figure 6C–D). The behavior in Figure 6B is similar to a saddle node bifurcation, but a more detailed analysis is required. As we decrease back to , the system remains in the new equilibrium, indicating that this equilibrium is very stable and has a very large basin of attraction. In biological terms, we may interpret this hysteresis effect as the newly formed neurons have committed to their new state and will not easily revert back to being bipotent.
Significantly, our analysis shows that increasing beyond a critical value can produce new cells with high levels of , which demonstrates that, based on our model, increasing the Notch decay rate can produce new ESNs. This directly relates to our consideration of the influence of microRNAs on Notch decay rates and ectopic ESN formation in the last two sections.
In the study of any model, it is important to determine which parameters have the greatest effects on the system. Our model is a high dimensional, nonlinear model with a large number of equilibria, so one would expect that the sensitivity of the model depends on the region of parameter space where the analysis is performed. Some equilibria will have large basins of attraction and will therefore be very robust to parameter changes, while other equilibria will have smaller basins of attractions and will be more sensitive. For this parameter sensitivity analysis, we examine variations of % in each of the parameters for our case where and , using the other parameter values from Table 1. This equilibrium is associated with a pattern of six neurons with 9 cells between each neuron, and we chose to focus on this equilibrium since this was the mean spacing and neuron count found experimentally and therefore should give us a general idea as to which parameters have a greater effect on our system. We established that the equilibrium for this system was stable and found the eigenvalues.
One measure for the sensitivity is the change in the value of the real part of the maximum eigenvalue. With the base parameters, we found . Figure 7A shows that increasing the coefficient of the negative feedback function, , has the greatest effect, and even results in the system going through a Hopf bifurcation. Decreasing the parameter has the next largest effect, which is not too surprising given that its parameter value is close to the Hopf bifurcation for that parameter. As we would expect, the parameters, , and have minimal effect on the eigenvalues, while the other parameters have more varied effects increasing or decreasing the stability. Figure 7A shows the effects of variations of % for all the parameters on the real part of the largest eigenvalue, .
Our study shows that in the case where and , the greatest instability lies in the center between two ESNs. This can be visualized by examining the eigenvector for . The largest level of away from the ESNs occurs at , ,… (see Figure 6C). The least stable levels of occur in the neighboring cells, such as and (see Figure 6C). Figure 7B–C provide information on how much a variation of % in a given parameter shifts the equilibrium values at and , where changes in amplitude are observed to be the largest. When a shift becomes sufficiently large at and a threshold is crossed, a new ESN forms in this location, completely changing the equilibrium values for and . Figure 7 shows that a description of parameter sensitivity for this system depends on the measure that is employed. Clearly, this system is most sensitive to the negative feedback coefficient, . However, the Hill coefficients relate to the degree of cooperativity for binding between Notch and Delta, which are intrinsic properties of the proteins not likely to change over development. As we alluded to before though, decreasing broadens the ESN count and spacing distributions, while increasing narrows the distributions (Figure S3). We found that a 10% change in caused a –25% shift in the ESN spacing and count distributions, suggesting that the system is indeed sensitive to this parameter (Figure S3). The least significant parameters are the kinetic constants, , , and . A 10% variation in results in a 10% change in ESN count and spacing distributions (Figure S3). The robustness of this model to variations in most of the parameters allows reasonable stability for patterning of ESNs, while providing flexibility to produce novel patterns when new adaptations are necessary, such as a need for more or less dense ESNs.
We previously showed that the microRNA miR-124 is expressed in the larval midline ESNs of Ciona intestinalis [33], [46]. We demonstrated that miR-124 is activated by proneural bHLH genes and negatively regulates Notch signaling by downregulating Notch and all three Hes genes [33], [34] through binding to canonical target sites in the corresponding transcript [33]. Mis-expression of miR-124 along the entire epidermal midline increases the number of midline ESNs presumably because of ectopic suppression of Notch signaling [33], although a detailed quantitative analysis was not performed.
Here we generated transgenic embryos using this same miR-124 construct from our previous studies (Epi::miR-124) [33], [46]. We electroporated increasing amounts of the transgene into Ciona embryos (, , or ; which we denote as Epi::miR-124+10, Epi::miR-124+20, or Epi::miR-124+30, respectively). In each case, we quantified the number and spacing of ESNs at 22 hours post-fertilization, and compared these results to control wild-type 22 hr embryos. We used immunohistochemistry to detect ESN cilia with an anti-acetylated tubulin antibody, and visualized the midlines with either an Ash or Delta fluorescent transgene reporter. We performed each experiment with independent biological replicates, and quantified a total of 17, 19 and 20 embryos for the miR-124+10, miR-124+20 and miR-124+30 experiments, respectively. We only quantified embryos for which we could clearly perform cell counts for both the dorsal and ventral midlines; since miR-124 overexpression produces kinked or twirled phenotypes that make counting difficult, we were not able to quantitate as many embryos as in the wild-type experiment.
Figure 8A–B shows a representative embryo, and the results of our Epi::miR-124 titration experiments are shown in Figure 8C–D. As we increased the amount of the miR-124 transgene electroporated into embryos, the mean number of ESNs per midline increased with a corresponding decrease in the mean ESN spacing. Note that the mean number of midline cells was very similar between the experiments (mean = 57.7, 57.0, 57.3, 58.9 in wild-type, +10, +20 and +30, respectively), indicating that miR-124 overexpression did not affect the number of midline cell divisions during development (see Figure S1). The largest difference occurred between wild-type and miR-124+10 embryos (difference in mean ESN counts = 2.44; difference in mean ESN spacing = −2.44); subsequent increases in miR-124 concentration had a linear effect on the number and spacing of ESNs (average difference in mean ESN counts = 1.45; average difference in mean ESN spacing = −1.43). Comparison of ESN count and spacing distributions and the associated minimum/maximum values among the different miR-124 concentrations also showed a shift towards an increasing number of ESNs per midline and decreasing inter-ESN spacing. In particular, the number of zero-spacing cases (i.e., adjacent ESNs) increased as the concentration of miR-124 was increased. A magnified region of the embryo in Figure 8A shows one such case of adjacent ESNs (Figure 8B), which we did not observe in wild-type embryos. This suggests that when expressed at high levels, miR-124 is able to mitigate the effect of lateral inhibition.
Since miR-124 downregulates Notch and Hes by base pairing to their transcript and likely mediating decay at the post-transcriptional level [33], [55]–[58], we proposed that miR-124 regulation of Notch/Hes could be modeled into the Notch decay rate, . An increase in miR-124 concentration would thus be reflected in our model by an increase in the value for . To test this hypothesis, we began with the value from our wild-type simulations () and ran Monte Carlo simulations (M = 1000) continuously increasing the value of to see if we could match the average number and spacing of high-Delta cells with the number of ESNs in our miR-124 titrations. In agreement with our experiments, we showed in the previous section that continuously increasing eventually resulted in the formation of new neurons, suggesting that production of extra midline ESNs could be explained by an increase in the Notch decay rate. Indeed, as we continued to increase , the average number of ESNs continually increased. Eventually, we found values for which both the mean ESN counts and inter-ESN spacing closely matched the observed values in each of the miR-124 titration experiments ( for miR-124+10; for +20; for +30, Figure 8C–D). The marginal increase in is greatest from wild-type to miR-124+10 embryos, correlating with the high marginal increase of ESN counts between these two samples. Importantly, the distributions of ESN counts and spacings closely fit the experiments, with the model also showing corresponding shifts in the distributions upon increasing gamma values (Figure 8C–D). The variance of the model is smaller than the corresponding miR-124 experiments, even though in the wild-type case the variances were similar between model and experiment. However, overall the model agrees very well with the miR-124 experiments, even more surprising given the fact that we can mimic the experimental ESN patterns with the tuning of just a single parameter. Coupled with extensive biological support [33], [55]–[58], we conclude that our model can accurately incorporate the experimental effect of miR-124 into the Notch decay rate term.
Notch signaling regulates the specification and patterning of sensory cell types not just in Ciona, but throughout metazoans (reviewed by [5], [10], [26], [53]). Examples of processes regulated by Notch-Delta signaling include the mechanosensory bristles (macrochaetes and microchaetes) of D. melanogaster [30], [31]; the inear ear hair cells of zebrafish [59]–[61], chick [41], [62], and mouse [41], [63], [64]; and the multiciliated cells derived from the respiratory airway epithelium in humans [65], [66]. Interestingly, the sparse patterning in Ciona appears also to be found in other animals [7], [30], [31], [35], [37], suggesting that long-range Notch-Delta signaling is also conserved.
Since the inner ear hair cells of vertebrates are likely evolved from the sensory neurons of invertebrates [26], [35], we originally hypothesized that miR-124 regulation of Notch signaling, as we described for Ciona [33], should be conserved. However, we found very little published evidence of miR-124 regulating miR-124 outside of ascidians other than miR-124 regulation of Hes1 in the mouse inner ear [67]–[69]. Our own bioinformatic analysis showed that miR-124 rarely targets Notch pathway genes in other organisms [33]. Interestingly though, miR-9 in Drosophila appears to regulate Notch signaling in a somewhat analogous fashion [70], suggesting that different organisms deploy different miRNAs to regulate Notch signaling [33]. This would suggest that incorporation of miRNA function into the Notch decay term of our expanded model may be relevant for other systems.
To determine if this might be the case, we first examined the literature to identify sensory neuron-expressed miRNAs in Drosophila, zebrafish, mouse, and human. For miR-124, we only found one study in mice where weak miR-124 expression was reported in the vertebrate inner ear [67]. However, many other miRNAs are highly expressed during mouse inner ear development [67], [71], [72]. In other bilaterians, different miRNAs are expressed in these sensory cells, with no obvious conservation of particular miRNA expression (Drosophila: [9], [73], [74], zebrafish: [71], human [65], [66]).
We then bioinformatically searched for canonical target sites of these sensory miRNAs in the of Notch pathway genes in these animals using a target prediction program we developed previously [33]. Through this, we discovered the presence of predicted target sites in the primary Notch receptor (Notch1) among vertebrates, as well as target sites for other Notch homologs in zebrafish and mouse (Figure 9). In agreement with our hypothesis, we found Notch1 target sites for different sensory miRNAs in each of the different organisms (miR-124 in Ciona, miR-15a in zebrafish, miR-30b, −100, −125b, −133a, −182 and 183 in mouse; and miR-34 and miR-449 in human airway epithelium). Among these, miR-34 and miR-449 targeting of Notch in human airway epithelial tissue has been experimentally verified [65]. We did not find any target sites for Notch in Drosophila, suggesting that such sensory miRNA regulation of the Notch receptor did not evolve until at least after ecdysozoans.
In agreement with previous reports, we also observed sensory miRNA target sites within many Hes homologs in Drosophila [9], [75]. However, whereas in Drosophila and Ciona almost all of the Hes homologs have target sites, we observed that in vertebrates predicted targeting of Hes is much more restricted (Figure 9). This may be explained by the fact that predicted targeting of the Notch receptor appears to be much more extensive in chordates (Figure 9). Since the Notch receptor is the initial effector of Notch signaling, miRNA-mediated suppression of Notch would relieve the need to target all of the downstream Hes factors. Another possible explanation is that the other Hes factors are not expressed in the sensory cells of vertebrates, and therefore their targeting by sensory miRNAs is not needed. Indeed, among the many Hes homologs in mice, only Hes1 and Hes5 are expressed in inner ear cells, of which Hes1 is the more highly expressed factor [27], [76].
Finally, we note that although in C. elegans there is no published evidence of Notch signaling regulating sensory neuron formation, the Notch homolog LIN-12 regulates the formation of some of the adjacent interneurons that relay signals from the sensory neurons [77]. Recent evidence suggests that the miR-51-56 family is ubiquitously expressed among neurons in C. elegans [78], and we bioinformatically found a canonical target site for this family of miRNAs in the LIN-12 . Although in this work we focused on sensory cell types in other organisms, since they are most analogous to the epidermal sensory neurons of Ciona, but it would be interesting to explore miRNA regulation of Notch signaling in other cell types.
Previous Notch-Delta models [12], [20]–[24] were based on early Notch signaling studies in Drosophila, Xenopus, and mouse [2], which suggested a checkerboard expression pattern whereby neighboring cells adopted alternate cell fates. This was supported by evidence in Drosophila that cells selected to become neurons activate Notch signaling in neighboring cells thereby preventing these cells from likewise adopting a neuronal fate [5]. This led to the classic Monk model [20], which provided the foundation for later models [12], [21]–[24].
However, more careful analysis has shown that the pattern produced by Notch-Delta signaling in some cases is not checkerboard. For example, the sensory microchaetes of the Drosophila thorax are initially formed at every other cell and prevent immediately neighboring cells from adopting a sensory fate via lateral inhibition. However, once the thorax has fully developed, these microchaetes become spaced about 4–5 cells apart. Meanwhile, the larger macrochaetes can be spaced dozens of cells apart [29], [30], [38]. In these cases, it has been suggested that dynamic filopodia extensions may provide a mechanism whereby the Delta ligand can activate Notch signaling in non-neighboring cells [28], [30]. Other examples of experimentally observed non-checkerboard patterning include sparse patterning of bristle cells in other fly species [38]; opposing gradients of Notch versus Delta expression along the apical-basal axis in the developing retina of both mouse and zebrafish [7], [37]; and gradient expression of Notch in the mouse inner ear [41]. These examples suggest the need for updated Notch-Delta models that can reproduce these non-checkerboard patterns. One such model has recently been developed for describing the sensory bristle patterning in Drosophila incorporating filopodia extensions [31]. This model requires dynamic lengthening and shortening of filopodia and incorporates data on several variables such as length of filopodia, lifetime of filopodia and sensitivity of Notch signaling to the Delta ligand specific to their experiments. More general models of juxtacrine systems have explored periodic patterning with longer wavelengths, producing sparser patterns also [42]–[44].
Here, we developed an expanded Notch-Delta model that builds upon the minimal equations established first by Monk and later Elowitz and colleagues [12], [20]. Our model incorporates a simple activity gradient that allows for long-range cell communication through juxtacrine (cell-cell) signaling. This is actually a long-range inhibition, mediated by local juxtacrine signaling, using a linear gradient term similar to Fickian diffusive flux. As mentioned earlier, specific examples of Notch-Delta patterning in Drosophila [28]–[30] and other fly species [38], as well as in the mouse inner ear [41], mouse retina [37] and zebrafish retina [37] all demonstrate that sparse or gradient neuronal patterns can arise from a field of neurocompetent cells. Unlike the model for Drosophila neuronal patterning [31], our model does not require the existence of dynamic filopodia extensions, and actually makes very few assumptions regarding the exact pattern of neurons and the underlying mechanisms responsible for neuronal patterning. Our model can produce a large number of possible equilibrium states and, although here designed for a linear array of cells, is easily adaptable to a planar field of cells. Therefore, we suggest that our model is adaptable and able to reproduce a variety of both sparse and dense spatial patterns, and should be useful for modeling other Notch-Delta systems.
In this study, we applied our model to the patterning of sensory neurons in the peripheral nervous system of Ciona intestinalis larvae. In a previous report [33], we found that the array of cells along the Ciona midlines are all neurocompetent and can be converted into neurons by inhibiting Notch signaling. However, in wild-type animals only a few of these cells are selected to become ESNs. Specifically, the spatial pattern of ESNs in the larvae of Ciona intestinalis is sparse and irregular, with variable ESN spacing ranging from one to thirteen cells between consecutive ESNs in wild-type animals. The large number of non-ESN cells found between one ESN and the next demonstrates the need for an extra term for producing long-range ESN patterns along the midline. This is the motivation for updating the previous Monk and Elowitz models with the addition of a Notch gradient term. For this study, we represent this long-range term as a simple activity gradient, and demonstrate that this is sufficient for explaining the patterning of ESNs in Ciona. We note that this is not necessarily a diffusion term, since Notch and Delta are membrane-bound and in most cases do not produce diffusible species. We chose a linear gradient over other possibilities, such as Hill function interactions [44], because it is the simplest and most generic form, and can be applied to a wide variety of biological and physical systems without assuming anything about the underlying mechanisms of long-range communication. One possible mechanism of long-range communication via Notch signaling in Ciona may be through the protein -fibrinogen, which is secreted from the tail notochord and is known to interact with Notch in the Ciona central nervous system [79]. -fibrinogen is similar to the fibrinogen-like protein Scabrous, which is involved in producing large-cell bristle spacings in Drosophila [28]. We will explore this and other possibilities in future studies and will update our model accordingly.
Finally, we provide a strong mathematical foundation for our model by performing rigorous stability analyses and bifurcation analyses of the key model parameters: the neuron spacing (), the Notch decay rate (), and the slope of the linear gradient (). The sensory neurons in Ciona derive from bipotent precursor cells along the tail midline, which adopt either an epidermal or neuronal fate [32]–[34]. Our eigenvector/eigenvalue analysis for different spacings between neurons demonstrates that the cells committed to becoming neurons occupy regions of high stability, while epidermal precursor cells more centrally located between consecutive neurons occupy regions of instability. These centrally located cells thus maintain their bipotent character. These cells may have very small basins of attraction for maintaining low levels of Delta and thus, are sensitive to perturbations and small changes in the parameters of the system. As we varied the parameters , , and , we discovered a threshold phenomenon whereby the system increasingly loses stability to a point where it jumps to a new equilibrium with these central cells becoming neurons. For and , these cells exhibit a hysteresis effect and remain committed to a neuronal fate (i.e., express high levels of Delta), even if the parameters are adjusted back to their original values.
An early study using both in situ hybridization and immunostaining demonstrated an apical-to-basal gradient of Notch expression within neuroepithelial precursor cells in the diencephalon, telencephalon, retina and spinal cord during chick development [36]. More recently, apical-to-basal expression gradients of Notch were also found within neuroepithelial cells in the zebrafish retina, where Notch-Delta signaling is active [37], [80]. The nuclei within these neuroepithelial cells are able to migrate along the basal-apical axis, and depending on where these nuclei are within the Notch gradient, after mitosis the daughter cells either remain in their precursor state or differentiate into neurons. Although these studies were examining intra-cellular gradients, this motivated us to consider the possibility that Notch gradients exist between cells along the midline. Intercellular gradients induced by cell-cell signaling relays have been well-established for TGF- family signaling [44], [81], and although not yet definitely shown to cause gradient patterns, signaling relays also exist in the context of Notch-Delta signaling through Notch activation of secondary relay ligands such as Jagged/Serrate [2], [41].
We found here (Figure S4) and also in previous studies [32]–[34] that Delta expression is restricted to the presumptive ESNs and is not expressed in the other midline cells, therefore a Delta-mediated gradient is not appropriate. Studies of the morphology of Ciona sensory neurons found no evidence for dynamic filopodia extensions in the PNS [51], and so the Cohen model is also not appropriate [31]. Conversely, Notch is expressed in all midline cells and, therefore, could mediate long-range communication [6]. Also, from our previous experiments [33], [34], we know that blocking midline Notch signaling using a dominant-negative form of the dowstream effector gene Suppressor-of-Hairless results in ectopic neuron formation along the entire midline. On the other hand, ectopic activation of Notch signaling along the entire midline through mis-expression of Delta causes a reduction in midline neuron formation and large regions without ESNs [34]. Thus, given our experimental knowledge in Ciona and knowledge of long-range patterning in other systems, our current hypothesis is that the Notch signal is somehow relayed from ESN-neighboring cells to more distant cells. Therefore, the most reasonable term to add to the original Collier model, given our experimental observations, would be a Notch activity gradient. From a dynamical systems perspective, this is also the simplest form in our model that can produce distal spacing patterns. This Notch gradient may be produced through lateral induction of secondary Notch ligands as in other animals [2], [41]. In Ciona, it is known that -fibrinogen interacts with Notch to regulate neuronal patterning in the central nervous system [79]. Given that a similar Notch ligand, Scabrous, is involved in producing long spacings in the Drosophila PNS [28], it is possible that -fibrinogen may also act as a Notch ligand in the PNS as well. We will be exploring these and other possibilities in the future. Overall, the linear Notch gradient term provides a simple initial model, which explains the Notch-Delta-mediated patterning of sensory neurons in Ciona based on our current biological knowledge of the Ciona PNS, and motivates future experiments and updated models.
The model is a high dimensional system of ordinary differential equations with many equilibria and 11 parameters, including the number of cells in the system. Scaling could be used to eliminate three parameters, but that still leaves 8 parameters. We provided detailed studies for the parameters , , and , which are significant in Ciona, and demonstrated when bifurcations occur, leading to new ESNs forming. In order to examine the stability of ESN patterning, we have conducted some initial bifurcation studies, finding Hopf bifurcations and indications of hysteresis effects through saddle node bifurcations. In the future, more detailed bifurcation studies will be performed to determine the exact type of bifurcation occurring when the miR-124-related parameter is varied. In addition, we performed a sensitivity analysis for all the parameters about an equilibrium of six high-Delta ESNs with a mean ESN spacing of nine cells to show the relative effects of each parameter as they varied, thus giving a local understanding of the most significant parameters. The model did prove to be quite robust for this equilibrium, producing similar ESN patterns for a range of each of the parameters. We do note that the system was sensitive to small changes in . This could suggest that an organism has limited variability in its cell to cell communication, or this could be a potential limitation of our model. More experimental evidence is needed to decide the precise nature of the long-range inhibition, and it is possible that our model will require additional nonlinear juxtacrine signaling functions from lateral induction and/or inhibition. For this current work, we have added a simple linear gradient term and have shown that this is sufficient for producing the long-range patterning of ESNs in Ciona.
In order to fit our experimental observations, future work needs to be done on modifying the local interaction terms. Monte Carlo simulations of our model produce too many one-cell spacings compared to what we observe in wild-type larvae. When we adjusted the Notch decay rate parameter, , in order to simulate the miR-124 overexpression experiments, the model was not able to produce the zero-spacing adjacent ESNs of miR-124 overexpressing embryos. There are several possibilities for these discrepancies. In the miR-124 experiments, wild-type ESNs endogenously produce miR-124, therefore the actual Notch decay rate is much higher within ESNs compared to the other midline cells. In our model we use a single 'average' value for all midline cells that does not take this variation in Notch decay rates into account. Indeed, if we increase significantly (), we are able to override lateral inhibition and produce adjacent high-Delta cells. Thus, a more appropriate model may be one that incorporates a spatially-varying , whose form perhaps follows a Gamma distribution. Another possibility is that the level of lateral inhibition (i.e., the strength of the Hill equations) is too strong in our initial model, and that tuning of the Hill coefficients may allow for production of adjacent ESNs more easily. Finally, there may be other yet unidentified local factors that counterbalance the feedback effect between neighboring cells, which we have not accounted for in our model. We will explore each of these possibilities in future studies.
Finally, although our model is motivated by our studies on Notch-Delta signaling in Ciona, we emphasize that it can also be applied to many other biological and physical systems. At its core, we have developed a general mathematical model involving two chemical species, and , which interact locally as well as over a distance. Local interactions involve a positive and negative feedback governed by Hill functions, which were originally derived by Goodwin [25] to model the reaction kinetics between two biochemical species and for which extensive experimental evidence exists [12], [20], [25]. Distal interactions are governed by a linear activity gradient, which is the simplest and most generic gradient form. Since the specific mode of distal interaction has not yet been determined in Ciona, this gradient is appropriate, since we do not assume these are diffusible species and are making no assumptions about the biological mechanism of long-range patterning. The presence of distal interaction greatly expands the number of possible equilibrium states of this system. Finally, as seen from our bifurcation analyses of several parameters, this is a high-dimensional system rich with at least hundreds of possible equilibrium steady states and a variety of interesting dynamics for which we have only begun to explore in this report. Since our model involves only two species and a minimal set of parameters, it is applicable not only to Notch-Delta systems, but is general enough to be applied to analogous biological and physical systems that exhibit both local and distal effects.
All of our transgene vectors were cloned in a pSP72 vector backbone (Promega) containing an SV40 Poly(A) site [33], [34], [46], [82]. To visualize expression in the midline, we used two promoter constructs fused in frame with an optimized form of yellow or red fluorescent protein [83]. The first, Ash, contained the conserved cis-regulatory and promoter region of the Acete-scute homolog, which showed expression along dorsal and ventral midlines in tailbud embryos [34]. The second construct, Delta, contained a conserved cis-regulatory and promoter region as well as the conserved first intron of the Delta2 gene, which is expressed in the Ciona PNS [34]. We generated transgenic embryos by electroporating 10-15 µg of each construct into fertilized and dechorionated embryos as previously described [46], [83]. Both of these constructs showed midline expression with occasional ectopic expression elsewhere in the epidermis, although the number of expressing midline cells varied from embryo to embryo. This is due to the fact that the genes themselves turn off early in the midline, although the fluorescent proteins have a half-life of and often remain expressed in the cells. DAPI staining of nuclei and acetylated tubulin antibody staining of cilia was performed as previously described [34]. Images were taken at 20× and 40× magnification with a Zeiss AxioPlan 2e fluorescent microscope equipped with an AxioCam HrM monochromatic camera.
Our expanded Notch-Delta model represented by the ordinary differential equations in (1) begins with a linear array of cells. Later we plan to modify this linear array to a dynamic array, which includes cell division. Each cell tracks activity levels of two biochemical species, and . Importantly, note that our model can be generalized to represent the signaling between any two biochemical species, although for this report we focus on Notch-Delta signaling. All the cells in the linear array interact with their nearest neighbors with the exception of the end cells. Here, our model uses average levels of species and as the missing neighbor for the end cells. The model localizes D inside the cell or expressed on the cell surface to signal only the neighboring cells. It is repressed internally by and activates neighboring cells to stimulate production of . The species also catalyzes the degradation of inside the cell. Both species have linear decay terms based on the natural half-lives of and . The production of depends on the level of in the neighboring cells. We also include a gradient term for based on the difference in Notch activity between the cells.
The functions and the parameters in the model given by the ODEs in (1) are common in biochemical control models. In the equation, the first function is a standard negative feedback or repression function. The parameters and are primarily scaling parameters in the production of . The most significant parameter is , which is the Hill coefficient and reflects the strength of the negative feedback. The higher the value of , the more effective works as a repressor in the production of . It is well-known that this parameter should significantly affect the stability of the system with larger values increasing instability. The parameter affects the half-life or linear decay of . From the indexing in the equation it can be readily seen that production and decay of is completely contained in the cell where is produced.
The equation is more complicated. The first term represents enhanced cis-inhibition of by inside the cell. Thus, accelerates the degradation of its repressor with a scaling parameter . The second term is a non-linear positive feedback or induction function, which has surface molecules of on neighboring cells signaling the production of . The parameters and are scaling parameters, while the parameter is the Hill coefficient representing the strength of the positive feedback. Again, the higher the value of , the more switch-like the behavior of this production term for by the levels of in neighboring cells. The parameter is the linear decay rate for . The last term is a gradient term for communication of between neighboring cells. This is a standard linear gradient term for flux of with a rate of between cells with differing activity levels of . The value of will affect the range of communication or signaling of with higher values of corresponding to longer range signaling.
The system (1) was coded and simulated in Matlab (R2008b, revision 20) using the solver. Stability and bifurcation analyses on the parameters , and , as well as all Monte Carlo simulations were performed using custom Matlab scripts.
For stability analysis we need to linearize the system (1). We let and and write the system (1) as follows:
We assume an equilibrium solution and , then we define the perturbed variables from the equilibrium as and . The linearized version is written where is the Jacobian matrix with
The submatrices , , , and are created with their row and column satisfying:
Submatrices and are relatively simple with only diagonal form. All diagonal elements for are . Since is the equilibrium value for , the diagonal element for satisfies:
The diagonal form of the matrices and reflect that the substance is confined to the cell. The diagonal elements in reflect the linear decay of in the model. The diagonal elements in reflect the production of , which is repressed by .
Since the variable of the model is produced and communicated based on the neighboring cells, the submatrices and are predominantly tridiagonal. The values and are the equilibrium values for and . For the submatrix , the diagonal elements are predominantly
This term reflects the enhanced degradation of by in the cell. The subdiagonal and superdiagonal elements come from the production term with most satisfying
This reflects the enhanced production of by in the neighboring cells.
For this version of the model we chose to make the boundaries substitute the average activity level for the end levels. This leads to small contributions in the and rows of . In addition to the terms listed above for we add the terms: and
For the submatrix , the diagonal elements are predominantly
This term reflects the enhanced degradation of by in the cell, the linear decay of , and part of the gradient term. The subdiagonal and superdiagonal elements come from the other terms of the gradient
Since the boundaries use the average activity level for the end levels, we obtain small contributions in the and rows of . In addition to the terms listed above for we add the terms: and
We noted that system (1) has many equilibria. For the Monte Carlo simulations we began the simulations with random low values (both and ) and allowed long simulation times for solutions to settle into a stable pattern of ESNs, which is determined by the ESNs, where . Most of these equilibria did not have their eigenvalues tested, so several of the stable patterns of ESNs, undoubtedly had eigenvalues with positive real parts, making system (1) unstable and leaving some cells to have low amplitude oscillations. (See Movie S2.) These oscillating cells could be considered bipotent, but the local environment remains sub-threshold, so they fail to convert to ESNs.
For the bifurcation studies in spacing, , the gradient parameter, , and the Notch decay parameter, , initial conditions were provided that favored a particular pattern. For some patterns with certain parameter values, the basins of attraction for the particular pattern were very large, and simulations easily settled to the desired pattern with initial conditions only roughly exhibiting the planned pattern. Nearer bifurcation points, the initial conditions required using equilibria from nearby (stabler) parameters. The equilibria used for stability analysis were found by a long time simulation "near'' a particular equilibrium. Subsequently, this simulated equilibrium had Newton's method with the Jacobian shown above applied to system (1) with the derivatives set to zero. The equilibrium results from the Newton's method were used for the local analysis described above to find eigenvalues and eigenvectors.
|
10.1371/journal.ppat.0040032 | The Landscape of Human Proteins Interacting with Viruses and Other Pathogens | Infectious diseases result in millions of deaths each year. Mechanisms of infection have been studied in detail for many pathogens. However, many questions are relatively unexplored. What are the properties of human proteins that interact with pathogens? Do pathogens interact with certain functional classes of human proteins? Which infection mechanisms and pathways are commonly triggered by multiple pathogens? In this paper, to our knowledge, we provide the first study of the landscape of human proteins interacting with pathogens. We integrate human–pathogen protein–protein interactions (PPIs) for 190 pathogen strains from seven public databases. Nearly all of the 10,477 human-pathogen PPIs are for viral systems (98.3%), with the majority belonging to the human–HIV system (77.9%). We find that both viral and bacterial pathogens tend to interact with hubs (proteins with many interacting partners) and bottlenecks (proteins that are central to many paths in the network) in the human PPI network. We construct separate sets of human proteins interacting with bacterial pathogens, viral pathogens, and those interacting with multiple bacteria and with multiple viruses. Gene Ontology functions enriched in these sets reveal a number of processes, such as cell cycle regulation, nuclear transport, and immune response that participate in interactions with different pathogens. Our results provide the first global view of strategies used by pathogens to subvert human cellular processes and infect human cells. Supplementary data accompanying this paper is available at http://staff.vbi.vt.edu/dyermd/publications/dyer2008a.html.
| Many pathogens, such as viruses and bacteria, cause disease in humans. Pathogen infections result in illness and death for millions of people each year. Pathogens communicate with human cells through physical interactions with various human proteins on the surface of the cell and within the interior of the cell. These interactions allow the pathogen to enter the host cell, manipulate important cellular processes, multiply, and invade other cells. In this paper, we compare interactions between human and pathogen proteins from 190 different pathogens to provide important insights into strategies used by pathogens to infect human cells. We show that both viral and bacterial proteins interact with human proteins that themselves interact with many human proteins or with human proteins that lie on many communication channels between other human proteins. Pathogens may have evolved to interact with these human proteins since they may control critical human cellular process. We also demonstrate that many viruses share common infection strategies, e.g., lengthening particular stages of the cell cycle, controlling programmed cell death, and interacting with the nuclear membrane to transfer viral genetic material into and out of the nucleus. Such studies may help us better understand the process of infection and identify better strategies to prevent or cure infection.
| Infectious diseases result in millions of deaths each year. Millions of dollars are spent annually to better understand how pathogens infect their hosts and to identify potential targets for therapeutics. An important aspect of any host-pathogen system is the mechanism by which a pathogen is able to invade a host cell. Within these complex systems, protein-protein interactions (PPIs) between surface proteins form the foundation of communication between a host and a pathogen and play a vital role in initiating infection [1]. PPI-mediated mechanisms of infection have been studied in detail for many pathogens [2–7]. However, many questions are relatively unexplored. What are the properties of human proteins that interact with pathogens? Do pathogens interact with certain functional classes of human proteins? Which infection mechanisms and pathways are commonly triggered by multiple pathogens? A significant hurdle to such global cross-pathogen comparisons has been the shortage of large-scale datasets of interactions between host and pathogen proteins. High-throughput experimental screens have been primarily used to identify intraspecies PPIs [8–16]. However, recent efforts to include host-pathogen PPIs in public databases have made it easier to acquire the data needed to address these important questions.
In this paper, we integrate experimentally verified human-pathogen PPIs for 190 pathogen strains from seven public databases [17–23]. We partition the strains into 54 different pathogen groups, where each group is made up of taxonomically related strains. We analyze the intraspecies network of PPIs between the 1,233 unique human proteins spanned by the host-pathogen PPIs, and find that pathogens, both viral and bacterial, tend to interact with hubs (proteins with many interacting partners) and bottlenecks (proteins that are central to many paths in the network) in the human PPI network.
We pay special attention to two networks of PPIs between human proteins: the proteins that interact with at least two viral pathogen groups (see Figure 1) and the proteins that interact with at least two bacterial pathogen groups (see Figure 2, noting that the figure also contains human proteins targeted by only one bacterial pathogen group). We used the Cerebral plugin [24] for Cytoscape [25] to render these images. We compute the Gene Ontology (GO) [26] functions enriched in each of these two sets of human proteins. Such enriched functions highlight human pathways that may be involved in infection mechanisms that are common to multiple pathogens. Examples of such processes and components include cell cycle regulation, I-κB kinase/NF-κB cascade, and the nuclear membrane. These functions shed light on a number of features shared by different pathogens: interacting with human transcription factors and key proteins that control the cell cycle; transport of genetic material through the nuclear membrane (in the case of viruses) to subvert the host's transcriptional machinery; triggering an immune response via toll-like receptors; and activation of NF-κB signaling. We discuss in detail the importance of these and other enriched functions, as well as the proteins they annotate and the pathogens they interact with. Overall, these results provide the first global view of aspects of human cellular processes that are controlled by and respond to pathogens.
Our results should be interpreted with caution since no single pathogen may target all the proteins and PPIs we analyze. In addition, data for bacterial pathogens are scarce. However, we suggest that piecing together targeted human proteins across multiple pathogens has the potential to provide insights into common molecular mechanisms of infection and proliferation used by different pathogens.
We use the term “pathogen group” to refer to a set of pathogen strains that are closely related taxonomically, i.e., they all belong to the same genus, or, in the case of viruses, the same family. We partition the 190 strains into 54 pathogen groups: 35 viral, 17 bacterial, and two protozoan. Nearly all of the 10,477 human-pathogen PPIs we collect are for viral systems (98.3%), with the majority belonging to the human-HIV system (77.9%). These human-pathogen PPIs involve 1,233 unique human proteins, of which 1,109 are known to interact with at least one other human protein. Of these 1,233 human proteins, 221 interact with at least two pathogen groups (182 with more than one viral pathogen and 20 with more than one bacterial pathogen).
Researchers have argued that the degree distribution of PPI networks is scale-free and follows the power law, i.e., the fraction of proteins in the network interacting with k other proteins is proportional to k−γ, for some γ greater than zero, typically between two and three [27,28]. One feature of such networks is that they are robust in the face of attacks on random nodes. For instance, the removal of random subsets of nodes increases the diameter of the network only gradually [29,30]. In this context, the diameter is defined as the average length of the shortest paths between all pairs of proteins. However, the selective removal of even a small number of nodes of high degree can dramatically change the topology of the network [29,30].
There is considerable debate on the origins of the scale-free property and whether this property is an artifact of experimental biases and errors [31–33]. Notwithstanding this debate, we reasoned that pathogens may have evolved to interact with human proteins that are hubs (those involved in many interactions) or bottlenecks (those central to many pathways) [34] to disrupt key proteins in complexes and pathways. (See Methods for a precise definition of “bottleneck.”) Our results support this hypothesis. Figure 3A displays the cumulative log-log plot of the degree distribution of four sets of proteins in the human PPI network: (i) all proteins, (ii) “Viral” set, the subset of proteins interacting with at least one viral pathogen group, (iii) “Bacterial” set, the subset of proteins interacting with at least one bacterial pathogen group, and (iv) “Multiviral” set, the subset of proteins interacting with at least two viral pathogen groups. We did not include the “Multibacterial” set of human proteins interacting with two or more bacterial pathogen groups in this analysis since there are only 20 such proteins. These plots show that across almost the entire range of degrees, proteins interacting with viral and bacterial pathogen groups tend to have higher degrees than human proteins not interacting with pathogens. Further, proteins interacting with at least two viral pathogens have higher degrees than proteins interacting with one or more viral pathogens. The betweenness centrality results display the same trend (see Figure 3B). Across the entire range of values, proteins interacting with viral and bacterial pathogens have higher betweenness centrality. These results suggest that pathogens may have evolved to interact with human hub and bottleneck proteins, perhaps because these proteins control critical processes in the host cell.
We used Gene Set Enrichment Analysis (GSEA) [35] to test whether the gaps we observed in Figure 3 are statistically significant. GSEA is a method developed to assess the significance of the differential expression of a pre-defined gene set in two phenotypes of interest [35]. GSEA ranks all genes by a suitable measure of differential expression (e.g., the t-statistic) and uses a modified Kolmogorov-Smirnov test to assess if the genes in the given set have surprisingly high or low ranks. Since distributions of the t-statistics of differentially expressed genes have been observed to follow a power-law distribution [36], we reasoned that GSEA may be appropriate to test whether the human proteins interacting with pathogens have surprisingly high degree or betweenness centrality.
Our GSEA results support the conclusions we draw from Figure 3 that pathogens preferentially interact with human protein hubs and bottlenecks: for each of the three sets of proteins plotted in Figure 3, GSEA yields a p-value of at most 3 × 10−5 (degree) and 2.3 × 10−4 (centrality). To alleviate the concern that the observed patterns may be artifacts of experimental biases or errors in the human PPI network, we repeated each of the analyses using two subsets of the human PPI network: a network composed of 13,324 PPIs detected only by high-throughput studies [14,15,37] and a network with 59,396 PPIs constructed using only manually curated interactions [20,23]. The top half of Table 1 summarizes these results. For all three networks, the viral set, the bacterial set, and the multiviral set are significant at the 0.05 level for both degree and centrality, with the exception of the multiviral set in the high-throughput network. Since 77.9% of the human-pathogen PPIs are for the human-HIV system, we repeated these analyses for each network after removing all human-HIV PPIs and obtained similar results (see the bottom half of Table 1). In Text S1, we discuss three analyses that show that the consistency in the GSEA results for degree and for centrality are unlikely to result from any correlation that may exist between a protein's degree and its centrality (Figure S1 and Table S1 accompany the discussion in Text S1). We note that Tables S2 and S3 of the supplementary data contain detailed information on the GSEA results for the groups in Figure 3 and for individual pathogen groups.
We computed over-represented GO terms in 58 sets of human proteins: the bacterial set, the viral set, the multibacterial set, the multiviral set, and the 54 sets of human proteins interacting with each of the 54 pathogen groups. Overall, we found 404 unique GO terms enriched in these sets. A complete list of enriched GO terms with images of the sub-networks spanned by the human proteins annotated with each term is available on the supplementary website.
We identified at least one enriched function in 21 pathogen groups. Analysis of these data identified 91 biclusters (see Methods for details), each containing between two and seven pathogen groups and between two and 40 enriched GO functions. We focus on two of the biclusters below. The biclusters demonstrate that our analysis can group different enriched functions together even if the effects of the interactions on the host cell or the participating host proteins are different.
Our first example is a bicluster spanning the three pathogen groups Adenovirus, HIV, and Papillomavirus and 23 GO functions. GO biological processes in the bicluster include “cell cycle process” and “regulation of cellular process.” GO cellular components in the bicluster include “membrane-enclosed lumen” and “pore complex.” The membrane-enclosed lumen is the space within a sealed membrane or between two sealed membranes. Proteins annotated with these functions include KPNA2, a karyopherin, the histone deacetylases HDAC1 and HDAC2, and a number of Transcription Factors (TFs). KPNA2 plays an important role in both the import and export of material through the nuclear membrane. Interactions with KPNA2 enable a virus to enter the nucleus and take over the host's transcriptional machinery [38–41]. HDACs play an important role in silencing gene expression by removing acetyl groups from histones, thus causing them to wrap more tightly around DNA and block the binding of TFs. The role played by pathogen-HDAC interactions varies among pathogen groups. In the case of Adenovirus, it has been suggested that the pathogen protein E1B interacts with HDAC1/SIN3 to produce an enzymatically active complex that may be capable of repressing the transcriptional activity of the human TP53 protein in order to block apoptosis [42]. In contrast, the E7 Papillomavirus protein binds to the HDAC complex to promote cell growth, eventually leading to cervical cancer [43].
The second example is a bicluster containing a virus (HIV) and three bacteria (Chlamydia, Neisseria, and Escherichia coli). This bicluster contains 11 GO functions including the biological processes “immune response,” “response to stimulus,” and “cytokine production.” Although these four groups of pathogens interact with proteins belonging to the same pathways, the functions of the interactions are different. In the case of the bacteria, these functions annotate such proteins as toll-like receptors (TLRs) and interleukin receptor-associated kinases (IRAKs), which are special classes of host proteins responsible for recognizing foreign material and activating an immune response. There are no reported interactions with these proteins and HIV, although some researchers suggest that the single-stranded RNA of HIV-1 may encode many TLR7/TLR8 ligands [44]. In contrast to the bacteria in the bicluster, HIV uses host proteins involved in immune response such as CD4, CCR5, and CXCR4 to gain entrance to the cell. HIV attaches to the host protein CD4, a T cell glycoprotein, and subsequently to host chemokine receptors CCR5 and CXCR4. These binding events cause conformational changes to host proteins that allow the membrane of the virus to fuse to the host cell membrane [1].
The biclustering analysis of the previous section suggests that specific sets of pathogen groups might trigger or target the same human pathways and processes. Encouraged by these data, we asked if there are infection pathways commonly targeted or triggered by at least two viral or bacterial pathogen groups. To answer this question, we constructed two networks of human proteins: one where every protein interacts with at least two viral pathogen groups and the other where every protein interacts with at least two bacterial pathogen groups. In each network, we included every PPI connecting two proteins in the network. Figures 1 and 2 display these networks. (Note that Figure 2 also contains human proteins that interact with only one bacterial pathogen group.) We computed the enriched GO functions in these two networks. We group and highlight some of the enriched functions and relevant sub-networks below. Throughout our discussion, we will refer to the localization of proteins in the four main regions of Figures 1 and 2: extracellular, the cell membrane, the cytoplasm, and the nucleus. For every GO function that we discuss, we mention its p-value and rank in the sorted list of all functions enriched in the corresponding network.
Our analysis highlights a number of important mechanisms that viral pathogens use to manipulate the human cell: (i) control the host cell cycle program to ensure the transcription of viral genetic material; (ii) utilize human TFs to promote the transcription of viral genetic material; (iii) target key human proteins that regulate critical cellular processes such as apoptosis; and (iv) subvert host machinery for transporting material across the nuclear membrane.
Although the number of human-bacteria PPIs gathered in this study is small (only 174), our methods identified an important subset of human proteins enriched for functions involved in immune response and interacting with multiple bacterial pathogen groups. Figure 6 displays a subset of the multibacterial set that is enriched in four GO functions: “immune system process” (p-value 1.397 × 10−9, rank 1/28), “response to wounding” (p-value 3.93 × 10−4, rank 8/28), “immune response” (p-value 0.002, rank 14/28), and “I-κB kinase/NF-κB cascade” (p-value 0.012, rank 18/28). The proteins contained in this image are located in the top-right corner of Figure 2.
These functions are tied together by the Toll-Like Receptors (TLRs) and the protein IRAK1 found in the network in Figure 6. TLRs are a special class of cell-surface proteins that play a role in recognizing the presence of a pathogen and activating an immune response against the pathogen. The TLR/IRAK complex stimulates the activity of NF-κB [86–88], a complex of proteins that act as a TF for activating the production of a set of proteins in response to stimuli such as stress, cytokines, and bacterial or viral antigens.
The human TLRs and IRAK1 protein interact with the pathogen proteins FLIC (E. coli), HSP60 (Chlamydia), and PIB (Neisseria) [20]. FLIC is a flagellin protein. TLR4 and TLR5 contain a specific innate immune receptor for recognizing bacterial flagella [5,89]. HSP60 is a heat-shock protein that stimulates an immune response via TLR2 and TLR4 [90]. PIB is an outer membrane protein that is known to be recognized by TLR2, TLR4, and TLR9 [7].
Another human protein included in this network is HLA-DRA, which is part of the major histocompatibility complex (MHC). The MHC plays an important role in the immune system. HLA-DRA belongs to the class II MHC; proteins in this class belong to the lysosomal compartment of the cell, which contains digestive enzymes that kill engulfed foreign particles such as viruses or bacteria. The two bacterial partners for HLA-DRA are Mycoplasma and Staphylococcus [91,92]. In the case of Mycoplasma, the interacting partner is the MAM superantigen, which is known to contribute to autoimmune disease by activating proinflammatory monokines such as interleukin 1β and the tumor necrosis factor α [93].
The networks in Figures 1 and 2 contain a number of other human proteins targeted by more than two pathogen groups. We discuss two of these proteins—STAT1 and EP300.
Viral pathogens also interact with other human proteins involved in immune response pathways that are not included in the network in Figure 6. An example is the human protein STAT1. When the cell recognizes the presence of foreign material, it activates an immune response as a defense mechanism to either remove the foreign material or cause the cell to undergo apoptosis. During this process, STAT1 is tyrosine- and serine-phosphorylated and forms a homodimer known as IFN-γ-activated factor (GAF). GAF migrates to the nucleus where it binds to specific cis-elements to drive the cell to produce interferons, agents that inhibit viral replication within other cells of the body [94]. STAT1 interacts with Adenovirus, HIV, and Hepatitis [95–97]. Hepatitis POLG is part of the pathogen core complex that allows the virus to avert host antiviral response by binding to host STAT1 and inhibiting its activity [98].
Within the nucleus, we see pathogens target the human protein EP300, a histone acetyltrans-ferase that regulates transcription via chromatin remodeling. EP300 interacts with Adenovirus, HIV, Papillomavirus, and Polyomavirus [99–102]. The pathogen Adenovirus targets human EP300 via E1A. E1A is an oncoprotein that stimulates cell growth and inhibits differentiation by binding to the EP300/CBP complex and deregulating cellular transcription programs [103]. Papillomavirus protein VE7 shares many functional and structural similarities with E1A and is an interacting partner of human EP300. The disruption of normal growth conditions brought about by the E1A-EP300 interaction leads to the development of cervical cancer [104]. In the case of HIV, the viral TAT protein targets human EP300. The resulting complex regulates TAT transactivating activity and may assist in the integration of viral genetic material into human DNA [105].
We have provided a general overview of the landscape of human proteins interacting with pathogens and demonstrated that pathogens preferentially interact with two classes of human proteins: hubs (i.e., proteins that interact with many other human proteins) and bottlenecks (i.e., proteins that lie on many shortest paths) in the human PPI network. We identified GO functions over-represented in human proteins interacting with pathogens. Biclustering analysis demonstrated that many sets of pathogen groups target the same processes in the human cell, even if they interact with different proteins.
We constructed networks of PPIs between human proteins that interact with at least two viral pathogen groups and with at least two bacterial pathogen groups. Consideration of the GO functions enriched in these networks provided insights into numerous pathways targeted or triggered by multiple pathogens: control and deregulation of the cell cycle; import of pathogen proteins into the nucleus in an attempt to subvert the host's DNA replication and transcription machinery; manipulation of host cellular programs such as apoptosis; immune response and activation of NF-κB pathways via the TLR/IRAK complex.
A striking aspect of this network is that human proteins that mediate pathogen effects are often proteins in cancer pathways (e.g., RB1, TP53, and STAT1). We note that only some of the pathogens targeting such proteins are known to cause cancer themselves (e.g., Herpesvirus and Papillomavirus). In fact, a number of parallels are becoming evident between infection and cancer; for instance, in the part that TLRs play in angiogenesis and their potential as targets for therapeutics [106,107] and the role that viruses may play in the development of inflammatory diseases and cancer [108]. Cell cycle regulators and many TFs have been extensively studied in the context of mediating tumor formation. Our observation that they are also communication vehicles for pathogens suggests that the link between pathogen infection and cancer may be worthy of further experimental studies.
An important outcome of such a comparative study is the identification of human proteins to target experimentally for developing therapeutics. We provide a file on the supplementary website that contains the degree, centrality, the number of pathogen interactors, and the most specific annotations in each of the three GO hierarchies for each human protein that interacts with at least one pathogen protein. We provide this data as a resource for researchers interested in prioritizing antiviral and antibacterial targets.
We reiterate that our results should be interpreted with caution since no single pathogen may target all the proteins we analyze. As interactions between host and pathogen molecules are discovered on genome-wide scales [109], computational analyses such as those presented in this paper may provide a more detailed understanding of the landscape of host pathways and processes that pathogens target.
We downloaded all datasets used in this study in August 2007. We gathered 10,477 experimentally detected and manually curated protein-protein interactions (PPIs) between human and pathogen proteins and 75,457 experimentally verified PPIs between human proteins from primary literature [109] and seven databases: the Biomolecular Interaction Network Database [21], the Database of Interacting Proteins [19], the Human Protein Reference Database [23], IntAct [18], the Molecular INTeraction database [17], the Munich Information Center for Protein Sequences [22], and Reactome [20]. Table 2 contains statistics on the experimental methods that yielded these PPIs and the literature support for the PPIs. These interactions cover 190 different pathogen strains. Two pathogens—HIV and Hepatitis—account for 88.4% (9,268) of the human-pathogen PPIs. To mitigate this bias, we merged pathogen strains into 54 groups based on taxonomic similarity: each group contains pathogens belonging to the same genus, or, in the case of viruses, the same family. The 54 pathogen groups contain 35 viral, 17 bacterial, and two protozoan groups. We constructed lists of unique human proteins interacting with each group. Table 3 summarizes the number of interactions acquired for each pathogen group. For some analyses, we consider a human PPI network assembled from unbiased high-throughput experiments [14,15,37] and a network constructed from only manually curated human PPIs [20,23]. These networks contain 13,324 and 59,396 interactions, respectively. We obtained functional annotations from the Gene Ontology (GO) [26].
We represent the set of known interactions between human proteins as an undirected graph G(V, E), where V is the set of nodes (proteins) and E is the set of edges (interactions). Let M be the set of pathogen groups. We say that a pathogen group P interacts with a human protein s if s interacts with a protein in P. For a pathogen group P ∈ M, we define VP ⊆ V to be the set of human proteins that interact with P. Let T = ∪P∈M be the set of proteins that interact with at least one pathogen. Let TV (respectively, TB) be the set of human proteins that interact with at least one viral (respectively, one bacterial) group. Let T(k)V ⊆ TV (respectively, T(k)B ⊆ TB) be the set of human proteins that interact with at least k viral (respectively, k bacterial) pathogen groups; by definition, T(1)V ≡ TV and T(1)B ≡ TB. We now describe in detail the tests we use to analyze TB, TV, T(2)B, T(2)V, and the 54 VP sets.
The degree of a protein in a graph is the number of interactions in which it participates, not including self-interactions. We plot distributions of the degrees of four sets of proteins in G: (i) V, the set of all proteins in G; (ii) TB, the set of all human proteins interacting with at least one bacterial pathogen group; (iii) TV, the set of all human proteins interacting with at least one viral pathogen group; and (iv) T(2)V, the set of human proteins interacting with at least two viral pathogen groups. In this analysis, we ignore T(2)B since it contains only 20 proteins. If the distributions of TB and TV are more biased towards high degree proteins than the distribution for V, then we hypothesize that viral and bacterial pathogens have evolved to interact with hub proteins in the human PPI network.
The degree of a protein captures only its local connectivity. Centrality captures both global and local features of a protein's importance in a network. In this paper, we use the notion of a protein's betweenness centrality [110]. A protein with high betweenness centrality is characteristic of a bottleneck in an interaction network (i.e., there are many paths that pass through this protein) [34].
We define the betweenness centrality bc(v) of a protein v as the fraction of shortest paths in G between all protein pairs (u,w) that pass through the protein v. Given u, v, w ∈ V, let σuw denote the number of shortest paths between proteins u and w. There may be multiple equally long paths between u and w that are shorter than any other path between u and w. Let σuw(v) denote the number of these that pass through v. Then the betweenness centrality of v is
In our analysis, we divide bc(v) by the number of pairs of nodes in G, yielding a quantity between 0 and 1. We use the algorithm devised by Brandes [111] to compute the betweenness centrality of all nodes in G. This algorithm runs in time proportional to the product of the number of nodes in G and the number of edges in G. As with the degree analysis, we plot distributions of the betweenness centrality for V, TB, TV, and T(2)V. If the distributions for TB, TV, and T(2)V are biased toward higher values of centrality than the distribution for V, we hypothesize that pathogens have evolved to interact with bottlenecks in the human PPI network.
Let L be the ranked list of the proteins in V, where we rank the proteins either by degree or by betweenness centrality. Given L and a predefined set S of proteins of interest (e.g., those interacting with HIV), we use GSEA to determine whether the proteins contained in S are randomly distributed throughout L or concentrated at the top. In the ranked list L, let li be the value (of degree or centrality) at index i; 1 ≤ i ≤ |L|. We abuse notation and say that an index i is an element of S if the protein whose rank is i belongs to S. First, we compute m = Σi∈Lli, the sum of all the values in L. Next, for each index i in L, we compute two values:
Thus, Phit(S, i) measures the weighted fraction of proteins with index at most i that are in S and Pmiss(S, i) measures the fraction of proteins with index at most i that are not in S. We handle multiple ranks with identical values by computing these two values only at the largest rank for each unique value in L. Finally, we define the enrichment score as the largest positive value of Phit(S, i) - Pmiss(S, i), i.e.,
A large positive value of es(S, L) indicates that the proteins in S have high degree or high betweenness centrality. Note that our modification of the original definition of the enrichment score [35] ensures that if S mainly contains proteins with low degree or betweenness centrality, then the score will be close to 0, since Phit(S, i) − Pmiss(S, i) will be negative for most indices. We record the rank i that yields es(S, L); the column titled “#proteins contributing” in Table S1 of the supplementary data displays these numbers. To compute p-values for an observed enrichment scores, we generate a null distribution of scores by repeatedly selecting |S| random nodes in L and computing the score for each random subset of nodes. We repeat this process 1,000,000 times and estimate the p-value for s as the fraction of random sets whose score is at least as large as s. We obtain our results by testing each of 57 sets: TB, TV, T(2)V, and the sets VP corresponding to each of the 54 pathogen groups.
We isolate functionally coherent subsets of human proteins among the sets TB, TV, T(2)B, T(2)V, and the sets VP corresponding to each of the 54 pathogen groups using a test for functional enrichment. Given the hierarchical structure of the Gene Ontology (GO) [26], we account for dependencies between annotations by using the method proposed by Grossman et al. [112]. Let S be a set of proteins of interest (e.g., the set of proteins interacting with HIV). We aim to compute GO functions that annotate a surprisingly large number of proteins in S. To this end, for each function f in GO, we count sf, the number of proteins in S annotated with f and spa(f), the number of proteins in S annotated by at least one parent of f. We also compute vf and vpa(f), the number of proteins in V annotated by f and by at least one parent of f, respectively. With these four counts in hand, we use the hypergeometric distribution to compute the probability pf(S, V) of drawing sf or more proteins from a set of vf marked proteins when we select spa(f) proteins at random from a universe of vpa(f) proteins:
We account for multiple hypothesis testing using the method of Benjamini and Hochberg [113]. We consider only functions enriched with a p-value of at most 0.05. Note that different enriched functions may annotate identical sets of human proteins. In each such case, we group the functions and associate the most enriched function (and its p-value) with the group. To report enrichment ranks, we sort the groups in increasing order of p-value. Although not discussed in this paper we repeat this analysis using T (rather than V) as the universe of proteins. With T as the universe, we expect to find functions that distinguish between the pathogens. The results with T as universe are available on our supplementary website.
We compute enriched functions in each of the 54 sets of human proteins interacting with each pathogen group. We construct a binary matrix whose rows are enriched functions and whose columns are pathogen groups. An entry is one in this matrix if and only if the function is enriched with a p-value of at most 0.05 in the pathogen. In this binary matrix, we define a bicluster to be a subset R of rows and a subset C of columns such that each row-column pair in R × C contains a one. We also require a bicluster to be closed, i.e., each row not in R (respectively, column not in C) contains a zero in at least one column in C (respectively, row in R). We use the Bimax algorithm to compute all closed biclusters in this binary matrix [114].
Table 4 contains a list of all the proteins discussed in this paper and their corresponding UniProt ids and descriptions. |
10.1371/journal.pcbi.1000768 | A New Approach for Determining Phase Response Curves Reveals that Purkinje Cells Can Act as Perfect Integrators | Cerebellar Purkinje cells display complex intrinsic dynamics. They fire spontaneously, exhibit bistability, and via mutual network interactions are involved in the generation of high frequency oscillations and travelling waves of activity. To probe the dynamical properties of Purkinje cells we measured their phase response curves (PRCs). PRCs quantify the change in spike phase caused by a stimulus as a function of its temporal position within the interspike interval, and are widely used to predict neuronal responses to more complex stimulus patterns. Significant variability in the interspike interval during spontaneous firing can lead to PRCs with a low signal-to-noise ratio, requiring averaging over thousands of trials. We show using electrophysiological experiments and simulations that the PRC calculated in the traditional way by sampling the interspike interval with brief current pulses is biased. We introduce a corrected approach for calculating PRCs which eliminates this bias. Using our new approach, we show that Purkinje cell PRCs change qualitatively depending on the firing frequency of the cell. At high firing rates, Purkinje cells exhibit single-peaked, or monophasic PRCs. Surprisingly, at low firing rates, Purkinje cell PRCs are largely independent of phase, resembling PRCs of ideal non-leaky integrate-and-fire neurons. These results indicate that Purkinje cells can act as perfect integrators at low firing rates, and that the integration mode of Purkinje cells depends on their firing rate.
| By observing how brief current pulses injected at different times between spikes change the phase of spiking of a neuron (and thus obtaining the so-called phase response curve), it should be possible to predict a full spike train in response to more complex stimulation patterns. When we applied this traditional protocol to obtain phase response curves in cerebellar Purkinje cells in the presence of noise, we observed a triangular region devoid of data points near the end of the spiking cycle. This “Bermuda Triangle” revealed a flaw in the classical method for constructing phase response curves. We developed a new approach to eliminate this flaw and used it to construct phase response curves of Purkinje cells over a range of spiking rates. Surprisingly, at low firing rates, phase changes were independent of the phase of the injected current pulses, implying that the Purkinje cell is a perfect integrator under these conditions. This mechanism has not yet been described in other cell types and may be crucial for the information processing capabilities of these neurons.
| Cerebellar Purkinje cells exhibit a wide range of dynamical phenomena. They are intrinsic neural oscillators, firing spontaneously and highly rhythmically in the absence of synaptic input, at a rate of 10–180 Hz [1]–[5]. They also exhibit intrinsic bistability [2], [3], which influences their responses to sensory stimulation [3]. In addition, interactions between spontaneously firing Purkinje cells can result in waves of activity travelling down the cerebellar folia [4], or in high frequency oscillations [6], which may contribute to the generation of precise temporal patterns in the cerebellar network [7]. Hence, the firing of Purkinje cells is highly time- and state-dependent, and thus they represent excellent targets for dynamical systems analysis.
The phase response curve (PRC; [8]–[12]) is a powerful tool to study neuronal dynamics at the cellular level. The PRC describes the effect of a brief perturbation on the firing phase of a neuron, and can be used to predict the response of a neuron to more complex stimulation patterns [8]–[12]. The shape of the PRC is linked to the type of neuronal excitability [13], [14], to oscillatory stability [15] and to network synchronization properties [16]–[19]. Studying Purkinje cell PRCs is therefore an essential step to explore their dynamic repertoire, probe their biophysical mechanisms, and to construct models of Purkinje cells to determine their role in information processing at the network level.
PRCs can be obtained by directly perturbing the membrane potential by short (infinitesimal) square current pulses [8]–[12] or synaptic conductance pulses [12], [19]–[22], and via indirect methods [23]–[25]. Using the direct method, infinitesimal PRCs are obtained by repeatedly injecting brief current pulses while neurons are firing action potentials (APs). Phase and phase perturbation are measured by using the AP immediately preceding the current pulse as a reference, and we refer to these PRCs as “traditional” PRCs throughout this paper. We show using electrophysiological experiments and in simulations that the interspike interval variability present in Purkinje cells introduces a systematic bias in this traditional calculation of the PRC. The bias results from loss of causality caused by the jitter of the APs surrounding the current pulse, and gives rise to an empty triangular region in the PRC plot, which we call the “Bermuda Triangle”. We introduce a method for calculating the PRC which corrects for this bias by using all spikes in the spike train as a reference, one at a time. We refer to PRCs obtained by this method as “corrected” PRCs. Note that in our study both “traditional” and “corrected” PRCs are calculated using the same experimental data: perturbation of the firing of Purkinje cells with brief square current pulses. Using the corrected method we show that the shape of the Purkinje cell PRC changes fundamentally depending on the firing rate of the neuron.
Somatic whole-cell patch-clamp recordings were made in current-clamp mode from spontaneously firing Purkinje cells in mouse cerebellar slices. To construct PRCs, a single brief depolarizing current pulse (amplitude, 0.05 nA; duration, 0.5 ms) was injected after a 100–150 ms baseline period (see Fig. 1A). Repeating this protocol many times should result in a homogenous sampling of phase space in spontaneously spiking neurons. The resulting change in interspike interval (ISI) relative to the mean ISI corresponds to the PRC value denoted by . Plotting, for each trial, as a function of the phase, , at which the pulse arrived shows the overall ISI shortening corresponding to a positive PRC (Fig. 1B, neuron firing at 180 Hz; see Materials and Methods). Three observations can be made. First, at late phases there is a triangular region entirely void of data points (outlined in green) which we call the “Bermuda Triangle”. This causes a negative bias of the running average at late phases (Fig. 1B, dashed red line). Second, the intrinsic variability in the ISI [1] of spontaneously firing Purkinje cells acts as a source of noise, giving rise to data points with . However, removing all points beyond 1 does not eliminate the negative bias (Fig. 1B, solid red line). Finally, many trials (typically more than 5000) were required to calculate the Purkinje cell PRC reliably, while PRCs in other cell types are normally obtained from 100–200 trials [8], [15]. The ISI variability in Purkinje cells [1] results in PRCs with low signal-to-noise ratio, increasing the bias at late phases and leading to a miscalculation of the PRC when this traditional method is used. Thus, a robust and unbiased method for the calculation of Purkinje cell PRCs in the presence of noise is required.
To better understand the negative deflection of the PRC at late phases, a control PRC (cPRC; see Materials and Methods) was calculated from the unperturbed voltage traces prior to the current pulse. The cPRC should be zero throughout all phases. However, the negative bias of the PRC at the late phases persisted in the cPRC (Fig. 1C). We conclude that it is not the result of the brief current pulse injection. Rather, it results from the inhomogeneous sampling of the phase in the presence of noise. Indeed, the phase histogram (Fig. 1C, lower panel) indicates that late phases are sampled less frequently.
To reproduce the effect of noise, PRCs were obtained from a Purkinje cell model [26] in which Gaussian current noise was added to reproduce the irregularity of real Purkinje cell spiking (example model neuron firing at its spontaneous firing rate of 27 Hz; see Materials and Methods). The model PRC exhibited the same negative deflection at late phases as observed in the experimental PRC (Fig. 1D, dashed red line). As before, removing all points for which the phase exceeds 1 did not eliminate the negative deflection (Fig. 1D, solid red line). Similarly, the cPRC in the model exhibited the same negative bias and the same inhomogeneous phase distribution (Fig. 1E) as the experimental cPRC. Therefore, the negative bias at late phases is a general feature of the traditional method for calculating PRCs, and must be due to the intrinsic ISI variability.
In order to explain how the ISI variability might affect the PRC calculation, we sketch twelve representative scenarios in which spike jitter due to noise causes misclassification of the phase and/or the PRC value. In these scenarios (shown in Fig. 2A–L), we jittered either the first or the second AP (Fig. 2A–L, black lines) with respect to a perfectly periodic cycle of firing (Fig. 2A–L, grey lines). We divided the sketches into three blocks depending on the phase of the current pulse within the cycle (Fig. 2 A, D, G, J: early phase; B, E, H, K: middle phase; C, F, I, L: late phase). The misclassification of phase and/or PRC value (arrows) becomes clear when comparing them against their deterministic counterparts. The jitter of the spike preceding or following the brief current pulse can lead to a loss of causality and hence to a drastic miscalculation of the PRC. The most serious consequences of the ISI variability due to noise occur in the scenarios illustrated in Fig. 2C and 2J, where the jitter causes the current pulse to fall into a different cycle of firing, resulting in a significant bias at the late and early phases of the PRC, respectively. Specifically, the “Bermuda Triangle” effect present in both model and experiment can be explained by means of the sketch in Fig. 2C: when the pulse arrives at late phases, and the AP jitter results in the pulse falling into the subsequent ISI as compared to the deterministic case, the resulting phase is small according to the new ISI boundaries. Due to causality, it is impossible for a PRC point to fall into the green “Bermuda Triangle” in Fig. 1, since for all points in the triangle the shortening of the ISI would be larger than the actual phase difference of the pulse to the end of the ISI. This explains the observation that phases are sampled less frequently in the late part of the ISI, and thus the PRC values are underestimated and the effect of ISI noise is not averaged out.
To visualize the resulting phase and PRC misclassification, we translated each of these twelve sketches onto a corresponding phase plot (Fig. 2M,N). This allows the resulting phase and PRC values of each of the twelve cases to be compared against their deterministic counterparts. More specifically, regularly spaced spike times were defined and jittered independently by noise taken from a Gaussian distribution. The known actual phase without noise was plotted against the sampled phase. The assumption that the process underlying spiking is perfectly periodic and that the presence of a spike does not reset this underlying process is made only for generating the data in Fig. 2M,N (and subsequently Fig. 3B–E), and only for purposes of illustration. In a purely deterministic scenario, the sampled phase is linearly dependent on the actual phase (Fig. 2M, points on the diagonal). This is also the case for occurrences in which the noise has no effect on the phase (e.g. the scenarios in Fig. 2A or 2B; yellow points in Fig. 2M). For any deviations of the sampled phase from the actual phase due to noise, the points are scattered across the plot (Fig. 2M, color coding as in A–L). Based on the same principles, the effect of noise in each of the twelve scenarios on the PRC plot is shown in Fig. 2N (color coding as in A–L).
To summarize, the bias at late phases of the PRC calculated using the traditional method is due to erroneous phase sampling, which results from the substantial ISI variability present in spontaneously firing Purkinje cells, and the loss of causality between the current pulse and the jitter in the times of either of its two surrounding APs.
Our new method to correct for the bias in the traditional PRC and obtain a homogeneous phase histogram is illustrated in Fig. 3A. The red spike immediately preceding the pulse is the one used as a reference () in the traditional method. In our new method, instead of using just the spike immediately preceding the current injection, each spike in the spike train is taken as a reference one at a time and the corresponding phase values (indicated under the arrows in Fig. 3A) are all taken into account (see also Materials and Methods). In this case, the two spikes prior to the stimulation pulse (red and black in Fig. 3A) predominantly contribute to the phase interval [0,1] of the PRC (Fig. 3B, red and black points). The impact of the pulse on the subsequent ISI, the PRC2, is then determined by the two spikes following the current pulse (Fig. 3A blue and cyan spikes; also compare [24]) and so on. It is worth emphasizing that even though more than one spike is included in the PRC calculation, the presence of each reference spike resets the phase to zero (). Our method restores periodicity in the spiking jitter as can be seen in Fig. 3B (all points, in analogy to Fig. 2M). By taking only the points according to the traditional calculation of the PRC, a sharp boundary is drawn (Fig. 3B, red) resulting in an inhomogeneous distribution of sampled phases (Fig. 3C, red). In contrast, by including the second spike prior to the pulse, spike jitter effects are averaged out (Fig. 3D). The bias at the late phases of the PRC plot observed when taking points according to the traditional calculation of the PRC (Fig. 3E, red) is thereby eliminated (Fig. 3E, all points), as is the bias in the cPRC (not shown).
In order to validate our new method, we applied it to neuronal models for which the PRC can be calculated analytically (from the adjoint [27]). PRCs of the Morris-Lecar model (parameters from [28]), in the presence and absence of noise, were compared with the analytically derived PRC (Fig. S1A). The PRCs calculated using both the traditional and our corrected method overlap perfectly (except near and , due to the finite time step and finite amplitude of the current pulse in the simulations), and match the analytically derived PRC. In the presence of noise, the PRC calculated by the traditional method is biased at late phases, as described above. Our new method eliminates most of this bias.
However, it has been shown that noise can directly affect the dynamics of neurons underlying the PRC, leading to changes in the PRC which are not due to measurement errors (e.g. in the Morris-Lecar model [29]). We therefore used an additional model, the non-leaky integrate-and-fire model, in which noise-dependent changes of dynamics can be excluded. When noise was introduced in this model, the traditional method resulted in a biased PRC, as compared to the analytically derived PRC and the PRC in the absence of noise. Again, our corrected method removed most of this bias (Fig. S1B). The same analysis was repeated in a leaky integrate-and-fire model. This shows that the “Bermuda Triangle” and its consequences on the PRC are the result of the traditional calculation of PRCs, separate from the effect of noise on the dynamics of the system (Fig. S1C).
Next, using the Purkinje cell model [26], we compare the result of our method (Fig. 4A, black) to the PRC obtained with the traditional method (Fig. 4A, red) and the deterministic PRC without noise (Fig. 4A, green). When the noise is increased, reflected by an increased coefficient of variation (CV) of ISIs, the traditional PRC deviates from the deterministic one and the bias becomes more pronounced (Fig. 4A, dashed red line). In contrast, our corrected method performs as well as with low CV (Fig. 4A, dashed black line). The strong bias at late phases is eliminated. In order to evaluate the performance of our method in comparison to the traditional method, we calculated the integral of the differences between PRCs and their deterministic counterparts (PRC error; Fig. 4B). As the CV increases, the PRC error shows larger increases using the traditional (Fig. 4B, red line) compared to our corrected method (Fig. 4B, black line).
In conclusion, the “Bermuda Triangle“ present in PRCs is due to shortcomings of the traditional method for calculating PRCs. The bias can for the most part be compensated for by taking the two spikes preceding the pulse as a reference, one at a time, instead of just the spike immediately preceding the pulse as in the traditional method.
Spontaneous firing frequencies of Purkinje cells range from 10–180 Hz both in vitro [1], [2], [4], [5] and in vivo [3], [30]. To test how the dynamics of Purkinje cells change according to the firing frequency, we recorded from cells firing spontaneously at low (15–40 Hz, n = 10) and high (55–180 Hz, n = 6) rates and calculated their PRCs using our corrected method.
A representative corrected PRC is shown in Fig. 5A (the same example of a rapidly firing (180 Hz) Purkinje neuron as in Fig. 1B). The PRC is positive, indicating that the brief current pulse causes an advance of the following spike (shortening of the ISI relative to the mean) with maximum displacement when the pulse arrives near the middle of the ISI. It is worth noting that the phase histogram is homogeneous (Fig. 5A, lower panel), suggesting that, with the corrected method, the ISI is equally sampled throughout. In order to study the effects of the brief pulse on the subsequent intervals we plotted the PRC2–5 (Fig. 5B; see Materials and Methods). PRC2 is negative, suggesting that the subsequent ISI is lengthened relative to the mean. A PRC2 with opposite sign to the PRC has been previously reported [24] and it is believed to be due to a compensatory effect on the current ISI length. Indeed, as seen from PRC2–5, these curves are negative and the effect dies out after about 4 ISIs.
In comparison, an example of a PRC of a slowly firing (30 Hz) Purkinje neuron is shown in Fig. 5C. The brief current pulse causes the same positive displacement of the following spike independently of its position within the ISI, resulting in a square PRC. The phase histogram is homogeneous, indicating that there is an equal probability for the pulse to arrive at each phase within the ISI (Fig. 5C, lower panel). In order to study the effects of the brief pulse on the subsequent intervals we calculated the PRC2–5 (Fig. 5D; see Materials and Methods). They were negative, similar to those of cells firing at a high rate, but exhibited larger fluctuations. It is interesting to note that the PRC phase advances occur at a different scale in the slowly and rapidly firing Purkinje cells. However, when converted back into time units, the PRC values are of the same order of magnitude in both cases (see below).
The PRCs of Purkinje cells exhibiting slow (15–40 Hz; n = 10) and rapid (55–180 Hz; n = 6) spontaneous firing were calculated using our corrected method. The PRCs switched from square (phase-independent) for lower frequencies (Fig. 6A) to phase-dependent for higher frequencies (Fig. 6B). The switch occurred at a frequency of approximately 50 Hz. The average PRC of all neurons firing at low rates (Fig. 6A, thick line) is phase-independent. To our knowledge, such a square PRC has not been previously reported. A square PRC can only be obtained if the cells act as perfect non-leaky integrators. In contrast, the average PRC of all Purkinje cells firing at high rates (Fig. 6B, thick line) exhibited a sharp peak. It is useful to compare these average PRCs (Fig. S2, thick black and red lines) with the biased ones obtained with the traditional method (Fig. S2, thick green lines). To quantitatively assess the switch in dynamics we plotted the peak-to-baseline ratio of the PRCs in relation to the firing rate (Fig. 6C; see Materials and Methods). This quantity essentially compares the extreme value in the first half of the PRC with the extreme value in the second half. The switch at a firing rate of approximately 50 Hz can be seen clearly in this representation. The switch becomes particularly apparent when both the phase and the phase shift of the PRC are plotted in units of time, and phases are aligned with respect to the second AP in the ISI (Fig. 6D). Then, the peaks of the PRCs measured at high firing rates coincide (red), indicating that an input signal causes an effect only in a 3 ms window prior to the output spike irrespectively of the precise firing rate of the cells in that group. This peak in the PRC is shown to give way to a larger phase-independent plateau (black) at low firing rates, in which incoming signals will affect the spiking of the cell regardless of the time at which they arrive. A transitory PRC (thin solid red lines in Fig. 6B and Fig. 6D, indicated by arrows) showing both a plateau at early phases and a peak at late phases was observed in a cell with intermediate firing frequency (55 Hz). To summarize, the PRCs of Purkinje cells largely depend on the intrinsic firing frequency of the cells: they are phase-independent at low firing rates (15–40 Hz) and phase-dependent at high frequencies (55–180 Hz).
The firing rate of a Purkinje cell changes depending on modulation of its inputs [31]–[35]. For example, during locomotion in cats the firing frequencies of Purkinje cells can increase from an average of about 40 Hz [34] to more than 100 Hz [35]. To test whether the switch in Purkinje cell dynamics can occur in the same cell, we recorded Purkinje cell PRCs while modulating their firing frequencies using injected current (n = 3; Fig. 6C, points labeled with two colors). We first recorded at the spontaneous firing frequency, and if the spontaneous frequency was low, we next increased the firing rate by injecting a positive constant current. Alternatively a negative constant current was injected if the spontaneous frequency was high. The PRCs for both fast and slow states were calculated (Fig. 7, color coding as in Fig. 6C). When Purkinje cell spiking was changed from slow (33 Hz) to fast (104 Hz), the originally square PRC (Fig. 7A), exhibited a sharp peak (Fig. 7B). This change in the PRC was reversible, as when the neuron was allowed to relax back to its intrinsic firing rate (40 Hz) the PRC returned to a square shape (Fig. 7C). Conversely, another neuron initially firing at a high rate (71 Hz) exhibited a peaked PRC (Fig. 7D), which was switched to a square shape by reducing its firing rate to 26 Hz via injection of hyperpolarizing current (Fig. 7E). When the neuron was then allowed to fire at its intrinsic firing rate (84 Hz) the sharp peak in the PRC reappeared (Fig. 7F). Therefore, the switch in Purkinje cell dynamics reflected in the switch of the PRC can also occur in the same cell.
We have shown that the traditional method for calculating PRCs results in a bias, particularly in neurons exhibiting high ISI variability. We developed a corrected method for calculating PRCs which removes most of this bias. Our method can be directly applied to noisy experimental data. We used this corrected approach to measure for the first time the PRCs of Purkinje cells at various firing rates. At high firing rates, Purkinje cell PRCs were phase-dependent; however, a phase-independent PRC was observed at lower firing rates. This suggests that Purkinje cells can behave as perfect integrators at low firing rates, which has important consequences for our view of the integrative properties of these neurons.
We have determined Purkinje cell PRCs by injecting brief current pulses and measuring the phase change in the subsequent neuronal spiking. Since at the typical spontaneous firing rates of Purkinje cells these phase changes were small compared to the spike jitter during spontaneous spiking [1], many trials were required. This revealed a general bias of the traditional method at late phases of the PRC in the presence of noise (Fig. 1). We characterized the effect in a model with and without noise, and showed that the bias is related to inhomogeneous phase histograms caused by interspike interval jitter (Fig. 1 and Fig. 3). To correct for this, we developed a new method, which recovers periodicity in the spike jitter due to noise (Fig. 3). We showed that this method homogenizes the phase sampling in the experimental data and removes most of the bias observed in the PRCs calculated using the traditional method (Fig. 4A). Our corrected approach can be directly applied to existing experimental data in order to measure PRCs under low signal-to-noise conditions. It should be applicable to a wide range of cell types, as neuronal noise and the resulting ISI variability are not restricted to Purkinje cells [36].
The use of indirect methods to obtain PRCs, for example from the spike triggered average [23] or the poststimulus time histogram (PSTH) [24] are possible alternatives to the traditional method. Here we have applied a correction to the traditional method, which resulted in reliable PRC measurements in Purkinje cells. Further alternative methods for calculating PRCs exist. For example, dynamic clamp was previously used to study hippocampal spike-timing-dependent plasticity in relation to PRCs [37]. In this special case, underlying subthreshold oscillations provide phase locking. Such a method is only applicable if phase information is accessible to the experimenter, independent of spiking. PRCs can also be calculated using Bayesian statistics [25], or by injecting trains of rectangular current pulses [38] and noisy inputs [11]. These methods result in periodic PRCs, but only because periodicity is imposed as part of the optimization (fitting) techniques employed. In conclusion, our method can be applied to noisy experimental data to calculate PRCs while avoiding possible bias or overfitting problems present in some of the currently available methods. A wide, comparative study will be required in the future to find out which methods for calculating the PRC yield the best results under different conditions.
Purkinje cells fire spontaneously and modulate their firing in response to synaptic input. The spontaneous firing rate of Purkinje cells varies from 10 to 180 Hz, but firing frequency can also be increased by the ∼150,000 parallel fiber synaptic inputs [39] or decreased by molecular layer interneurons during the execution of motor tasks such as smooth-pursuit eye movements [40], maintenance of posture [41] and locomotion [35], [42]. For example, the rate of Purkinje cell firing can exhibit a consistent temporal relationship with wrist movement [31] or be monotonically related to eye velocity during smooth-pursuit eye movements [40].
How is the integration of single inputs affected by the firing rate of the Purkinje cell? We have addressed this question by measuring the PRC at different firing rates. Using our new approach, we determine experimentally the PRCs of cerebellar Purkinje cells and show that their shape changes significantly depending on the firing rate (compare Fig. 5A and Fig. 5C). At high firing frequencies (>50 Hz) Purkinje cell PRCs are monophasic (Fig. 6B). However, at low firing rates (<50 Hz), Purkinje cell PRCs become phase-independent (Fig. 6A). To the best of our knowledge, this is the first study to report a phase-independent PRC in a mammalian neuron.
It was previously reported in a spike-frequency adaptation model of cortical neurons that an increase in firing frequency causes a shift of the PRC peak from rightward skew to the centre with a decrease in amplitude [24], implying that the integrative properties of this model neuron change depending on the firing rate. Specifically, it was suggested that the model cell acts like a coincidence detector at low firing rates and more of an integrator at higher firing rates [24]. Purkinje cells appear to show the opposite behaviour, acting as perfect integrators at low firing rates.
The shape of the PRC is thought to be linked to the type of excitability of the neuron. Neurons with type I excitability, whose f-I curves are continuous, are thought to display purely positive PRCs while neurons with type II excitability, characterized by a discontinuity in the f-I curve at the onset of firing, exhibit biphasic PRCs [11], [13], [14]. While biphasic PRCs intuitively result in resonator behavior, neurons with purely positive PRCs act as integrators of synaptic input [11], [13], [14], [43]. Although Purkinje cells exhibit type II excitability [2], [44], [45], their PRCs are positive at all firing rates, implying that they are integrators rather than resonators. These findings suggest that the type of excitability of a neuron is not strictly correlated with the PRC shape. Similarly, Tateno and Robinson [15] showed that low-threshold spiking, fast spiking and non-pyramidal regular spiking interneurons can exhibit both purely positive and biphasic PRCs which do not always strictly correspond to the type of excitability of the neuron.
The shape of the PRC has functional implications for the integration of synaptic inputs. At high firing rates, Purkinje cells are most sensitive to inputs during the last 3 ms of their firing cycle (Fig. 6D), imposing a strict relationship between the timing of the input and the timing of spike output, with direct consequences for network dynamics. It has been shown theoretically that oscillators which are described by type I PRCs and are coupled by excitatory synapses tend not to synchronize [16]. However, the opposite is true for inhibitory coupling between oscillators [16], [46], such as coupled Purkinje cells. Indeed, theoretical and experimental evidence indicates that Purkinje cells tend to synchronize via inhibitory inputs [4], [6], [7].
As the firing rate of Purkinje cells decreases, and the levels of synaptic and intrinsic conductances and currents are modified, the PRC switches from monophasic to phase-independent (Fig. 6C). The phase-independent PRCs at low firing rates suggest that Purkinje cells integrate their synaptic inputs independently of their timing within the interspike interval (Fig. 6A). Our results therefore support the idea that at low firing rates, Purkinje cells cannot read out the timing of their inputs, which would exclude the use of a temporal code. Instead, in this regime they are well suited for rate coding.
What are the biophysical mechanisms responsible for the switch in PRC behaviour at different firing rates? To generate an entirely flat PRC would require a neuron to effectively completely compensate for its leak conductance. This is illustrated by the example of the PRC of a simple leaky integrate-and-fire neuron in which the leak conductance was eliminated (Fig. S1B and C). However, this absence of leak is unlikely to occur in real Purkinje cells, and the biophysical implementation remains unknown. PRCs qualitatively similar to those observed in our experiments at high firing rates can be generated by the Purkinje cell model of Khaliq and colleagues [26] (Fig. S3A). However, when the firing rate is lowered in the model, no qualitative switch in the shape of the PRC can be observed. A hint to the mechanisms underlying the switch in the experiment is provided by using the model of Akemann and Knöpfel [47] (a further development of the Khaliq et al. model): at low firing rates a ‘shoulder’ appears in the early phases of the PRC (Fig. S3B). However, none of these models fully capture the experimentally determined switch in Purkinje cells, perhaps reflecting the fact that both of these models represent dissociated Purkinje cells. Thus, our experimental results could aid the refinement of existing models in order to capture the full dynamic behaviour of Purkinje cells.
In conclusion, our experimental findings indicate that Purkinje cells display different dynamic behavior depending on their firing rate. At high firing rates these neurons act as coincidence detectors of synaptic inputs, with maximal sensitivity at the late phases of the interspike interval. In contrast, at low firing rates Purkinje cells are not suited for precise coincidence detection, but instead appear to perfectly integrate their inputs independently of their position within the interspike interval. Thus, at high firing rates Purkinje cells can transmit information via a temporal code whereas at low firing rates they are well-suited for rate coding.
All procedures were approved by the U.K. Home Office.
Twelve- to fifteen-day-old L7-tau-gfp mice [48] were anaesthetised using isoflurane, decapitated and their brains were transferred to ice-cold low Ca2+ artificial cerebrospinal fluid (ACSF) containing (in mM): 125 NaCl, 26 NaHCO3, 25 glucose, 2.5 KCl, 26 NaH2PO4, 0.5 CaCl2 and 3 MgCl2, saturated with carbogen (95% oxygen and 5% carbon dioxide gas). 230-µm-thick sagittal brain slices from the cerebellar vermis were cut on a VT1200S microtome (Leica Microsystems) and were transferred to normal ACSF containing the following (in mM): 125 NaCl, 26 NaHCO3, 25 glucose, 2.5 KCl, 26 NaH2PO4, 2 CaCl2 and 1 MgCl2, again bubbled with carbogen. The slices were incubated for 30–40 minutes at 37°C and were then allowed to cool to room temperature. Thick-walled, filamented, borosilicate glass electrodes (Harvard Apparatus Ltd.) were pulled to a tip resistance of 4–5 MΩ (PC-10 microelectrode puller, Narishige). Cells were visually identified with the aid of an upright infrared differential interference contrast (IR-DIC) microscope (Axioskop, Carl Zeiss) and a video camera (C2400-07, Hamamatsu). Purkinje cell somatic whole-cell patch-clamp recordings were obtained using an internal solution containing the following (in mM): 130 methanesulfonic acid, 10 HEPES, 7 KCl, 0.05 EGTA, 2 Na2ATP, 2 MgATP, 0.5 Na2GTP and 0.4% biocytin, pH-adjusted to 7.3 with KOH. All recordings were performed at 34.5±1°C in the presence of carbogen-bubbled ACSF supplemented with GABAA receptor blocker SR95331 (10 µM). Recordings were made with an Axoclamp 2B amplifier (Axon Instruments) and were filtered at 3 kHz and sampled at 50 kHz using an ITC-18 DAC board (Instrutech) and Axograph 4.9 (Axon Instruments). Series resistance and pipette capacitance were carefully monitored and compensated throughout the experiment. Methanesulfonic acid was obtained from Fluka, and other chemicals from Sigma-Aldrich and BDH Chemicals.
Data were analyzed with MATLAB (The MathWorks). To determine how spike timing during spontaneous firing is shifted by a brief perturbation, we injected rectangular current pulses of 0.5 ms duration and 50 pA amplitude, after a baseline of 150 ms (50 ms) of spontaneous firing in subsequent trials of 350 ms (100 ms) for a slowly (rapidly) firing cell. A control PRC (cPRC) was calculated using the unperturbed part of the voltage traces and assuming a current pulse injection (0 pA amplitude) after 50 ms (25 ms) of spontaneous firing in subsequent trials of 350 ms (100 ms) for a slowly (rapidly) firing cell. The cPRC should be zero throughout all phases.
The dynamics of a neuronal oscillator can be reduced to a single variable: the phase . is calculated by dividing the time from the previous spike by the period of the oscillation; it increases linearly from 0 to 1 between two spikes. Depending on the phase of the stimulus, a change in phase, , of subsequent spiking will occur.
Traditional method: A brief current pulse is injected at a random time. The spikes before and after it are identified. is calculated by the difference between the unperturbed and the perturbed [8]–[12]. When the unperturbed is defined as the mean ISI (), a point on the PRC plot becomes:(1)where denotes the ISI which contains the brief current pulse and is the PRC point calculated in reference to the spike just prior to the stimulus. is the time between the pulse and the preceding spike . The resulting curve is a plot of against . The curve is positive (negative) when the injected current advances (delays) the next spike. In the experimental and model (with noise) PRCs, we refer to raw data as the estimated measurements (‘points’) on the PRC plot. A moving average was calculated with a Gaussian kernel over the raw data.
Corrected method: A major problem with the traditional method is the loss of periodicity of the sampling reference (Fig. 3B), which results in an inhomogeneous sampling of phases in the presence of spike jittering. In order to restore periodicity, points unaffected by the stimulation pulse can be added to the ensemble of PRC points, which allows the spiking jitter to average out properly. These points can be obtained from the same data by adding PRC values when the preceding ISI is taken into account:(2)When preceding and subsequent ISIs are taken into account as in:(3)and(4)periodicity in the spiking jitter is restored, phases are sampled homogeneously and the cPRC becomes flat. In the resulting plot, the phase interval ranges from to and the PRC component affecting directly the interval corresponds to all points in the phase interval [0,1], termed PRC1. Successive PRC2–5, correspond to phase intervals [−1,0], [−2,−1], [−3,−2] and [−4,−3], respectively and indicate how are affected by the pulse.
Peak-to-baseline ratio: In order to distinguish the phase-independent PRCs from the phase-dependent ones, PRCs were classified according to the peak- to-baseline ratio. Inspired by Tateno and Robinson [15], local extrema at the two halves of the PRC (i.e. for and ) were calculated and were denoted as early () and late () respectively. The peak-to-baseline ratio is then defined as:
Simulations were performed in NEURON [49] using a model of Purkinje cells consisting of a single compartment [26], [47]. The model includes seven voltage-gated conductances (a resurgent Na+ current, fast and slow K+ currents, P-type Ca2+ current, Ca2+-activated K+ current and the hyperpolarization-activated current Ih) and one voltage-independent conductance (Ileak), based on voltage clamp measurements from Purkinje cells [26]. The membrane surface area of the neuron was modified (×13) to reproduce input resistance values close to those observed in Purkinje cells (80 MΩ). In order to mimic the noise observed in Purkinje cells, noisy current input drawn from a normal distribution with (mean) and (standard deviation) was injected at each time step of the simulation (every 25 µs) into the soma. The noise injection resulted in a coefficient of variation of ISIs of 0.05, which is comparable to the values measured in real Purkinje cells in the experiments presented here (see also [1]). Current pulses of 0.5 ms duration and 250 pA amplitude were injected after 2500 ms, at a time at which spike jitter had randomized spiking phase. Data shown is taken from more than 15000 trials.
Additional neuron models were used in the supplementary parts of the manuscript. For Fig. S1, the Morris-Lecar model was directly implemented using parameters from [28]. The adjoint was calculated using XPPAUT [27] and the PRCs with noise were integrated in MATLAB (The MathWorks). The parameters for the leaky integrate-and-fire model were: a membrane time constant of , a reset potential of , a threshold potential of , a membrane resistance of , and a steady driving current of (to result in 50 Hz firing) and was simulated at time steps of . For the non-leaky integrate-and-fire model the time constant was set to infinity and , otherwise the same parameters were used. An alternative model for Purkinje cell firing was used for Fig. S3 which also includes a resurgent Na+-current and modified voltage-gated K+-conductances [47]. In this model, current pulses of 0.5 ms duration were injected at amplitudes of 10 pA in the low firing rate (33 Hz) case and 60 pA in the high firing rate (111 Hz) case. Simulation results were analysed in the same way as the experimental data.
|
10.1371/journal.pntd.0004594 | Genetic Susceptibility and Predictors of Paradoxical Reactions in Buruli Ulcer | Buruli ulcer (BU) is the third most frequent mycobacterial disease in immunocompetent persons after tuberculosis and leprosy. During the last decade, eight weeks of antimicrobial treatment has become the standard of care. This treatment may be accompanied by transient clinical deterioration, known as paradoxical reaction. We investigate the incidence and the risks factors associated with paradoxical reaction in BU.
The lesion size of participants was assessed by careful palpation and recorded by serial acetate sheet tracings. For every time point, surface area was compared with the previous assessment. All patients received antimicrobial treatment for 8 weeks. Serum concentration of 25-hydroxyvitamin D, the primary indicator of vitamin D status, was determined in duplex for blood samples at baseline by a radioimmunoassay. We genotyped four polymorphisms in the SLC11A1 gene, previously associated with susceptibility to BU. For testing the association of genetic variants with paradoxical responses, we used a binary logistic regression analysis with the occurrence of a paradoxical response as the dependent variable.
Paradoxical reaction occurred in 22% of the patients; the reaction was significantly associated with trunk localization (p = .039 by Χ2), larger lesions (p = .021 by Χ2) and genetic factors. The polymorphisms 3’UTR TGTG ins/ins (OR 7.19, p < .001) had a higher risk for developing paradoxical reaction compared to ins/del or del/del polymorphisms.
Paradoxical reactions are common in BU. They are associated with trunk localization, larger lesions and polymorphisms in the SLC11A1 gene.
| Buruli ulcer is an infectious disease of skin, subcutaneous fat and sometimes bone, mainly affecting children in West Africa. It is considered as one of the Neglected Tropical Diseases but the disease occurs also in moderate climates like South East Australia and Japan where it may also affect adults. Once a patient has started antibiotic treatment, lesions may increase in size even if the antimicrobial treatment is effective; this is highly confusing for doctors and patients as they may think that treatment actually fails. The cause of Buruli ulcer is Mycobacterium ulcerans, related to other mycobacteria that cause disease in man, like leprosy and tuberculosis. Using data from two different studies in West Africa, we show that these paradoxical reactions are associated with trunk localization and that they occur more often in larger lesions. The chance to develop these reactions appeared partly inherited: carrying the homozygous ins/ins genotype of 3’UTR TGTG 285 polymorphism in the SLC11A1 gene increased the risk of paradoxical reactions. Vitamin D is important for the immune defense against infections by mycobacteria. Vitamin D blood concentrations were not associated with paradoxical reactions; patients generally did well, and we did not need corticosteroid immune suppression to overcome these reactions.
| The neglected tropical disease Buruli ulcer (BU) is the third most frequent mycobacterial disease in immunocompetent persons after tuberculosis and leprosy [1–2]. It is caused by Mycobacterium ulcerans. Central to the pathogenesis is the immunosuppressant and necrosis inducing toxin mycolactone.
During the last decade, an antibiotic regimen of eight weeks of streptomycin and rifampicin was introduced [3,4]. Earlier studies reported the success of this antimicrobial treatment with or without surgery [5–7]. A clinical trial showed that antimicrobial treatment was highly effective in patients with small lesions (cross-sectional diameter < 10 cm), of which 96% healed without surgery [8].
However, during or after antibiotic treatment the BU lesions may worsen. This could be caused by treatment failure [9–11], but might also be due to an inflammatory response caused by treatment-induced recovery of the immune system, i.e. a paradoxical reaction. Paradoxical reactions have been described in tuberculosis and in leprosy [12,13]. Recent studies have recognized the existence of paradoxical reactions in BU [11,14]. In Australia, one in five BU patients appear to have a paradoxical reaction. Most cases occurred between three and ten weeks after the start of treatment [9]. In a trial in Ghana, most of the cases with a paradoxical reaction (>30%) were reported at week eight after the beginning of antimicrobial treatment [15]. The diagnosis of paradoxical response is difficult; no serological markers have been identified to differentiate paradoxical reactions from treatment failure [15]. Paradoxical reactions can be defined clinically by worsening of existing lesions, or the appearance of new lesions, and histologically by the appearance of intense inflammation in lesions [9]. Importantly, in most areas endemic for BU, histology is not available. In Africa, very few studies have addressed paradoxical reactions in BU [10,14] as well as its risk factors. In Australia, edematous lesions, use of amikacin and age above sixty years old were strongly associated with paradoxical reactions. In addition to sociodemographic and clinical features, we suggest genetic factors may influence the occurrence of paradoxical reactions as well. As paradoxical reactions are hypothesized to reflect an exaggerated immune response, genes involved in the immune response in infectious diseases might play a role. For BU, a polymorphism in the innate immune SLC11A1 gene (formerly known as NRAMP1) was previously found to be associated with increased susceptibility to BU [16]. Furthermore it has been shown that 1,25(OH)2D3 suppresses the Th1 response by down-regulating the production of pro-inflammatory cytokines [17–19]. So it is possible that polymorphisms in SLC11A1 gene as well as vitamin D are also related to paradoxical reactions.
In West Africa, most of the patients are below age 15 [20] and amikacin is not used to treat BU but very few patients receive antimicrobial treatment without streptomycin, the parent aminoglycoside drug. As the patient demographics and treatment regimen in West-Africa are widely different from that of Australia, it is important to look at the risk factors for developing paradoxical reactions in BU in this region. In Ghana, paradoxical reactions were described in patients with M. ulcerans infection with early lesions (duration < 6 months), limited to 10 cm cross-sectional diameter [14]; large lesions that are common in west Africa were not included in that study. Our study focuses on the risk factors associated with paradoxical reactions in patients with both small and large BU lesions, during and after antimicrobial treatment, and examines the influence of genetic factors as well.
In the present study, we included participants of two randomized clinical trials in Ghana and Benin. The BURULICO drug trial with patients enrolled between 2006–2008, was a randomized controlled trial for the treatment of early (duration less than 6 months), limited (cross-sectional diameter, 10 cm) M. ulcerans infection [clintrials NCT00321178]. In this trial, patients were randomized to receive either 8 weeks of streptomycin and rifampicin or 4 weeks of streptomycin and rifampicin followed by 4 weeks of clarithromycin and rifampicin. Participants in this study that had their BU lesions healed at time point 52 weeks after initiation of antimicrobial treatment were earlier studied for possible paradoxical reactions [14]. The second trial is a randomized trial on timing of the decision on surgical intervention for BU patients treated with rifampicin and streptomycin [clintrials NCT01432925]. All included patients (2011–2015) had confirmed M. ulcerans infection by direct microscopy following acid-fast staining or Polymerase Chain Reaction (PCR), and all received 8 weeks of antimicrobial therapy with rifampicin and streptomycin. For both trials, patients who were pregnant, children below five years old, patients not compliant with the antibiotic therapy, and patients with osteomyelitis, were excluded from the study. For the current study population, 150 of 241 participants of the BURULICO study, and 91 of the Burulitime study contributed (S1 Dataset).
For all patients, we recorded demographics and clinical data from the trial databases. In addition, we recorded the progression of the size of the lesion size by measurement at regular intervals. For both trials, measurements were available for the first 12 weeks at two-week intervals. In the BURULICO trial, lesions were measured at 14, 21, 27 weeks after start of treatment, and for the timing of surgical intervention trial, measurements were available at 16, 20, and 28 weeks after starting treatment. For analyses, the measurements at 14 and 16 weeks, at 21 and 20 weeks, and 27 and 28 weeks were considered to be equivalent time points.
We considered an increase in lesion area of more than 5% between two consecutive measurements as a clinically relevant change. We defined a paradoxical reaction as 2 consecutive increases in lesion size after 1 initial decrease. We additionally performed all analyses (post-hoc) using a less strict definition of two consecutive increases without an initial decrease.
For associations of clinical and patient characteristics with paradoxical responses, we used t-tests or Mann-Whitney U tests for accordingly and Χ2 tests for categorical variables. For testing the association of genetic mutations and variants with paradoxical responses, we used a binary logistic regression analysis with the occurrence of a paradoxical response as the dependent variable.
The protocol was approved by the Committee on Human Research, Publication, and Ethics of the Kwame Nkrumah University of Science and Technology and the Komfo Anokye Teaching Hospital, Kumasi (CHRPE/07/01/05), by the Ethical Review Committee of Ghana Health Services (GHS-ERC-01/01/06) and by the provisional national ethical review board of the Ministry of Health Benin, nr IRB00006860. Written and verbal informed consent or assent was obtained from all participants aged ≥12 years, and consent from parents, caretakers, or legal representatives of participants aged ≤18 years. All data were analyzed anonymously.
A total of 241 patients were included, 150 from Ghana, and 91 from Benin; 61% were female. The mean (SD) age was 16.2 (13.2) years. On presentation, 45% of patients had an ulcer, 23% had a plaque, and 13% had a nodule; 29% had a WHO category I lesion, 55% a category II lesion, and 16% a category III lesion. The median (IQR) surface area of the lesion on presentation was 20.6 (6.6; 43.5) cm2; 49% had a lesion on the lower limb, 43% on the upper limb, and 8% on the trunk.
Paradoxical reactions, as defined by an initial decrease of the lesion followed by two consecutive increases occurred in 22% of cases. Most paradoxical reactions occurred between weeks 8 and 12 (Fig 1). When using a definition that did only require two consecutive increases without an initial decrease, 26% of patients had a paradoxical response, and the frequency distribution of the initiation week of paradoxical reaction did not differ substantially. All cases that had a paradoxical reaction healed without additional treatment.
Paradoxical reactions were significantly related to the site of lesion (p = .039 by Χ2): 44% of patients with a lesion on the trunk had a paradoxical response, compared to 24% of patients with a lesion on the upper limb, and 17% with a lesion on the lower limb.
Paradoxical reactions were also significantly related to WHO category at presentation. Ten percent of patients with a category I lesion had a paradoxical response, compared to 27%, and 23% of patients with a category II and category III lesion, respectively (p = .021 by Χ2). Paradoxical reactions were not significantly related to patient age or gender. They were also not related to the type of lesion, duration of lesion before presentation, or white blood cell count at presentation (Table 1). For the participants in the BURULICO trial, paradoxical reactions were not related to treatment arm (8 week streptomycin vs 4 weeks streptomycin followed by 4 weeks clarithromycin). The pulse and temperature at the time of paradoxical response did not differ from the pulse at presentation by paired samples t-test, and did not differ from the average pulse and temperature of those not classified as having a paradoxical response at the respective week. Using the less strict definition, the same pattern of results emerged, where paradoxical reactions were significantly related to the site of the lesion (p = .024 by Χ2) and WHO category at presentation (p = .009 by Χ2), but to none of the other variables.
Vitamin D deficiency was found in 15% of participants. The mean (SD) vitamin D level was 66.5 (19.1) for the patients who had paradoxical reaction and 68.3 (17.1) for those who did not; 38% of patients with a vitamin D deficiency had a paradoxical reaction, compared to 23% of patients without a deficiency (p = .134 by Χ2). In the post-hoc analysis using the less strict definition of a paradoxical response, 33% of patients with a vitamin D deficiency had a paradoxical reaction, compared to 17% of patients without a deficiency (p = .082 by Χ2).
31% of patients with a 3’UTR TGTG ins/ins polymorphism had a paradoxical response, compared to 13% of patients with a ins/del or del/del polymorphism (OR 0.14, 95% CI: 0.05–0.44). 5’(CA)n microsatellite length, INT4 G/C polymorphism and D543N G/G polymorphism were not significantly related to paradoxical responses (Table 1). Using the less strict definition of a paradoxical response in a post-hoc analysis, a similar pattern of results emerged.
This is the first prospective study in West Africa addressing risk factors associated with paradoxical reaction in BU. In our sample, paradoxical reactions were common, and significantly associated with trunk localization, larger lesions and genetic factors. Currently, there is no standard definition of paradoxical reactions in BU. Histological aspects [9] suggested from Australia is not feasible in rural West Africa where most BU cases occur [2]. All patients included in this study healed without changes in therapy (no change in antibiotics, no corticosteroids). This strongly supports our suggested definition and suggests that cases in our study represent true paradoxical reaction and not progressive disease secondary to antibiotic failure.
We found a 22% incidence of paradoxical reactions (2 consecutive increases after 1 initial decrease and healing without surgery or a change in antimicrobial therapy), which is similar to a previous study from Australia [9].
In our study, most paradoxical reactions occurred between week 8 and 12—slightly later than the Australian study, where most paradoxical reactions occurred between week 3 and 10 [9,11]. In the case reports from Benin paradoxical reactions occurred between 12 and 409 days after completion of antibiotic treatment [10].
Mycolactone, the exotoxin produced and secreted by M. ulcerans, has been proposed as the major cause of immune suppression [28–32]. Perhaps, the period between week 8 and 12 in which most paradoxical reactions occurred coincides with the elimination of most M. ulcerans organisms, with an arrest in the production and subsequently, a strong decrease in tissue concentration of mycolactone. The increase of the lesion then reflects an inflammatory response against the microbes—or microbial antigens of dead bacilli—already present in tissue which initially failed to elicit a host immune response [25–27,30].
We found several risk factors associated with paradoxical reactions. The incidence appeared to increase in larger lesions. One explanation of this may be that smaller lesions heal before eight weeks when most of the paradoxical reaction occurs. Another possibility is that larger lesions have a higher bacterial load than small lesions. We showed that lesions localized on trunk were significantly associated with paradoxical reaction, even when controlling for the size of the lesion. More than 4 out 10 patients (44%) with lesion on the trunk had paradoxical reaction compared to 24% and 17% for the upper limb and lower limb respectively. The increased incidence of paradoxical reactions on the trunk might be due to a difference in local immune responses and body temperature.
Our study shows that paradoxical reactions were not significantly associated with patient age or type of lesion. This finding contrasts with Australian patients in whom associations between paradoxical reactions and age and edema were reported [9]. This might be due to differences in the study populations. In affluent countries like Australia, with a steeper population pyramid, BU mainly affects the elderly in Australia [31], while in West Africa, most patients are children [32].
Paradoxical reactions were not associated with the white blood cell count or patients’ vital parameters such as the temperature and the pulse rate. We argue that an increase of pulse, temperature or white blood cell count is indicative of an additional disease or super-infection, which should be further investigated. Whether paradoxical reactions were associated with aminoglycoside use, as has been shown for amikacin in Australia, could not be examined for streptomycin use because all study participants had been exposed to this drug, for 4 or 8 weeks. One might speculate that this effect seen in amikacin might in fact reflect a decrease in paradoxical reactions by using antimicrobial drugs like macrolides that have been associated with immuno-modulatory effects [33].
We also show for the first time that paradoxical reactions to M. ulcerans infection are associated with genetic factors. Carrying the homozygous ins/ins genotype of 3’UTR TGTG polymorphism in the SLC11A1 increases the risk of paradoxical reactions in BU. Earlier studies have shown that genetic factors can influence the innate immune response to mycobacterial antigens, such as infectious disease susceptibility genes, e.g., SLC11A1, HLA-DR, vitamin D3 receptor, and mannose binding protein [34,35]. In BU no associations were found with the 3’UTR TGTG ins/del polymorphism and developing BU [16]. However in tuberculosis, it was reported that participants who were heterozygous for two SLC11A1 polymorphisms (INT4 and 3’UTR) were at highest risk of tuberculosis [35]. A meta-analysis [35] has shown that the TGTG ins/ins 3’UTR genotype protected against tuberculosis, compared to the del/del genotype. We interpret our data such that the protective TGTG ins/ins 3’UTR genotype in the SLC11A1 gene may induce a stronger immune response during M. ulcerans infection. In turn, this stronger immune response might increase the risk of paradoxical reactions once BU develops. It has been reported that genetic variation in SLC11A1 affects susceptibility to others mycobacterial diseases such as leprosy and tuberculosis [35–37]. However, no study addressed the genetic risk factor for paradoxical reaction in tuberculosis or leprosy.
In this study, we report for the first time that paradoxical reactions are not associated with vitamin D level. Vitamin D deficiency has been found to be associated with susceptibility to tuberculosis [38]. Very few studies address vitamin D and paradoxical reactions in tuberculosis. Clearing of pathogens with anti-tuberculosis treatment and a delayed negative feedback on macrophage activation due to low 1,25(OH)2D production from vitamin D deficiency can lead to excessive granuloma formation and an exacerbated inflammatory response [39]. In our sample, the means of vitamin D level in patients with or without paradoxical reactions were similar.
All included patients in this study healed without any change in therapy. In earlier studies corticosteroids were used to treat paradoxical reactions [9,40,41]. We would indeed caution for use of corticosteroids West Africa, as other infections like tuberculosis and strongyloidiasis may worsen.
This study has some limitations. There are no standard definitions of paradoxical reactions in BU that we could use to validate our definition. Our definition is clinical and did not include histological aspects, which may lead to a lack of accuracy. However we believe that our cases accurately represent paradoxical reactions since all patients healed without any additional therapy. Secondly, we excluded co-infected patients with Buruli ulcer and HIV. This may have reduced the incidence and severity [42].
Paradoxical reactions are common in BU–and it is important that these should be differentiated from antimicrobial treatment failure. These paradoxical reactions are associated with trunk localization, larger lesions and certain polymorphisms in the SLC11A1 gene. There was no apparent need to change therapy or add steroids.
|
10.1371/journal.pntd.0002953 | A Randomized, Single-Ascending-Dose, Ivermectin-Controlled, Double-Blind Study of Moxidectin in Onchocerca volvulus Infection | Control of onchocerciasis as a public health problem in Africa relies on annual mass ivermectin distribution. New tools are needed to achieve elimination of infection. This study determined in a small number of Onchocerca volvulus infected individuals whether moxidectin, a veterinary anthelminthic, is safe enough to administer it in a future large study to further characterize moxidectin's safety and efficacy. Effects on the parasite were also assessed.
Men and women from a forest area in South-eastern Ghana without ivermectin mass distribution received a single oral dose of 2 mg (N = 44), 4 mg (N = 45) or 8 mg (N = 38) moxidectin or 150 µg/kg ivermectin (N = 45) with 18 months follow up. All ivermectin and 97%–100% of moxidectin treated participants had Mazzotti reactions. Statistically significantly higher percentages of participants treated with 8 mg moxidectin than participants treated with ivermectin experienced pruritus (87% vs. 56%), rash (63% vs. 42%), increased pulse rate (61% vs. 36%) and decreased mean arterial pressure upon 2 minutes standing still after ≥5 minutes supine relative to pre-treatment (61% vs. 27%). These reactions resolved without treatment. In the 8 mg moxidectin and ivermectin arms, the mean±SD number of microfilariae/mg skin were 22.9±21.1 and 21.2±16.4 pre-treatment and 0.0±0.0 and 1.1±4.2 at nadir reached 1 and 3 months after treatment, respectively. At 6 months, values were 0.0±0.0 and 1.6±4.5, at 12 months 0.4±0.9 and 3.4±4.4 and at 18 months 1.8±3.3 and 4.0±4.8, respectively, in the 8 mg moxidectin and ivermectin arm. The reduction from pre-treatment values was significantly higher after 8 mg moxidectin than after ivermectin treatment throughout follow up (p<0.01).
The 8 mg dose of moxidectin was safe enough to initiate the large study. Provided its results confirm those from this study, availability of moxidectin to control programmes could help them achieve onchocerciasis elimination objectives.
ClinicalTrails.gov NCT00300768
| Around 100 million Africans live in onchocerciasis endemic areas. Control of onchocerciasis as a public health problem and possibly even elimination of onchocerciasis infection relies on annual community-directed mass treatment with ivermectin. Given concerns about possible emergence of ivermectin resistance of the parasite Onchocerca volvulus and elimination of infection in areas where very high numbers of vectors can result in continued parasite transmission even when only few parasites are present in only a few people, research for drugs with higher effect on the parasite remains important. A series of non-clinical and clinical studies was planned to find out whether moxidectin, a veterinary anthelminthic, is sufficiently safe for mass treatment and has a better effect on the parasite than ivermectin. We report here results from the first study in infected people conducted to assess in small numbers of individuals the adverse reactions to the killing of parasites by moxidectin. A single dose of 8 mg moxidectin reduced skin parasite numbers better and for a longer time than ivermectin. The frequency and severity of adverse reactions was so low that a larger study to better characterize the adverse reactions to moxidectin and compare its efficacy with that of ivermectin was initiated.
| Onchocerciasis is caused by the filarial nematode Onchocerca volvulus and is transmitted among humans through the bites of blackfly vectors, in Africa mainly by Simulium damnosum s.l.. Around 99% of people at risk live in Sub-saharan Africa. The African Programme for Onchocerciasis Control (APOC) estimated 89 million Africans at risk and 37 million infected in 19 APOC countries [1] based on rapid epidemiological mapping [2].
Since its launch in 1995, APOC and the public health systems have established annual community-directed treatment with ivermectin (CDTI) to eliminate onchocerciasis as a public health problem in the 17 APOC countries with areas where onchocerciasis is meso- and/or hyperendemic [3]. At the time, it was considered impossible for CDTI to interrupt transmission across Africa [4]. Thus, the UNICEF/UNDP/World Bank/World Health Organization Special Programme for Research and Training in Tropical Diseases (WHO/TDR) continued research for drugs or drug combinations which could eliminate onchocerciasis infection (e.g. [5]–[7]). Today, prospects for elimination of infection with CDTI appear better [8]–[10]. Questions remain as to whether CDTI alone can eliminate onchocerciasis in highly endemic areas [11]–[13].
Moxidectin, a milbemycin macrocyclic lactone, is registered worldwide as an anthelmintic in cattle, sheep, swine, horses and dogs [14]. Initiation of the clinical development of moxidectin was based on (i) published data [15]–[18] and investigator reports to TDR from in vitro (O. volvulus, O. lienalis, O. gutturosa) and in vivo models (O. cervicalis in horses, O. lienalis in mice, O. ochengi in cattle, Brugia pahangi in dogs and jirds) of onchocerciasis and lymphatic filariasis and (ii) toxicology data from development as a veterinary drug [19]. The objective of the development for human use is to assess through a series of non-clinical and clinical studies whether moxidectin could be safe for mass treatment for onchocerciasis control with an efficacy which modelling studies suggest could result in permanent interruption of transmission of O. volvulus after substantially less rounds of mass-treatment than ivermectin. Data from moxidectin use for veterinary parasites [14], in vivo models of human filarial infections and on the effects of ivermectin, an avermectin macrocyclic lactone, on O. volvulus, suggest several possible effects of moxidectin on O. volvulus. These include a microfilaricidal effect combined with an embryostatic effect after a single dose and/or an adult worm (macrofilariae) sterilizing effect and/or a macrofilaricidal effect upon repeated exposure. Furthermore, repeated exposure to moxidectin might reduce the life time of the macrofilariae as discussed by Geary and Mackenzie [20] for the effect of long term treatment with ivermectin. Given a half life of 20–40 days in healthy volunteers [21]–[25] (around 20 days in the participants in this study, unpublished data, manuscript in preparation), a potential effect of moxidectin on the viability or development of transmitted L3 larvae could also be considered.
The data from two studies in healthy volunteers [21], [24] and all toxicology data available at the time resulted in the decision to initiate the first study in individuals infected with O. volvulus [19], [26]. It was not known whether the putative microfilaricidal activity of moxidectin would be associated with a combination of severe Mazzotti reactions (i.e. the complex, acute inflammatory response of the body to the effect of the drug on microfilariae), similar to those seen after diethylcarbamazine treatment which make diethylcarbamazine unsuitable for mass treatment [27]–[32]. Consequently, the study was designed to determine in a small number of participants whether moxidectin induces severe reactions at a frequency suggesting that development of moxidectin should be discontinued. If that was not the case, the study was designed to determine the moxidectin dose(s) with an adverse reaction profile suitable for further clinical testing. Further clinical testing in a large number of participants would allow to better define moxidectin's safety profile and to quantify the difference in the effect of moxidectin and ivermectin on skin microfilariae levels.
The participant-safety driven design resulted in a study duration of ≥1.5 years. Therefore, participant follow up was expanded beyond that required for assessment of Mazzotti reactions to obtain pharmacokinetic data as well as initial data on moxidectin's effect on the parasite relative to that of ivermectin.
This paper summarizes the safety data with focus on statistically significant differences to ivermectin, presents the effect on the skin microfilariae (mf) and reports the results of the histological examination of the macrofilariae from subcutaneous nodules excised 18 months after treatment.
This study was approved by the Ghana Food and Drugs Board, the Ghana Health Service Ethics Review Committee and the WHO Ethics Review Committee. Study conduct according to the principles laid down in the Declaration of Helsinki and in compliance with Good Clinical Practice and the protocol was monitored regularly.
Participants gave informed consent to study participation and testified to this by signature or thumbprint, as specified in the protocol approved by the Ethics Committees, in the presence of an independent literate witness in their villages before initiation of any study related procedures.
The severity of many Mazzotti reactions correlates with the skin mf density [31]. Therefore, each of three dose levels of moxidectin (2 mg, 4 mg, 8 mg, established on the basis of the pharmacokinetic data from healthy volunteers [21], [24], 34–136 µg/kg or 0.05–0.21 µmol/kg for 59 kg body weight) was evaluated sequentially in three cohorts of participants with different levels of skin mf density and ocular involvement pre-treatment. In the first cohort, participants had a skin mf density <10 mf/mg skin and no ocular involvement (subsequently referred to as ‘mildly infected’). In the second cohort, participants had a skin mf density of 10 mf/mg to 20 mf/mg skin and the sum of microfilariae in both anterior chambers of the eye had to be ≤10 (subsequently referred to as ‘moderately infected’). Participants of the third cohort had skin mf density >20 mf/mg skin without or with any level of ocular involvement (subsequently referred to as ‘severely infected’) (Figure 1).
In each mildly and moderately infected cohort, 16 participants were planned to be enrolled and randomized in a ratio of 3∶1 to moxidectin or 150 µg/kg ivermectin (as per ivermectin labelling for use in onchocerciasis). This provided 4 ivermectin treated participants as concurrent controls for the safety data in the planned 12 moxidectin treated participants in each cohort. To increase the probability of detection of adverse events in participants with high skin mf density (who are most likely to experience Mazzotti reactions [31]), 32 severely infected participants were planned to be enrolled for each moxidectin dose level and randomized 3∶1 to moxidectin∶ivermectin. This provided 8 ivermectin treated participants as concurrent controls for the safety data of 24 moxidectin treated participants. Across all 9 cohorts, the planned number of participants resulted in 48 participants in each treatment group for comparison of the safety data as well as the effects on the parasite. Figure 1 shows the number of participants actually treated in each cohort and provides the screen failure reasons which resulted in these numbers being lower than the planned numbers.
Mazzotti reactions usually subside within one or two weeks after treatment [30]. Since no data on Mazzotti reactions following moxidectin treatment were available, the decision to treat the next cohort within one moxidectin dose level was made based on the safety data obtained during the first month after treatment of the previous cohort. Progression to the next dose level was decided upon based on all data available to Month 1 follow up of the last cohort at the previous dose level (Figure 2). To further decrease potential risk to participants, all participants remained in the study center for 18 days after treatment. Subsequent follow up to 18 months was on an outpatient basis.
Participants were recruited from onchocerciasis endemic villages between 0°30′ and 0°45′E, 6°45′ and 7°0′N within the River Tordzi basin in the Volta Region of South-eastern Ghana. The vast majority (90%) of participants came from the villages Honuta-Gbogame, Kpedze-Anoe, Togorme, Aflakpe, Luvudo, Kpoeta-Ashanti and Hoe, the remainder from 11 other villages in the area (Figure 3). This area was not included in vector control activities under the Onchocerciasis Control Programme because at the time of the OCP it was forested. Simuliid species were Simulium yahense and Simulium squamosum [33]. At the time of this study, the area was not yet included in the ivermectin mass distribution programme of the National Onchocerciasis Control Programme because it is overall hypoendemic with small meso- or hyperendemic foci. The area is not endemic for lymphatic filariasis or loiasis.
A total of 172 of 196 planned individuals meeting the intensity of infection criteria described above but otherwise regarded as healthy based on physical examination, electrocardiography, medical and medication history, serum biochemistry, haematology and semiquantitative urinalysis participated in the study. Volunteers with a history of or current neurological or neuropsychiatric disease or epilepsy, orthostatic hypotension at screening, hyperreactive onchodermatitis and antifilarial therapy within the previous 5 years as well as pregnant and breastfeeding women were excluded. Women of child-bearing potential who wanted to participate had to agree to contraception (depo-medroxyprogesterone acetate or levonorgestrel implants) during the first 150 days after treatment. The pre-treatment evaluations included those detailed in the footnote to Table 1 and height measurement. Vital signs were obtained 12 times during the pre-treatment evaluations and the mean was used to assess changes post-treatment.
The 3 mg ivermectin tablets (purchased from Merck and Co. Inc), 2 mg moxidectin tablets developed for human use, as well as placebo were provided by Wyeth in identical looking capsules. Inactive ingredients of the moxidectin tablets were microcrystalline cellulose, anhydrous lactose, sodium croscarmellose, sodium lauryl sulfate, colloidal silicon dioxide, and magnesium stearate. Placebo capsules contained the inactive ingredients of the moxidectin tablets. Each participant received 4 capsules provided in an envelope labelled only with subject identifying information, resulting in participants and investigative team being blinded.
In each cohort, participants were stratified by sex and randomly allocated by a pharmacist in a ratio of 3∶1 to receive a single oral dose of moxidectin or ivermectin (150 µg/kg) based on computer-generated randomization schedules with a block size of 4 provided by the sponsors. The pharmacist prepared for each participant an envelope which contained 4 capsules, including 1, 2 or 4 capsules containing a 2 mg moxidectin tablet or 2, 3 or 4 capsules containing a 3 mg ivermectin tablet and the complementary number of placebo capsules. The pharmacist gave the sealed envelopes to the investigative team and was not otherwise involved in the study.
Treatment was administered on day 1 between 7:00 and around 7:40 under observation by members of the investigative team after an overnight fast and vital sign measurement.
Treatment effects were evaluated daily during the first 18 days and 1, 2, 3, 6, 12 and 18 months after treatment (Table 1).
Safety outcomes: Adverse events (AEs), including (i) clinically significant changes in laboratory values from pre-treatment, (ii) clinically significant adverse changes from pre-treatment in systemic or ocular symptoms detected through examinations, spontaneous reporting by participants or questioning of participants, and (iii) changes in vital signs, whether clinically significant or not. AEs were to be classified as serious (SAE) if they met the criteria in the SAE definition in the ‘Guideline for Good Clinical Practice’ of the ‘International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use’, i.e. if they resulted in death, persistent or significant disability or incapacity, a congenital anomaly or a birth defect or were life-threatening or required inpatient hospitalization or prolongation of an existing hospitalization [34] or if they were important medical events, including cancer, that could jeopardize the subject and require medical or surgical intervention to prevent one of the outcomes above. The severity of all AEs was graded (grade 1–4) according to the Onchocerciasis Chemotherapy Research Center (OCRC) Common Toxicity Criteria (OCRC CTC) version 2.0 (Supporting Table S1), an expansion of the criteria developed at OCRC for quantitation of Mazzotti reactions [35], [36]. Severity of AEs not included in the OCRC CTC was graded according to the National Cancer Institute Common Toxicity Criteria version 2.0 [37]. All AEs were categorized by the Principal Investigator (KA) as Mazzotti reactions, non-Mazzotti adverse drug reactions or drug-unrelated adverse events.
Efficacy outcomes: Skin snips were obtained with a 2 mm Holth or Walser corneoscleral punch. One snip each was taken from the right and left iliac crest as well as the right and left calf for a total of 4 snips/participant pre-treatment, 8 days and 1, 2, 3, 6, 12 and 18 months after treatment. Each skin snip was weighed on an electronic balance to an accuracy of 0.1 mg. Snips were incubated individually in a well of a 96-well plate with flat bottom for at least 8 hours in 0.9% normal saline at approximately 22°C. Microfilariae were counted under an inverted microscope at a magnification of 60× supplemented by 100× when necessary. The skin mf density of each participant at each follow up time point was calculated as the arithmetic mean number of mf/mg across the four skin snips. The change from pre-treatment in O. volvulus skin mf densities was calculated as the difference between skin mf density at follow up and pre-treatment in absolute terms and as the percentage of pre-treatment density for each follow up time point. The proportion of participants with undetectable levels of skin mf was calculated as the proportion of participants without mf in each of the four skin snips.
The palpable nodules from the participants who attended the 18 month follow up visit and agreed to their excision were excised at that time (35/38 in the 2 mg moxidectin group, 37/37 in the 4 mg moxidectin group, 24/37 in the 8 mg group, 30/42 in the ivermectin group). Nodules were processed as described previously [6], [38]. The histological assessment was based on the examination of three 4 µm sections obtained along the longest axis of each nodule in a way that each third of a nodule was sampled. Worms were examined and classified by a single observer (SKA) according to the criteria described previously [38], [39]. Each macrofilaria identified was classified by sex and as alive, moribound, dead or dead and calcified based on the presence or absence of normal or degenerate cuticle, hypodermis and muscles, of normal, degenerate and/or pigmented intestine and genital tract wall, deposits in the body cavity, nearly complete resorption of the worm, calcification of organs or of the worm, and presence of giant cells. Female macrofilariae were furthermore classified as young or not. The reproductive status of each female macrofilaria was classified based on the presence or absence of an empty genital tract, normal or degenerate oocytes, morulae, gastrulae, coiled mf, stretched mf, sperm in the uterus, polymorphous material, presence of oocytes and remant mf only, or mixed development. Gravid female macrofilariae were categorized as in full production of developmental stages, as in less than full production of developmental stages, as having resumed production of developmental stages and showing the presence of pre-microfilarial stages (morulae and gastrulae) or as going out of embryonic production and showing the presence of coiled mf (pretzels) and stretched mf. Furthermore, the number of mf in the nodule was determined. The reproductive status of male macrofilariae was categorized based on an empty genital tract and presence or absence of normal or degenerate spermatogonia, primary spermatocytes, secondary spermatocytes, spermatids, or spermatozoa [38], [39].
The sample size for each pre-treatment intensity of infection category (mild, moderate, severely infected) and treatment arm was chosen to minimize the number of participants exposed to moxidectin while providing a relatively high probability of detecting frequent adverse events. The probabilities of events with given true event rates occurring with a given sample size can be calculated using the binomial distribution [40]. The planned sample size for each pre-treatment intensity of infection category combined with the 3∶1 moxidectin∶ivermectin randomization ratio provides for each moxidectin dose level a probability of >90% to detect at least 1 event with a true event rate of 20% among the 12 mildly or moderately infected participants, at least 1 event with a true event rate of 10% among the 24 severely infected participants and at least 1 event with a true event rate of 5% among the total of 48 participants with any intensity of infection exposed to a specific moxidectin dose or ivermectin.
The planned 48 participants per treatment group chosen in view of the adverse event detection probabilities, provided approximately 92% power to detect a statistically significant difference between any 2 treatment groups when the percentage of participants with undetectable levels of skin mf was approximately 99.9% in one and 80% in the other group. This estimate is based on a 2-sided Fisher's Exact test with Type I error of 0.05 without adjustment for multiple comparisons which was not necessary due to the hierarchical way comparisons for efficacy were performed (see Statistical Methods).
A total of 172 of 196 planned participants were included in the study and treated (Figure 1). The difference between the number of planned and treated participants is due to the fact that it was not possible to recruit within the protocol specified timelines for each cohort 16 participants who met the eligibility criteria. Screen failure reasons included not meeting criteria for intensity of infection for the cohort for which screening was conducted (56%), laboratory values outside the protocol specified range (26%), ocular disease not meeting the criteria for the cohort for which screening was conducted (7%), hypertension (6%) and others (5%, including age outside protocol specified range, orthostatic hypotension, pregnancy, weight lower than specified in the protocol, history of neurologic/neuropsychiatric disorder/epilepsy).
Of the 172 treated participants (mITT population), 6 discontinued from the study before the 18 months follow up examination (Figure 1), including 1 who died due to a snake bite, 1 who decided to withdraw from the study and 4 who were lost to follow up, i.e. could not be located despite several attempts to find them.
Table 2 shows the demographics and O. volvulus infection related pre-treatment characteristics. Testing for differences between treatment groups in age, sex, height, weight and subjects by intensity of infection showed no statistically significant differences. With the eligibility criteria precluding ocular involvement in approximately 25% of participants (mildly infected cohorts) and limiting it in a further approximately 25% (moderately infected cohorts), ocular involvement overall was very low. Table 2 also shows the number of onchocercal and non-onchocercal nodules determined at histological assessment of all excised nodules at Month 18.
Table 3 shows the number of subjects with different categories of AEs, with Mazzotti reactions for which Fisher's exact test between at least one moxidectin treatment group and the ivermectin treatment group resulted in a p-value of <0.05, and with severe symptomatic postural hypotension (SSPH).
The skin mf densities in the individual participants before (0) and after treatment are shown in Figure 5. Figure 4 shows the percent reduction from pre-treatment in mean mf density after treatment in each treatment group across all participants who completed the study (e-mITT population) and after treatment with ivermectin and 8 mg moxidectin for the subgroup with >20 mf/mg skin pre-treatment.
The decrease in skin mf density after ivermectin treatment (for overview of decrease seen in other studies see [44]) was also observed after treatment with 2 mg, 4 mg or 8 mg moxidectin, but was faster and more extensive (Figure 4, Figure 5). From day 8 onward, the decrease in skin mf density from pre-treatment was significantly higher after treatment with any of the moxidectin doses than after treatment with ivermectin in both the e-mITT and the mITT populations (<0.005). Supporting Table S3 shows descriptive statistics and the results of the statistical analysis of the data for the e-mITT population.
The faster and more extensive decrease in skin mf density was reflected in the proportion of participants with undetectable levels of skin mf (Figure 6). This proportion was significantly higher than in the ivermectin group already on day 8 in the 4 mg and 8 mg moxidectin groups (p<0.02) and from month 1 onward in all moxidectin groups in both the e-mITT and the mITT populations (p<0.01). Supporting Table S4 shows the descriptive statistics and the results of the statistical analysis of the data for the e-mITT population.
In three ivermectin treated participants in the e-mITT population (skin mf density pre-treatment 29.7–62.5 mf/mg), as well as in one participant (skin mf density pre-treatment 19.3 mf/mg) who discontinued from the study after the 12 month follow up, the reduction in skin mf density from pre-treatment to day 8 was <60% and the skin mf density was >6% of the pre-treatment value 3 months after treatment. Thus, the reduction in skin mf density in these participants did not meet the criteria for an ‘adequate parasite response’ to ivermectin defined by Awadzi and coworkers (≥60% reduction in skin mf level from pre-treatment level on day 8 post-treatment, skin mf density ≤6% of pre-treatment density 3 months post-treatment) [41]. These participants are referred to below as ‘Suboptimal Microfilariae Responders’ (SOMR) to be distinguished from ‘Suboptimal Responders’ reported in other studies [41], [42], [45], [46] in whom the reduction in skin mf levels is as expected, but the increase in skin mf levels following the initial decrease is faster and/or more extensive than considered consistent with ‘adequate parasite response’ by Awadzi and coworkers (skin mf density ≤40% of pre-treatment density at 12 months after treatment). Figure 4 B shows the percentage reduction from pre-treatment across all severely infected ivermectin treated participants who completed the study as well as after exclusion of the SOMRs. The conclusions from the statistical analysis of the change in skin mf densities between moxidectin and ivermectin treated groups did not change when the three SOMRs in the e-mITT population were excluded. A response to ivermectin not meeting the criteria for adequate response was also observed in four other ivermectin treated participants in the e-mITT population. These participants are not considered here as SOMRs because the intensity of infection pre-treatment (10.2 mf/mg, <1 mf/mg for three of them) was below that from which the criteria were derived and at low intensity of infection the variability between mf counts in individual snips can have too large an impact on mean skin mf density to support definitive conclusions, even when the skin snip weight is taken into account. No participant treated with moxidectin showed a response fitting the criteria for SOMRs.
Figure 7 A and B show for the e-mITT population and for the subgroup of severely infected participants excluding the three SOMRs, respectively, the time from treatment to recorded skin mf density nadir. Figure 7 C and D show the time of start of recorded sustained increase in skin mf density. In the ivermectin group, three (7.7%) participants showed an increase in skin mf levels as early as 2 months after treatment (0.1–2.53 mf/mg) while three other participants reached the nadir only at 3 or 6 months (undetectable – 2.89 mf/mg). Overall, 85% of ivermectin treated participants who completed the study had undetectable levels of skin mf recorded at one point, including 76% of severely infected participants.
An increase in skin mf levels was first observed among 2 mg moxidectin treated participants 3 months after treatment (n = 1, 0.14 mf/mg skin), among 4 mg treated participants 6 months after treatment (n = 4, 0.08–0.69 mf/mg skin) and among 8 mg moxidectin treated participants 12 months after treatment (n = 12, 0.24–3.3 mf/mg skin, 0.5%–12% of pre-treatment value). The proportion of participants with undetectable levels of skin mf (Figure 6) was statistically significantly higher among moxidectin treated than ivermectin treated participants through month 12 for the 2 mg and 4 mg doses (p<0.02) and through month 18 for the 8 mg dose in both the mITT and the e-mITT population (p<0.05, Supporting Table S4).
The reduction in mean skin mf densities relative to pre-treatment densities was statistically significantly higher among moxidectin than ivermectin treated participants through month 12 for the 2 mg moxidectin dose (p<0.0001) and through month 18 for the 4 and 8 mg moxidectin dose in both the mITT and the e-mITT population (p<0.005, Supporting Table S3).
The average annual reduction in skin mf density from pre-treatment was 88% (median 94%, range 24%–99%, average 89% if the SOMRs are excluded) after ivermectin treatment, 97% (median 98%, range 81%–99%) after 2 mg moxidectin, 98% (median 99%, range 90%–99%) after 4 mg moxidectin and 98% (median 99%, range 96% to 99%) after 8 mg moxidectin.
A total of 245 nodules were excised 18 months after treatment. Histological assessment showed 214 (87.3%) nodules to be onchocercal, including 46/56 (82.1%) nodules from ivermectin treated participants, 66/76 (86.8%) nodules from 2 mg moxidectin treated participants, 57/62 (91.9%) nodules from 4 mg moxidectin treated participants and 45/51 (88.2%) nodules from 8 mg moxidectin treated participants.
Non-onchocercal nodules (1–2/subject) included lipoma, lymphnodes and granulomas around foreign bodies. The types of non-onchocercal nodules include those found in other studies, but the frequency of non-onchocercal nodules was higher than observed in these studies [47], [48]. In 30/166 (18.1%) of participants who completed the study, the number of nodule sites palpated at Month 18 was lower than that palpated pre-treatment, suggesting that some of the nodules palpated pre-treatment were not onchocercal or onchocercal nodules whose resorption had been completed by Month 18.
The number of nodule sites palpated at month 18 was higher than pre-treatment by 1 in 23/163 (14.1%) and by 2 in 4/163 (2.5%) of participants who completed the study and agreed to nodule palpation. Investigator observations suggest that participants becoming more aware of nodule sites after the pre-treatment examination and nodule palpation for preparation of nodulectomies at month 18 follow up occurring under better lighting conditions contributed to the higher number of nodules palpated after than before treatment. An increase in the number of palpable onchocercal nodule sites due to new infections is also possible but unlikely to be significant given that some of the ‘additional’ nodule sites were already observed 1–6 months after treatment.
The number of excised onchocercal nodules/participant ranged from 0 (no palpable nodules or all palpable nodules were non-onchocercal) to 8. There was no trend suggesting a relationship between the age or sex of the participant and the number of excised onchocercal nodules (Figure 8 A), the skin mf density pre-treatment (Figure 8 B) or the skin mf level when the body surface area was taken into account [49] (data not shown).
Figure 9A shows the pre-treatment skin mf densities by sex of the host vs. the number of onchocercal nodules for the 135 participants with a number of palpable nodule sites at month 18≤ the number of palpable nodule sites pre-treatment and who had agreed to excision of all palpable nodules, or who had no palpable nodules. High skin mf densities were observed in some participants with 0 or only 1 onchocercal nodule. This indicates that the palpable onchocercal nodules can represent only a small fraction of the nodules in the body as previously concluded by others (e.g. [50] and references therein). Pre-treatment skin mf densities of some participants with 0 or 1 excised onchocercal nodule were several times higher than those in some of the participants with ≥3 excised onchocercal nodules. This suggests that the fraction of onchocercal nodules accessible for excision varies significantly between individuals. Figure 9B shows for the same set of participants as for Figure 9A that there was no correlation between the number of live or the total number of live and dead female macrofilariae and the skin mf density pre-treatment. This suggests that the macrofilariae accessible through nodulectomy represent only a fraction of those present in the body as well as considerable inter-individual variability in this fraction. Since assessment of macrofilariae was only a secondary objective, nodules and macrofilariae were assessed by only one parasitologist (SKA). Given a level of agreement on the onchocercal nature of nodules and total number of female macrofilariae of 98% and 88%, respectively, between SKA and another parasitologist in a previous study [7], it is unlikely that reading of the slides from this study by a second parasitologist would have resulted in a significantly better correlation between number of nodules or total number of female macrofilariae and skin microfilariae than shown in Figure 9. Furthermore, a high degree of uncertainty about the true female adult parasite burden has been deduced previously from examination of the relationship between macrofilariae in palpable nodules and skin mf density in Burkina Faso and Liberia [51]. Figure 10 shows for each treatment group the skin mf density 18 months after treatment vs. the number of excised live female, live young female and live male macrofilariae. In all treatment groups, undetectable levels of skin mf as well as skin mf densities ≥5 mf/mg occurred in some participants with 0 or 1 live female or male macrofilaria as well as in some participants with ≥4 live female macrofilariae. This suggests again that the excised macrofilariae are not necessarily representative of the macrofilariae in the body.
Summary statistics of the results of the histological assessment are provided in Supporting Table S5. The percentage of female macrofilariae assessed as dead and/or dead and calcified was 50%, 36.5%, 32.2% and 27.7% in the ivermectin, 2 mg, 4 mg and 8 mg moxidectin treatment groups, respectively. Since a single dose of 150 µg/kg ivermectin does not kill macrofilariae (see e.g. [52]–[54]), this indicates a pre-treatment imbalance in the proportions of live and dead female macrofilariae between the ivermectin treatment group and the moxidectin treatment groups. Even if the excised macrofilariae were representative of all macrofilariae in the body, this imbalance makes conclusions about the relative effect of ivermectin and moxidectin on macrofilariae viability impossible. Consequently, the only conclusion on the effect of moxidectin on the macrofilariae the histology data support is that a single dose of 2 mg, 4 mg or 8 mg had neither sterilized all excised macrofilariae to month 18 nor killed all excised macrofilariae by 18 months follow up. Thus, the histology data do not provide any indication of the biological basis of the differences in the skin mf densities seen between treatment groups.
The clinical development plan for moxidectin includes (i) five pharmacokinetic and safety studies in healthy volunteers [21]–[25], (ii) the study reported here, (iii) a large (Phase 3) study in 1500 O. volvulus infected individuals ≥12 year old to determine adverse reactions, including those with a true frequency too low to have been detected in the study reported here, and to assess the relative efficacy of 8 mg moxidectin and ivermectin in individuals from different areas in Africa, and (iv) a paediatric pharmacokinetic and safety bridging study as per discussions with the European Medicines Agency.
The primary role of the study reported here was to determine in a small number of infected individuals whether moxidectin-associated Mazzotti reactions are infrequent enough or have a level of severity which allows to give moxidectin to several hundred people in the Phase 3 study. Review of the blinded data obtained to 1 month after treatment of the last cohort by the sponsors and an external advisory committee, as well as review by the external advisory committee with access to the treatment codes, resulted in the conclusion that 8 mg moxidectin is safe enough to initiate the Phase 3 study which compares 8 mg moxidectin to ivermectin. The investigators agreed based on their blinded assessment that the safety profile in the study was not different from what they had seen in previous studies with ivermectin using similarly close monitoring of participants for Mazzotti reactions. It is noteworthy that the publications of the safety data from these studies did, in contrast to Table 3 here, not include the total number of patients who had experienced at least one Mazzotti reaction (see e.g. [5], [55], [56]). The Phase 3 study has been completed (NCT00790998).
In the study reported here, there were no significant adverse drug reactions other than Mazzotti reactions, which is consistent with the data obtained in healthy volunteers [21]–[25].
The Mazzotti reactions pruritus, rash, increased pulse rate and decreased mean arterial pressure after 2 minutes standing still following ≥5 minutes supine occurred significantly more frequently among 8 mg moxidectin than ivermectin treated participants. The majority of these reactions were mild or moderate and all, including severe ones, resolved without treatment. This contributed to the decision to initiate the Phase 3 study.
While Fisher's exact test did not return a p value<0.05 for the comparison of the frequency of severe symptomatic postural hypotension (SSPH) after moxidectin and ivermectin treatment, consideration is given here also to SSPH. SSPH is a transient phenomenon observed at the Onchocerciasis Chemotherapy Research Center (OCRC) when a person cannot tolerate standing still for 2 minutes following ≥5 minutes supine (see OCRC Common Toxicity Criteria in Supporting Table S1). SSPH disappears rapidly after lying down [43] and OCRC staff observations of participant behaviour following an SSPH episode show that it does not interfere with resumption of their normal daily activities. In OCRC studies, SSPH incidence among participants treated with 150 µg/kg ivermectin was variable and sometimes high (e.g. 11% [57] or 22% [58]). In contrast, SSPH incidence reported from large scale use of ivermectin is in most cases low or 0, even in hyperendemic areas, when higher than standard doses were used or when significant decreases in standing MAP were measured [43], [59]–[64]. SSPH has furthermore not been an impediment to CDTI covering by now >75 million people [65]. OCRC experience shows that SSPH occurs when study participants get up (e.g. as part of the procedure to assess postural hypotension or after a bed rest) and immediately or shortly thereafter have to stand still (e.g. during the OCRC procedure or while urinating). SSPH is not usually seen when participants get up and move around naturally. This link between SSPH and standing still could explain the discrepancies between the results in OCRC studies and reports from large scale use of ivermectin. Underreporting by the treated individuals during large scale use is another possible explanation consistent with the OCRC observations that the symptoms of SSPH (severe dizziness, weakness, faintness) are shortlived and do not interfer with resumption of normal activities. This could result in SSPH not being a ‘memorable’ experience reported to the investigators. Consequently, the data to date do not suggest that a higher frequency of SSPH after moxidectin vs. ivermectin treatment, if shown to be statistically significant in the Phase 3 study (which used OCRC procedures), will be an impediment for evaluating moxidectin in community studies. These studies would allow to assess the potential significance of SSPH for mass treatment when appropriate advice is given to participants as in the early community studies of ivermectin [59].
Four participants treated with ivermectin did not show a decrease in skin mf levels meeting the criteria of adequate response to the microfilaricidal effect of ivermectin defined by Awadzi et al. [41]. Given that the participants were recruited from an area in Ghana without mass-treatment with ivermectin at the time of recruitment for this study, it is unlikely that this reflects drug pressure-induced selection of parasites with low susceptibility to ivermectin's microfilaricidal activity. It, thus, needs to be considered that the variability of the response of O. volvulus to the microfilaricidal activity is larger than had been observed by Awadzi et al. at the time they analysed their available data to derive criteria for adequate parasite response [41]. A larger variability of the response of O. volvulus to the embryostatic effect of ivermectin than considered ‘adequate’ in some studies [41], [66] has been deduced from a meta-analysis of the data from a large number of clinical and field studies [67].
In this study, a single dose of 2 mg, 4 mg or 8 mg moxidectin resulted in a significantly lower skin mf density and higher proportion of participants with undetectable skin mf earlier and for a longer period of time than ivermectin. Detectable levels of skin mf 18 months after treatment in 83%, 73% and 65% of participants treated with 2 mg, 4 mg and 8 mg, respectively, show that a single dose of moxidectin did not prevent skin repopulation with mf in all participants and thus did not kill or sterilize (permanently or to month 18) all macrofilariae in all participants. The histology data do not allow further conclusions about the biological basis of the long term differences in skin mf densities between the treatment groups, because of the pre-treatment imbalance between treatment groups in the percentage of dead and/or dead and calcified female macrofilaria (Supporting Table S5) and because the macrofilariae in the palpable onchocercal nodules were not representative of all macrofilariae in the body (Figures 9, 10). This is not surprising given the small sample size, chosen based on safety considerations, and not in view of allowing conclusions on treatment differences in the effect on the macrofilariae.
Determining a sample size sufficient to conclude with a pre-specified power and significance level that the effect of two treatments differs by at least a pre-specified amount requires a good estimate of the effect size in the comparator arm and the variance in the efficacy parameter (see e.g. [68], [69]). For macrofilariae, the comparator effect size may vary with the endemicity of the area and treatment history of the study area and participant. The effect size variance depends on biological and methodological factors which include: (i) inter-individual variability in the fraction of total nodules in the body which the excised nodules represent (Figure 9A), (ii) variability in the fraction of total worms of each category (female, male, live, dead, …) which the excised worms represent and which is not necessarily the same as the fraction of total nodules (Figure 9, Supporting Table S5), (iii) variability between the macrofilariae within each participant in the variable evaluated (e.g. age, reproductive activity [70]), (iv) method-specific factors such as for histology the extent to which the number of sections per nodule examined permits representative characterization of all worms in the nodule, quantitative vs. semiquantitative assessment, inter-observer variability (see e.g. [7], [71]) and (v) variability in macrofilariae exposure and susceptibility to the drug (e.g. macrofilariae age dependent, nodule location dependent). Analysis of the pooled raw data from different past studies may help quantitate some of these variabilities and provide the basis for calculating sample sizes for future studies. This may also help resolve the question of the extent to which cumulative doses of ivermectin affect macrofilariae reproductive capacity and viability. As pointed out for the reproductive capacity, this has significant implications for whether, where and how elimination of O. volvulus infection with ivermectin can be achieved [72].
The pre-treatment imbalance in this study in the percentage of female macrofilariae assessed as dead and/or dead and calcified between the ivermectin and moxidectin treatment groups (Supporting Table S5) shows the importance of stratification of study participants for randomization to treatment groups by number or at least proportion of dead and live macrofilariae in the palpable nodules (determined with a non-invasive method such as ultrasonography [73]) to reduce the probability of false conclusions from post-treatment data.
Given the absence of data on the biological basis for the low skin mf levels 12–18 months after moxidectin treatment, the relative value of moxidectin and ivermectin for reducing disease transmission will be considered without any assumptions about this basis. The transmission model developed by Duerr and coworkers [12], [13] includes examination of the effect of a drug which reduces skin mf density irrespective of the effect on the macrofilariae. It shows that mass-treatment with a skin mf reducing drug can lead to elimination by increasing the threshold biting rate (TBR), i.e. the annual biting rate (ABR, number of vectors taking a blood meal from one person/year) below which onchocerciasis cannot remain endemic. With increasing efficacy of the intervention, quantified as the annual average reduction (AAR) in skin mf density in the population, the TBR increases in a non-linear manner: in an area with a TBR without intervention of 700 bites per person per year, the model predicts a TBR of approximately 1200, 2000 and 5000 when the AAR is around 65%, 80% and 95%, respectively. The AAR after treatment with ivermectin and 8 mg moxidectin in this study were 88% and 98%, respectively. The AARs after treatment of communities would be different because of different distribution of pre-treatment skin mf densities and treatment coverage. However, and provided the relative superiority of 8 mg moxidectin over ivermectin is confirmed in the Phase 3 study, moxidectin would have a higher AAR than ivermectin suggesting that, according to this model, moxidectin could lead to elimination in areas with higher ABRs than ivermectin.
CDTI occurs at a time chosen by the community and is in many cases furthermore dependent on logistical conditions (e.g. availability of funds for activities the public health system needs to conduct to enable CDTI). This results in CDTI in areas with seasonal transmission not always happening within the time window optimal for achieving elimination, i.e. a time that results in the lowest skin mf levels in the population when the vector population is largest. The longer period of undetectable skin mf levels after moxidectin compared to ivermectin would reduce the negative impact of community treatment occurring outside the optimal time window on transmission.
|
10.1371/journal.ppat.1003583 | Phosphoproteomic Analyses Reveal Signaling Pathways That Facilitate Lytic Gammaherpesvirus Replication | Lytic gammaherpesvirus (GHV) replication facilitates the establishment of lifelong latent infection, which places the infected host at risk for numerous cancers. As obligate intracellular parasites, GHVs must control and usurp cellular signaling pathways in order to successfully replicate, disseminate to stable latency reservoirs in the host, and prevent immune-mediated clearance. To facilitate a systems-level understanding of phosphorylation-dependent signaling events directed by GHVs during lytic replication, we utilized label-free quantitative mass spectrometry to interrogate the lytic replication cycle of murine gammaherpesvirus-68 (MHV68). Compared to controls, MHV68 infection regulated by 2-fold or greater ca. 86% of identified phosphopeptides – a regulatory scale not previously observed in phosphoproteomic evaluations of discrete signal-inducing stimuli. Network analyses demonstrated that the infection-associated induction or repression of specific cellular proteins globally altered the flow of information through the host phosphoprotein network, yielding major changes to functional protein clusters and ontologically associated proteins. A series of orthogonal bioinformatics analyses revealed that MAPK and CDK-related signaling events were overrepresented in the infection-associated phosphoproteome and identified 155 host proteins, such as the transcription factor c-Jun, as putative downstream targets. Importantly, functional tests of bioinformatics-based predictions confirmed ERK1/2 and CDK1/2 as kinases that facilitate MHV68 replication and also demonstrated the importance of c-Jun. Finally, a transposon-mutant virus screen identified the MHV68 cyclin D ortholog as a viral protein that contributes to the prominent MAPK/CDK signature of the infection-associated phosphoproteome. Together, these analyses enhance an understanding of how GHVs reorganize and usurp intracellular signaling networks to facilitate infection and replication.
| Systems-level evaluations of infection-related changes to host phosphoprotein networks are not currently available for any gammaherpesvirus (GHV). Here we describe a quantitative phosphoproteomic analysis of productive GHV replication that demonstrates alterations in the phosphorylation status of more than 80% of host phosphoproteins and identifies 18 viral phosphoproteins. Systematic bioinformatics analyses reveal a predominance of MAPK and CDK signaling events within infected cells and suggest a virus-induced reorganization of signal-transduction pathways within the host phosphoprotein network. Functional experiments confirmed that CDKs and ERK MAPKs facilitate efficient viral replication and identify transcription factor c-Jun as a potential downstream target contributing to MHV68 replication. Finally, we identify the viral cyclin D ortholog as a major pathogen-encoded factor contributing to the MAPK/CDK signature of the infected cell phosphoproteome. These data provide new insight into both viral and host factors that regulate phosphorylation-dependent signaling during lytic GHV replication and offer a new resource for better defining host-pathogen interactions in general.
| Post-translational modification of proteins by phosphorylation and dephosphorylation regulates numerous functional properties, including activation status [1], stability [2], protein-protein interactions [3], and subcellular localization [4]. Such signals regulate the majority of cellular processes ranging from cell-cycle progression [5], [6] to terminal differentiation of specific cell types [7] to activation of intracellular signals that trigger both local and organismal antimicrobial responses [8]. Following infection of host cells, viruses and intracellular bacteria manipulate cellular signaling to facilitate replication. Pathogen-directed signaling may mobilize enzymatic pathways to provide nutrients or energy necessary for the large increase in macromolecular biosynthesis [9] or reorganize host components to direct packaging, envelopment, or egress [10]. In defense, host-cell sensing of microbial infection may trigger signaling cascades aimed at hindering pathogen replication and alerting neighboring cells to the present danger [8]. Pathogens also may encode factors to deregulate anti-microbial signaling pathways in order to prevent detection or elimination by host immune responses [11].
Recent innovations coupling affinity-based phosphopeptide enrichment with high-resolution mass spectrometry followed by systematic bioinformatics analyses have enabled systems-level evaluations of phosphorylation-dependent signaling cascades in cells or tissues responding to discrete stimuli, such as epidermal growth factor receptor stimulation or DNA damage responses (DDR) [12], [13]. Such analyses revealed that >90% of detectable phosphorylation sites on cellular phosphoproteins were not previously identified [13] and that critical regulatory phospho-motifs and phosphorylated effector proteins remain to be identified, even for extensively studied signaling cascades [12], [13]. Currently, systems-level phosphoproteomic analyses to define infection-associated alterations in protein phosphorylation status during viral infection are lacking. Thus, while hypothesis-driven and intuition-based studies have identified many phosphorylation-dependent signaling events that regulate viral replication and host responses to infection, it is likely that the vast majority of infection-associated changes in host protein phosphorylation status are not yet known. This highlights a critical gap in our current understanding of virus-host interactions. Importantly, the identification of unappreciated signaling pathways and/or effector proteins usurped or inhibited by pathogens in infectious disease states may reveal new targets for pharmacologic intervention.
Gammaherpesviruses (GHVs) are members of the Herpesviridae family of large double-strand DNA viruses [14]. GHVs include the human pathogens Epstein-Barr virus (EBV) and Kaposi sarcoma-associated herpesvirus (KSHV or HHV-8); non-human primate viruses herpesvirus Saimiri (HVS), rhesus rhadinovirus (RRV), and rhesus lymphocryptovirus (rhLCV); and rodent pathogens wood mouse herpesvirus (WMHV), rodent herpesvirus Peru (RHVP), and murine gammaherpesvirus-68 (γHV68 or MHV68). Like all herpesviruses, GHVs exhibit two distinct phases of their infectious cycles. The productive replication phase (also termed lytic replication) is characterized by robust viral gene expression leading to viral DNA replication and the production of infectious progeny virions. In contrast, latent infections are characterized by restricted viral gene expression and indefinite maintenance of the viral genome as episomal DNA. GHVs characteristically establish lifelong latent infections of lymphocytes, thereby placing the host at risk for lymphoid and other cancers, especially in settings of immunocompromise such as HIV infection or immunosuppression for organ transplants [15], [16], [17].
In contrast to EBV and KSHV, MHV68 – which provides a tractable small animal model for evaluating GHV pathogenesis – undergoes robust productive replication in cultured cells. Further, given the ease of generating MHV68 mutants and the capacity to perform controlled experimental infections of various wild-type (WT), knockout, and transgenic mice, MHV68 offers an attractive system for understanding the virus-host dynamic during productive viral replication [18], [19], [20]. Akin to what is hypothesized to occur during primary EBV infection of humans, acute MHV68 infection of mucosal epithelia following intranasal inoculation is necessary for viral dissemination and latency establishment in distal reservoirs [21], [22], [23]. While the importance of DDR, mitogen-activated protein kinase (MAPK), and inhibitor of kappa-B kinase (IKK) signaling in MHV68 replication recently was demonstrated [24], [25], [26], [27], an understanding of how the kinases involved in these responses influence the overall phosphoprotein milieu within the host cell is not known. The breadth of host or viral proteins targeted by specific viral signaling proteins with critical roles in pathogenesis, such as the conserved herpesvirus protein kinase (CHPK) ORF36 [28] or the viral cyclin D ortholog, encoded by gamma-2-herpesviruses such as KSHV [29] and MHV68 [30], also has not been globally evaluated. Hence, our understanding of GHV-regulated signaling events – as well as host cell responses to infection – in facilitating productive viral replication and ultimately pathogenesis is incomplete. To this point, a better appreciation of phosphorylation-dependent signaling during GHV replication may illuminate novel targets for the treatment of infectious mononucleosis, the acute phase malady of primary EBV infection [31], or Kaposi sarcoma, a KSHV-related cancer for which lytic viral replication is thought to be a driver of disease [32], [33].
In this study, we use a comparative, quantitative phosphoproteomic analysis to define phosphorylation-related signaling events regulated during productive MHV68 infection. We identified more than 400 host phosphoproteins that are induced or repressed in infected cells, as well as 17 viral phosphoproteins. Quantitative analyses indicated that a vast majority of definable phosphopeptides was regulated during GHV infection. Unbiased bioinformatics analyses and complementary biochemical approaches predicted the importance of extracellular-signal related and cyclin-dependent kinases (ERK and CDK, respectively) in facilitating viral replication, and pharmacologic inhibition and shRNA knockdown experiments confirmed the predictions. Finally, we identified the MHV68 cyclin ortholog as a key contributor to many of the virus-associated signaling changes observed. Together, these data and analyses provide a novel systems-level resource to better inform and enable studies of pathogen-host interactions.
To gain a more complete understanding of GHV-induced changes in host cell signaling networks, we devised an experimental approach integrating high-resolution mass spectrometry, predictions based on orthogonal bioinformatics analyses, and hypothesis-driven functional tests (Fig. 1). We first performed time-course immunoblot analyses using phospho-residue-specific antibodies to determine timepoints at which MHV68 elicited robust changes in protein phosphorylation patterns during productive infection of mouse fibroblasts. These experiments revealed the greatest qualitative differences in phosphoprotein detection between control and infected samples at 18 h post-infection (Fig. 2A and S1). At this timepoint, phosphopeptides were selectively enriched from mock-infected and infected cells using titanium dioxide (TiO2) chromatography. Three separate enrichments were performed for each of two independent experiments, and peptides were sequenced by HPLC-coupled high-resolution MS/MS analysis using collision-induced dissociation on an LTQ Orbitrap hybrid mass spectrometer. The analysis identified 13,801 peptides, ca. 80% of which contained detectable phosphorylated residues. As only 30% of the cellular proteome is estimated to be phosphorylated [34], this indicates the enrichment procedure was robust. The processed and sorted data sets are available as Table S1. Despite identifying peptides from 14% fewer distinct proteins in infected samples, the absolute numbers of phosphopeptides identified for control and infected samples was essentially equivalent (5632 vs. 5472, respectively), which indicates that the enrichment and analysis procedures did not introduce bias. Notably, approximately 15% of phosphopeptides identified from infected cells are specific to the infected state (i.e., only detected in infected analytes), 4% of which derive from viral proteins, whereas 11% of control phosphopeptides were not detected in infected cells (Fig. 2B). Extended to the protein level, we identified a total of 989 individual proteins representing 845 proteins from control cells and 728 proteins from infected cells (Fig. 2B). 584 of the identified proteins were present in both control and infected samples, while 261 or 144 proteins were unique to either control or infected cells, respectively (Fig. 2B). 22 of the 144 infection-specific proteins (ca. 15%) were derived from MHV68.
For quantitative analyses, the relative intensity (a quantitative measure of protein abundance) for each identified peptide was determined using MaxQuant [13]. Of the 2428 unique peptides identified, 86% exhibited greater than 2-fold change in peptide intensity between control and infected cells. A dot plot of summed peptide intensity vs. intensity ratio demonstrates the extent to which common phosphopeptides – those identified in both control and infected samples – are up or down-regulated during MHV68 infection (Fig. 2C). By comparison, recent studies to define phosphoproteomic changes induced by discrete stimuli, such as growth factor stimulation or DNA damage, found that less than 15% of phosphoproteins were regulated (enhanced or decreased phosphorylation) in response to stimulus [12], [13]. Similarly, Hela cell challenge with Salmonella also elicited a more modest change in host-cell phosphoprotein status than MHV68 infection, with ca. 24% of phosphopeptides being regulated [35]. The breadth of differential phospho-protein regulation induced by MHV68 is further highlighted by gene ontology (GO) analyses, which revealed that infection-associated changes in protein phosphorylation status were evident in essentially all ontologically-grouped protein classes, rather than effecting specific types of proteins (Fig. 2D). Interestingly, all MS-identified proteins annotated to the “kinase” protein class exhibited phosphorylation status changes during MHV68 infection (Fig. 2D). This also was true for all proteins annotated to “protease” and “oxidoreductase” protein classes (Fig. 2D). Thus, MHV68 infection effects dramatic changes in the host-cell phosphoproteome on a scale not previously observed in comparable phosphoproteomic analyses.
Several host proteins are notable among those specifically induced in or lost from infected cells. The promyelocytic leukemia protein (PML) is the single most abundant phosphoprotein detected in control cells, yet absent from infected cell analytes (Fig. 2E). This finding is consistent with PML being targeted for degradation by MHV68 tegument protein, ORF75C [36], [37]. Infected samples, on the other hand, displayed highest levels of phosphorylated linker histone H1 variant 1, MAPK1/ERK2, and TAR DNA binding protein (TARDBP), a protein originally identified as a host factor that enhances HIV transcription [24]. MAPK3/ERK1 and MAPK substrate c-Jun also were exclusive to infected cells. Although roles for ERK1/2 in MHV68 replication have not been described, ERK1/2 were previously implicated in facilitating KSHV replication [38], [39], [40]. Further, we recently demonstrated c-Jun phosphorylation and related AP-1 transcription factor activation during MHV68 replication [24]. Thus, several MS-defined phosphoproteins identified in our study are consistent with previously published findings, thereby providing confidence in the phosphoproteomic data sets obtained.
We also identified 18 viral phosphoproteins and non-phosphorylated peptides derived from 5 viral proteins (Table 1). Of the viral phosphoproteins identified, to our knowledge only tyrosine phosphorylation of ORF21/thymidine kinase (TK) was previously reported for MHV68 [41]. In addition to defining the specific phospho-tyrosine residues previously reported, we also identified numerous phosphorylation events on serine and threonine residues within ORF21/TK (Table 1). Six of the MHV68 phosphoproteins we identified [ORF8/gB, ORF21/TK, ORF25/major capsid protein (MCP), ORF27/gp48, ORF39/gM, and ORF45] correspond to homologous phosphoproteins present in purified EBV virions [42]. Phosphorylation of ORF21/TK [41], ORF45 [43] and viral DNA polymerase processivity factor ORF59 [44] also were demonstrated during KSHV infection. Interestingly, phosphorylation of pUL44, the HCMV homolog of ORF59 [45], [46], [47], [48], and glycoprotein B (gB) phosphorylation, during both HCMV and HSV1 infection [49], [50] suggest that some of these phosphorylation events may be functionally conserved among all herpesvirus subclasses. Also identified were phosphorylated peptides for ORF36, a conserved herpesvirus protein kinase which by analogy to HCMV and HSV should be capable of autophosphorylation [28]. The identification of discretely phosphorylated residues in viral proteins, especially those conserved for other herpesviruses, provides a resource for future functional studies to define roles for specific phospho-motifs and related kinases in the processes of infection, persistence, and pathogenesis.
We next performed a series of comparative immunoblot analyses to validate the MS data. In agreement with MS determinations, ERK1/2 was potently phosphorylated following MHV68 infection of murine fibroblasts, while total ERK1/2 levels remained unchanged (Fig. 3A). Infection-associated c-Jun phosphorylation on Ser73 also was readily detected. As for differentially regulated phosphoproteins (i.e., those present at differing abundance in both control and infected MS data sets), quantitative MS analyses indicated that MHV68 infection enhanced phosphorylation of the myristoylated alanine-rich C-kinase substrate (MARCKS), a classic protein kinase C (PKC) targeted scaffold protein (highlighted in Fig. 2C). By immunoblot, infection-related enhancement of MARCKS phosphorylation manifested as a retarding of MARCKS mobility during SDS-PAGE (Fig. 3A). In contrast to proteins for which phosphorylation was enhanced during infection, PML protein only was detected in control cell lysates (Fig. 3A), which is consistent with the presence of PML phosphopeptides in control, but not infected, MS analyses. As stated above, this finding is consistent with PML protein being targeted for degradation by the MHV68 tegument protein, ORF75C [36], [37].
To validate viral phosphoproteins, we evaluated whether ORF21/TK and ORF59, the two most abundant viral phosphoproteins identified (Table 1), were phosphorylated during MHV68 infection. For ORF21, phosphoproteins were captured by immunoprecipitation with p-Thr, p-Ser, and p-Tyr-specific antibodies. Immunoblot analyses readily detected ORF21 in phospho-specific immunoprecipitates (Fig. 3B). Interestingly, ORF21 mobility in SDS-PAGE corresponded to the abundant ca. 80 kDa phosphoprotein detected in infected cell lysates (see Fig. 2A). Likewise, p-Thr and p-Ser-specific antibodies also recognized immunoprecipitated ORF59 (Fig. S2). These data provide complementary biochemical evidence that supports the identification of ORF21 and ORF59 as phosphorylated viral proteins by MS. Taken together, the results of these experiments biochemically validate the presence or absence of several viral and host phosphoproteins identified through global phosphoproteomics analyses. Thus, these data provide additional confidence in the robustness of the MS data sets.
Because our MS analyses focused on timepoints associated with robust infection-induced alterations in total phosphoprotein profiles, we next sought to correlate infection-associated phosphoproteomic changes with specific stages in the viral replication cycle. Cells were mock-infected or infected with WT MHV68, UV-inactivated (UVI) MHV68, ORF50-null (50.Stop) MHV68, or WT MHV68 in the presence of cidofovir, a nucleoside analog that blocks viral DNA replication and hinders progression into the late phase of the viral replication cycle [51]. Due to disruption of the gene encoding the viral transactivator protein, RTA, ORF50-null MHV68 arrests at the immediate-early (IE) gene expression stage [52]. Consistent with time-course analyses suggesting that the majority of infection-associated phosphoproteomic changes occur during early-to-late stages of infection (Fig. S1), ERK1/2 phosphorylation, c-Jun phosphorylation, and the retardation of MARCKS mobility only were evident in cells infected with WT MHV68 or WT MHV68 in the presence of cidofovir (Fig. 3C). The findings that UVI and ORF50-null MHV68 did not elicit changes in the phosphorylation status of the proteins tested indicate that active viral gene expression and progression through the replication cycle, not simple internalization of the virus or presumed initiation of IE transcription, respectively, are important for infection-associated induction of host phosphoproteins. Additionally, while cidofovir reduced viral protein production as evidenced by immunoblot analyses with MHV68 antiserum (Fig. 3C) and ca. 20-fold reduction in viral titers (not shown), cidofovir did not inhibit ERK1/2, c-Jun, and MARCKS phosphorylation (Fig. 3C). These data suggest that viral DNA replication, and most likely late viral gene expression, do not play major roles in the infection-related signaling events evaluated. These data agree with a previous study in which JNK1/2 and c-Jun phosphorylation occurred during MHV68 infection despite inhibition of viral DNA synthesis with phosphono-acetic acid [24]. Further, while virus-related signaling undoubtedly occurs coordinate to other steps in the infection process, results of these experiments support the notion that our phosphoproteomic analyses offer an accurate representation of signaling events corresponding to the early-to-late phases of the viral replication cycle.
Finally, in an effort to link cell culture observations to the in vivo setting, we evaluated c-Jun phosphorylation in infected cells during acute MHV68 replication. Mice were intraperitoneally (IP) inoculated with recombinant MHV68 expressing a histone H2B-YFP fusion protein as a fluorescent marker to enable detection of infected cells by flow cytometry [53]. Four days after IP inoculation with MHV68, a timepoint at which robust productive viral replication is occurring in the spleen [54], splenocytes were harvested and processed for flow cytometry to detect H2B-YFP and c-Jun phosphorylated on Ser73. The H2B-YFP gating strategy and an analogous proof-of-principle experiment in productively infected fibroblasts are provided in Figure S3. Compared to splenocytes from mock-infected animals and H2B-YFP-negative cells from infected animals, H2B-YFP-positive cells exhibited a significant increase in the detection of phosphorylated c-Jun, with an approximate 2-fold increase in mean fluorescence intensity (Figs. 3D and 3E). This approximated the induction of c-Jun phosphorylation observed in productively-infected fibroblasts in culture (Fig. S3). The finding that H2B-YFP-negative cells from infected animals exhibit phospho-c-Jun signals approximating those of mock-infected animals is important, because it demonstrates that c-Jun phosphorylation during acute MHV68 infection occurs specifically in infected cells, rather than through a non-specific bystander process such as immune activation [55], [56]. These findings provide an important experimental link suggesting that signaling parallels exist between productive replication in culture and during MHV68 pathogenesis in vivo.
Having validated the initial MS data sets, we next performed a battery of bioinformatics analyses to determine the functional consequences of phosphoproteomic changes during MHV68 infection and potential signaling pathways involved. Given the extent to which infection altered the phosphorylation status of host proteins, we first sought to determine whether infection elicited a rearrangement of the global cellular phosphoprotein network. A central tenet of systems biology analyses is the assertion that most if not all components of a complex network exert some influence over other components of the system [57], [58]. As such, network analyses predict that changes in the “interactions” (i.e., functional association as defined by many disparate properties) between specific molecules, especially those positioned downstream of multiple stimuli or between functional protein modules, can have dramatic influences on the propagation of signals through, and ultimately the response of, a particular system [57], [58]. While numerous linear analyses of signaling pathways that regulate GHV replication have been performed, how infection by GHVs modulates the host phosphoprotein network is not known. We therefore performed comparative network analyses to provisionally identify critical host molecules and protein complexes that were manipulated and reorganized during MHV68 infection.
We first performed STRING (Search Tool for the Analysis of Interacting Genes/Proteins) analyses [59] to establish global phosphoprotein interaction networks for all proteins detected in either control or infected cells (Figs. S4 and S5, respectively). The STRING algorithm links genes or proteins into networks based on published functional or informatics-predicted interactions [59]. Phosphoprotein networks were then processed in Cytoscape [60], and interconnected proteins (referred to as nodes) were color coded based on MaxQuant intensities to denote detection and/or relative abundance. Node size correlates to betweenness centrality (BC), a measure of a protein's capacity to connect disparate protein modules in the network [61], thus allowing a visualization of nodes that likely have a strong influence on signal propagation through the networks. The width of connecting arrows between nodes (known as edges) represents the STRING-defined confidence value of the predicted interaction. Disconnected nodes, those that did not “interact” with any other proteins evaluated (see Fig. S4 and S5), are not shown.
Upon gross inspection, it was immediately evident that control and infected protein networks were topologically unique (compare Figs. 4 and 5, note the repositioning of highlighted common nodes during infection relative to control). Interestingly, several proteins specifically induced (c-Jun, MAPK1/ERK2, MAPK3/ERK1, and others) or lost (AKT1, PCNA, Smad2, and others) from the infected phosphoproteome have high BC values indicated by relatively large node size (Figs. 4 and 5). These findings suggest that the induction or repression of specific proteins during infection alters the flow of information from protein to protein within the cell [57], [58]. MCODE analyses to identify functional protein clusters within the networks [62] identified either unique clusters or clusters in which the composition of proteins represented varied according to infection (Figs. 4 and 5). Indeed, none of the highest scoring clusters were identical, which further indicates reorganization of functional protein modules in the network during MHV68 lytic replication. Finally, GO analyses utilizing the Cytoscape plugin BiNGO [63] indicate that induction or repression of specific proteins during MHV68 infection alters many functionally grouped biological processes represented in the phosphoprotein network (Table S2). Of note, GO IDs associated with chromatin organization, RNA production and localization, and negative regulation of cell death were unique or overrepresented during MHV68 infection, whereas biological processes associated with hormone-related signaling pathways, ubiquitylation, and signal transduction were absent in comparison to control networks (Table S2). These findings suggest that infection-directed changes in the phosphorylation status of specific proteins redirects the flow of information through the cellular phosphoprotein network to effect broad functional changes to specific biological processes.
We next performed a comparative analysis of Kyoto encyclopedia of genes and genomes (KEGG) pathways [64] that were represented in control and infected phosphoprotein networks (Fig. 6A). This analysis revealed a high degree of overlap between mock and infected data sets for thirteen KEGG-defined pathways, although the majority of redundant pathways do contain several proteins whose presence or abundance was influenced by infection. This point is illustrated by extracting proteins represented in the “spliceosome” KEGG pathway from the global phosphoprotein network. Five spliceosome proteins were absent from infected cells (solid blue), two were only detected in infected cells (solid red), and thirteen exhibited differing abundance between control and infected cells (blue or red outlines, respectively) (Fig. 6B). The phosphorylation status of only five spliceosome proteins remained unchanged. KEGG analyses also predicted several functional protein modules exclusively lost or induced during infection. For instance, control cells contained a high number of phosphoproteins annotated to the ubiquitin-mediated proteolysis pathway. Indeed, 5 of the 11 proteins represented were absent from infected cells, while the remaining 6 were less abundant during infection (Fig. 6C). Conversely, the MAPK signaling pathway was exclusively represented by infected phosphoproteins, including MAPK3/ERK1, MAPK1/ERK2, MAP4K4, c-Jun, and epidermal growth factor receptor (Fig. 6D). These data indicate that MHV68 infection modulates the phosphorylation status of specific functional protein modules despite a seemingly global redistribution of ontologically grouped protein phosphorylation.
Distinct protein kinases phosphorylate specific amino-acid motifs present on target proteins. To determine if distinct phospho-motifs were overrepresented in the infection-associated data set relative to control identities, we segregated all individual phosphopeptides according to their presence or absence within control or infected cells. Sequence logos generated for each data set using the ICE-LOGO resource (http://iomics.ugent.be/icelogoserver/logo.html), a weighted representational analysis, demonstrate general differences in phosphorylated sequences between control and infected systems (Figs. 7A and 7B, respectively). Distinct phospho-motifs in each data set were identified using the Motif-X algorithm [65] (Fig. 7C). Interestingly, only two shared phospho-motifs were overrepresented in both control and infected phosphopeptides, S*XXE and XS*PX (Fig. 7C), where the asterisk designates the phosphorylated residue, relative to the background Mus musculus phosphoproteome. Control phosphopeptides exhibited more promiscuity in motif representation, with eight specific phospho-motifs identified as being enriched. A number of motifs present in control peptides were characterized by acidic residues downstream and an Arg upstream of the phospho-acceptor, including an AKT-related RXXS* motif (Fig. 7C). In contrast to control peptides, only three unique phospho-motifs were enriched during MHV68 infection (Fig. 7C). Each of these exhibited Pro-directed phospho-acceptors reminiscent of CDK and MAPK target sequences, including a classic CDK motif characterized by a basic residue at the -2 amino acid position relative to the phosphorylated residue [66]. Of note, detection of phosphorylated CDK1 and CDK2 was reduced during infection compared to control samples (Table S1). However, the CDK1/2 phosphorylations detected in our MS analyses correspond to deactivating the post-translational modification [67], [68]. Hence, a relative decrease in abundance within infected cells corresponds to CDK1/2 activation during infection, which was previously documented during productive MHV68 infection [69]. Additionally, the ERK1/2 phosphopeptides detected are indicative of activation (Table S1) [70]. Thus, the infection-associated phosphoproteome exhibits a strong CDK/MAPK phosphorylation signature, which differs from that of uninfected cells. This finding is in agreement with the identification of activated ERK1/2 and CDK1/2 by MS during infection and KEGG biological pathway predictions.
To directly test these predictions, we performed comparative immunoblot analyses using phospho-motif-specific antibodies directed against the AKT RXXS* target motif or MAPK/CDK XS*P/S*PXK motifs following mock-infection or infection with MHV68 (Fig. 7D). Although two unique RXXS*-containing proteins were prominent in infected cells, infection resulted in a general reduction in the number of detectable AKT-phosphorylated proteins. In contrast, infection resulted in enhanced detection of MAPK/CDK motif-containing proteins (Fig. 7D). As was the case with general phosphoprotein immunoblotting (Fig. S1), detection of MAPK/CDK phospho-motifs increased over time during infection (Fig. S6), which is consistent with the observation that ERK1/2 activation occurs during the early-to-late phase of the MHV68 replication cycle (See Fig. 3). These data independently verify the motif-based bioinformatics prediction that infection substantially alters the cellular signaling network, revealing that MAPK/CDK-related phosphorylation events are predominant in the infected system, while other signaling pathways apparently are repressed. These findings strongly suggest prominent roles for CDK and/or ERK signaling during MHV68 infection.
To gain insight into specific host proteins potentially regulated by ERK1/2 and/or CDK1/2 activity during infection, we performed high-confidence group-based phosphorylation scoring (GPS) analyses [71] to define infection-specific phosphopeptides that contain ERK1/2 and CDK1/2 motifs. The GPS analysis data table includes predictions for all available kinases for infection-specific phosphopeptides (Table S3), although other kinase-motif predictions are not discussed here. GPS analysis predicted 220 unique phosphopeptides derived from 171 host proteins that contain CDK1 and/or CDK2 motifs and 150 unique phosphopeptides derived from 103 host proteins for ERK1 and/or ERK2 (Fig. 7E and Table S3). 98 of the 253 predicted targets contained high-confidence motifs for both CDKs and ERKs, while 155 proteins were distinct targets – 150 CDK and 5 ERK (Fig. 7E). STRING analyses identified only 11 of these proteins as CDK or ERK interactors (Fig. 7E). Remarkably, of the 23 most abundant infection-specific host phosphoproteins illustrated in Figure 2D (not including ERK1 or ERK2), 18 contain GPS-predicted phosphorylation events on CDK and/or ERK target motifs. Of these, only c-Jun was a STRING-defined interaction partner for both CDKs and ERKs, which highlights it as a prioritized candidate for functional studies. These data strongly support the hypothesis that CDK and ERK-related signaling are predominant during productive MHV68 infection and further suggest that this serves to regulate a core set of proteins in the cellular phosphoprotein signaling network.
Three independent bioinformatics analyses strongly predict the importance of CDK and/or MAPK activity in productive MHV68 replication. First, GO and KEGG pathways analyses reveal an over-representation of proteins annotated to MAPK signaling present in the infected cell phosphoproteome. Second, infected cells exhibit a MAPK/CDK kinase motif signature. Finally, a very high percentage of infection-specific host phosphoproteins contain strong CDK and/or ERK phospho-acceptor motifs. To directly test whether CDK and ERK activity control MHV68 replication, we evaluated the capacity of pharmacologic inhibitors of CDK and ERK activity to block MHV68 replication in single-step growth curves. Target cells were pretreated with the CDK inhibitor roscovitine [72], [73], MEK inhibitor U-0126 [74], or ERK inhibitor 5-iodotubercidin [75], [76] prior to infection with MHV68, and viral titers were evaluated by plaque assay at 24 h post-infection. Compared to vehicle and untreated control infections, both roscovitine and 5-iodotubercidin treatments led to greater than 60-fold reduction in output titers. However, U-0126, which acts on MEK kinases 1 and 2 upstream of ERK activation [74], only minimally affected virus production (Fig. 8A). This suggests that MHV68-induced ERK activation occurs independent of MEK1/2. These data provide biological evidence that MHV68 usurps host CDK and ERK kinases for productive replication.
While roscovitine is a highly specific CDK inhibitor [72], [73], 5-iodotubercidin is a more promiscuous kinase inhibitor also capable of inhibiting adenosine monophosphate kinase and haspin [77], [78]. Highly specific chemical inhibitors of ERK activity are not currently available [79]. To more definitively evaluate roles in MHV68 replication, we utilized shRNAs targeting either ERK1 or ERK2 to knockdown ERK expression in 3T3 fibroblasts. At the same time, we also knocked down c-Jun expression in an effort to establish a possible downstream target of CDKs and/or ERKs required for viral replication. Compared to control shRNA knockdown cells, all of the shRNAs tested influenced the efficiency of MHV68 replication on some level (Fig. 8B). Interestingly, shRNAs targeting ERK1 slightly delayed the onset of viral replication (see 24 h and 48 h timepoints), but did not significantly reduce titers at later timepoints. In contrast, knockdown of c-Jun and ERK2 expression led to an overall reduction in output titers from 48–96 h post-infection, approximating 10-fold less efficient viral yield by 96 h post-infection (Fig. 8B). It is notable, however, that inhibition of viral replication was not absolute, but rather manifested as a relative deficit over time. This result may be a consequence of incomplete knockdown of the targeted proteins, or it also is possible that c-Jun, ERK1, and/or ERK2 function to enhance the efficiency of MHV68 replication, but are not absolutely required. It is worth noting that each of 4 unique shRNA constructs targeting either ERK1 or ERK2 reduced the expression of both ERK isoforms recognized by the antiserum used to evaluate knockdown efficiency (not shown). We reason this effect stems from the presence of a long stretch of highly homologous nucleotide sequence present in both isoforms. Thus, as immunoblot data suggest, it is likely that both ERK isoforms were depleted in these experiments, which complicates direct functional interpretations for particular ERK isoforms. Nonetheless, in conjunction with pharmacologic inhibition data, these findings provide strong evidence in support of the bioinformatically-predicted hypothesis that CDK and MAPK signaling promote MHV68 replication.
While the bioinformatics approaches employed above facilitated the identification of host molecules involved in GHV replication, it was not yet clear whether viral signaling proteins also contributed to infection-associated changes in the cellular phosphoprotein network. MHV68 encodes two kinases (ORF21 and ORF36) and an ortholog of cellular D-type cyclins (v-cyclin) capable of stimulating CDK activity (ORF72) [30], [69]. We therefore evaluated the capacities of WT MHV68, and transposon mutant viruses in which ORF21, ORF36, and ORF72 had been disrupted [80] to elicit phosphorylation of ERK1/2, c-Jun, and MAPK/CDK-motif containing proteins during infection. In comparison to mock-infected cells, all of the viruses tested potently induced ERK1/2 and c-Jun phosphorylation (Fig. 9A). However, the ORF72 transposon mutant did not elicit robust phosphorylation of MAPK/CDK-motif-containing proteins, while the other viruses did (Fig. 9A), thus provisionally identifying v-cyclin as a viral molecule that contributes to the infection-associated phosphorylation signature. As a more discrete test of this hypothesis, we infected cells with a recombinant virus containing a targeted disruption of ORF72 (ORF72-null) and its genetically repaired WT control, ORF72-MR [81]. While ORF72-MR infected cells exhibited enhanced MAPK/CDK-motif phosphorylation relative to mock infection, the ORF72-null virus did not elicit robust MAPK/CDK-motif phosphorylation (Fig. 9B). Thus, these data indicate that v-cyclin is a pathogen-encoded molecule that plays a prominent role in directing infection-related phosphorylation events.
Data presented in this manuscript describe a first-of-its-kind global phosphoproteomic analysis of gammaherpesvirus infection. The results offer new insight into infection-driven alterations in cellular phosphoprotein networks and requirements for productive viral replication. Quantitative analyses suggest that the vast majority of detectable phosphopeptides are regulated during GHV infection, and the identification of more than 400 induced or repressed host phosphoproteins and 18 viral phosphoproteins substantially increases the knowledge base of proteins phosphorylated during GHV infection. Complementary biochemical, bioinformatic, and pharmacologic inhibition studies depict a virus-induced redirecting of host-phosphoprotein interaction networks and functional pathways, likely dictated by ERK and CDK host proteins and a virus-encoded D-type cyclin ortholog. Together, these data demonstrate the potential power of systems-level analyses to define critical signal transduction pathways usurped during intracellular pathogen replication.
Our study identified a total of 405 proteins that were only detected in either control (266) or infected cells (144 – 22 viral and 122 host). One of the proteins absent from infected cells, PML, is subject to ubiquitin-mediated degradation during MHV68 infection [36], [37]. This is also true for EBV, HSV and HCMV [82]. For HSV and HCMV, PML degradation limits an intrinsic host response to infection that represses viral gene expression [83], [84]. It is intriguing to speculate that other phosphoproteins not detected during infection also are degraded in order to limit inhibitory host-cell responses to MHV68 infection. Of course, the lack of detection in our phosphoproteomic analyses may also reflect a simple loss of phosphorylation, perhaps through phosphatase activity or viral inhibition of an upstream kinase. Indeed, phosphorylated FOXK1 was differentially detected between control and infected cells, yet expression of FOXK1 protein actually remains unchanged during MHV68 infection (J.A.S. and J.C.F., unpublished result). Likewise, infection-specific detection of ERK1/2 and c-Jun was a result of induced phosphorylation, rather than enhanced expression. Having established the framework here, in next generation experiments we envision pairing global differential proteomics techniques, such as stable-isotopic labeling of amino acids in cell culture (SILAC), with phosphoprotein enrichment to simultaneously determine changes in protein expression levels with induction or repression of phosphorylation. Kinetic analyses that combine these approaches would enable a dynamic assessment of how GHVs manipulate host protein expression levels and phosphorylation-dependent signaling events to gain control of the host cell. Moreover, such combined approaches would readily lend themselves to comparative studies aimed at determining specific contributions of viral signaling proteins, like v-cyclin, or host kinases, such as CDKs and ERKs.
The extent to which infection alters the phosphorylation status of specific proteins is remarkable and is dramatically illustrated in the GO protein class and global network analyses presented in Figures 2D and Figures 4 and 5 respectively. Indeed, 86% of proteins we identified exhibited intensity changes of greater than two-fold. By comparison, analogous phosphoproteomic analyses suggest that fewer than 15% of phosphoproteins are differentially regulated during cellular responses to DNA damage or growth factor receptor signaling [12], [13], 24% in response to Salmonella infection of cultured cells [35], and 14% induced by HIV binding to cells [85]. We hypothesize that these comparative differences in phosphoprotein regulation reflect the extent to which an intracellular pathogen must usurp multiple host cell biosynthetic systems and evade innate immune detection during viral replication. As an obligate intracellular parasite, a herpesvirus must commandeer host cell machinery involved in transcription, DNA replication, nuclear import and export, translation, and vesicle transport, while limiting or redirecting cell death, antiviral, and immunomodulatory host-cell responses to infection. The finding that infection alters the composition of protein clusters within the host phosphoprotein network may provide insight into mechanisms by which GHVs, and perhaps intracellular pathogens in general, gain control of functional modules within the host cell to facilitate viral replication. We hypothesize that GO analyses further illustrate this point, revealing how infection-related phosphorylation appears directed toward proteins involved in distinct biological processes, such as nucleosome organization and ribosome or rRNA regulatory processes (Table S2). One might also hypothesize that the absence of phosphoproteins in specific GO classes or KEGG pathways during MHV68 infection, such as those involved in ubiquitination, signal transduction, and cell death (Table S2), illuminate host processes shut down by virus-directed events.
Along these same lines it is important to consider the possibility that some of the signaling events we observed reflect the host cell response to infection. Roles for kinases in propagating and enforcing antiviral responses have been extensively studied [55], [86], [87], [88], and recent systems-level analyses demonstrate broad reorganization of host cell transcription and signaling networks following exposure to immuno-stimulatory microbial products [89]. From this study and numerous others, functions of MAPKs clearly influence the host cell response to infection. Given the strong MAPK signature present during MHV68 infection, it is possible that, beyond ERK, JNK [24], or Tpl2/Cot1 [26], other MAPKs that do not overtly facilitate viral replication influence the phosphoprotein network as part of the innate host-cell response to infection. Thus, it will be of interest to elucidate if and how other MAPKs or unrelated innate-immune kinases influence the infection-associated phosphoproteome.
A key feature of the MHV68 phosphoproteome is that it offers direct insight into specific host signaling pathways usurped by MHV68 to facilitate infection. The data also highlight several interesting parallels with other herpesviruses. The concurrence of several independent bioinformatics analyses in highlighting the prominence of ERK/MAPK and CDK-related phosphorylation during MHV68 infection was impressively emphasized by the detection of ERK/CDK motif phosphorylation on 55% of infection-specific host phosphoproteins (Table S3, compare to Table S1). Functional tests using pharmacologic inhibitors and shRNA knockdown confirmed the importance of CDK and ERK signaling in MHV68 replication (Fig. 8). With regard to ERK, a number of previous studies have demonstrated presumably biphasic roles for ERK in the KSHV lytic replication cycle. In the first phase, ras/raf-MEK-ERK signaling pathways are capable of promoting reactivation from latent infection by promoting immediate-early viral gene expression [38], [39], [40]. As one would expect given the involvement of MEK in this canonical pathway, this phase of ERK activation – and consequently KSHV reactivation downstream of ras/raf, MEK, or chemical induction with TPA – is inhibited by treatment with the pharmacologic MEK inhibitor U-0126 [38], [39], [40]. In contrast, a second phase of ERK activation during the KSHV lytic cycle is mediated by ORF45, a multifunctional tegument protein that stabilizes a ternary complex composed of ORF45, ERK, and p90 ribosomal S6 kinase [43], [90]. In agreement with the finding that MHV68 replication (Fig. 8) and ERK activation (not shown) were not inhibited by U-0126 treatment, ORF45-directed ERK activation also is insensitive to MEK inhibition [43], leading us to hypothesize that MHV68 ORF45 may similarly promote ERK activation. Indeed, MHV68 and KSHV ORF45 proteins are functionally interchangeable in facilitating viral replication [91]. Hence, in addition to enhancing a general understanding of ERK functions in GHV infection, the identification of previously unknown putative ERK-phosphorylated proteins within infected cells may provide new insight pertaining to ORF45-directed enhancement of viral gene expression, translation, and viral egress [92], [93], [94]. It will also be of interest to determine if predicted ERK phosphorylation sites we identified on ORF45 (Table 1 and Table S3) are bona fide ERK targets, and whether they exert functional control over ORF45 complex formation during viral replication.
As a group, herpesviruses are thought to usurp CDK signaling in order to provide an S-phase-like environment amenable to replicating the viral DNA genome. For instance, reactivating EBV drives high S-phase cyclin expression and CDK activity, while at the same time inhibiting host DNA replication [95], possibly through induction of a DNA damage response (DDR) [96] and/or inactivation of the MCM4-6-7 helicase complex [97]. Further and in agreement with our data, pharmacologic inhibition of CDK activity with roscovitine also inhibits lytic replication of EBV [98], HCMV [99], [100], and HSV1 [101], [102], which strongly suggests that usurping cyclin/CDK activity is a universal requirement of herpesviruses. However, roles for CDK activity in promoting GHV replication have been minimally explored. And, although a few EBV targets of cyclin B/CDK1 are known [103], information as to host substrates of cyclin/CDK activity during lytic GHV replication are lacking. In this regard, identification of potential CDK-phosphorylated proteins during MHV68 infection may reveal how GHVs direct CDK activity to foster efficient viral replication.
A related question asks which viral factors contribute to the CDK phosphorylation signature during lytic GHV replication. As a homeostatic cellular process, cyclin/CDK activity is tightly controlled on several levels. This includes transcriptional regulation of cyclin genes, phosphorylation-dependent activation and inactivation of CDKs, and direct inhibitory interactions of cyclin/CDK holoenzyme complexes with CDK inhibitors (CKIs), such as p21 and p27 [67]. At the host transcriptional level, both MHV68 and KSHV LANA proteins induce transcription of cellular cyclin genes [104], [105]. Although the functional significance of LANA-mediated induction of host cyclins during productive gamma-2-herpesvirus replication has not been specifically tested, it is interesting to note that LANA-null MHV68 exhibits attenuated replication both in culture and in vivo [106], [107] that is dependent on its transcriptional regulatory capacity [108]. Additionally, conserved herpesvirus protein kinases (CHPKs) in GHVs and beta-herpesviruses, which are required for efficient viral replication [11], [27], [109], [110], exhibit CDK-like functions, most notably pRb phosphorylation and the capacity to complement temperature sensitive yeast CDC28 (S. cerevisiae CDK ortholog) mutants for growth [111], [112]. Further, BGLF4, the EBV CHPK, exhibits partially overlapping substrate specificity with cyclin B/CDK1 in vitro [103].
Finally, gamma-2-herpesviruses, including MHV68 and KSHV, encode an ortholog of cellular D-type cyclins [29], [30]. Viral cyclins exhibit an expanded capacity to interact with host CDKs [69], [113], [114] and are resistant to inhibition by CKIs [115]. Indicative of their capacity to stimulate cell-cycle progression [116], [117], v-cyclins are oncogenic when expressed in mice as a transgene [116], [118], [119], and MHV68 v-cyclin is singularly required for pRb phosphorylation during lytic MHV68 infection [69]. Further, the KSHV cyclin ortholog is thought to play initiating and sustaining roles in KSHV-related cellular transformation [33], [120], [121]. The finding that cells infected with v-cyclin-null MHV68 exhibit reduced MAPK/CDK phosphorylation (Fig. 9) strongly suggests that v-cyclin is a major contributor to the MAPK/CDK signature of lytic MHV68 infection. Although v-cyclin is not absolutely required for viral replication in cell culture, v-cyclin-null MHV68 exhibits attenuated acute replication, delayed latency establishment, and a severe reactivation defect in vivo [81], [122]. An elegant study using recombinant MHV68 viruses in which v-cyclin was exchanged with cellular cyclins A, D, or E demonstrates overlapping or redundant functions for host and v-cyclins in some, but not all, aspects of MHV68 pathogenesis [123], which may explain why v-cyclin is not necessary for MHV68 replication in culture [81], but roscovitine potently blocks viral replication (Figs. 8 and 9). Moreover, v-cyclin expression is necessary for MHV68 transformation of primary B cells in culture [124] and lymphoproliferative disease and lethal pneumonia in vivo [125], [126]. While further validation clearly is necessary, it is tempting to speculate that the CDK-motif containing proteins identified in this report are critical host-cell targets of v-cyclin that influence GHV pathogenesis.
Together, the data presented herein enhance our understanding of the GHV-host interaction. In addition to defining new proteins and hypotheses for experiments to foster a more complete understanding of basic mechanisms of GHV replication and pathogenesis, the identified ERK and CDK-predicted phosphoproteins may encompass new host targets for therapeutic interventions. Our data strongly support the further evaluation of ERK and CDK inhibitors as treatments for lytic cycle-associated GHV diseases, such as IM or KS. Toward defining the pathogen-host interaction in general, it will also be of interest to determine whether global reorganization of the host phosphoprotein network is a phenotype shared with unrelated intracellular pathogens, such as RNA viruses or bacteria. If so, are common signaling pathways or macromolecular complexes targeted? And, could these common pathogen-exploited host proteins serve as novel candidates for new generalized treatments? The approaches we describe should be readily adaptable to other systems. Thus, our studies lay the foundation for future comparative analyses of this sort, as well as for defining differences and commonalities between de novo lytic GHV replication and reactivation, or comparative studies with alpha and beta-herpesviruses.
Mouse experiments performed for this study were carried out in accordance with NIH, USDA, and UAMS Division of Laboratory Animal Medicine and IACUC guidelines. The protocol supporting this study was approved by the UAMS Institutional Animal Care and Use Committee (Animal Use Protocol 3270). Mice were anesthetized prior to inoculations and sacrifice to minimize pain and distress.
Swiss-albino 3T3 fibroblasts (referred to as 3T3 fibroblasts throughout) were purchased from ATCC. All cells were cultured in Dulbecco's modified eagle medium supplemented with 10% fetal calf serum (FCS), 100 units/ml penicillin, 100 µg/ml streptomycin, and 2 mM L-glutamine (cMEM). Serum starvation involved culturing of cells in media containing 0.5–1% FCS for 18–24 hours prior to infection or treatment. Cells were cultured at 37°C with 5% CO2 and ∼99% humidity. Wild-type MHV68 was strain WUMS (ATCC VR1465), WT BAC-derived MHV68 [127], or BAC-derived MHV68-YFP [128]. ORF50-null MHV68 (50.STOP) was previously described [52]. UV-inactivation of WT MHV68 was accomplished by diluting virus stock to 1×107 PFU/ml and autocrosslinking in 60 mm plates using a Stratalinker prior to infection. Disruption of viral gene expression was confirmed by immunoblot analyses to detect viral proteins. Cells were infected by low-volume adsorption of viruses to the monolayer. The time of adsorption for all experiments was considered t = 0. Inocula were removed after 1 h, and cells were cultured in a normal volume of serum starvation medium.
1×107 mock-infected or infected cells were harvested 18 h post-infection by scraping in cold phosphate-buffered saline (PBS). Cells were pelleted at 700 g for 5 min, and snap frozen in liquid N2. Cell pellets were lysed on ice in 500 µL buffer containing 50 mM Tris (pH 7.5), 50 mM NaCl, 0.05% surfactant (Promega) supplemented with protease and phosphatase inhibitor cocktails (Thermo Scientific) with vortexing every 5 to 10 min for 30 min. Insoluble debris was removed by centrifugation at 11000 g for 11 min. Protein concentration in the resulting solution was determined by BCA assay. 1 mg of protein in solution was concentrated by centrifugation through a 3 kDa filter (Amicon) and rinsed with 500 µL buffer containing 0.025% surfactant and 25 mM ammonium bicarbonate (ABC). The concentrated protein mixture was then diluted to 900 µL total volume in 25 mM ABC. Proteins were reduced in 5 mM DTT for 20 min at 60°C, followed by alkylating in the dark with 25 mM iodoacetamide for 30 min at 25°C. Buffer exchange to 25 mM ABC was performed by centrifugation through 3 kDa filters and the resulting concentrate was diluted to 900 µL total volume in 25 mM ABC. Trypsin diluted in 0.01% trifluoroacetic acid (TFA) was added to the protein mixture (1∶50 w/w) and incubated overnight at 37°C. Digests were quenched with 0.1% TFA. A 20 µL aliquot of the quenched trypsin digest was set aside for analysis.
Phosphopeptide enrichment was based on a previously described method [129]. Digested peptide samples (from 1 mg total protein) were desalted using Sep-Pak columns. Sep-Pak columns were primed with a 75/25 mixture of buffers B/A (Buffer A - 2% acetonitrile (ACN), 0.1% formic acid; Buffer B - 75% ACN, 0.1% formic acid) and rinsed with 2 mL Buffer A. Peptide samples were passed through Sep-Pak columns, followed by rinsing with 2 ml Buffer A. Peptide were eluted with 75/25 Buffer B/A mixture and desiccated in a speed vac. TiO2 beads were pre-incubated in Loading Buffer 1 (LB1 – 65% ACN, 2% TFA, saturated glutamic acid) at a ratio of 1 mg beads to 20 µl LB1. 10 µl of TiO2 bead slurry was added to each desalted peptide sample and agitated for 10 min. Beads were collected by centrifugation at 3000 rpm for 30 sec, and the enrichment was repeated twice more with a fresh aliquot of TiO2 beads for each peptide solution. Thus, three successive enrichments were performed for each sample. Beads and phosphopeptides were washed three times for 10 min with agitation using 800 µl Wash Buffer 1 (65% ACN, 0.1% TFA), followed by 3 identical washes with 800 µl Wash Buffer 2 (65% ACN, 0.5% TFA). Phosphopeptides were eluted by incubation in Elution Buffer 1 (300 mM NH4OH, 50% ACN) for 10 min with agitation, followed by identical treatment with Elution Buffer 2 (500 mM NH4OH, 60% ACN). Eluted phosphopeptides were desiccated in a speed vac.
Peptide samples acidified to 0.1% formic acid final concentration were analyzed by nano-LC/MS/MS technique on an ion trap tandem mass spectrometer (MS). An auto-sampler was used for automatic injection of tryptic peptides from a 96 well plate to the NanoLC 2D system (Eksigent). Peptides were separated by reverse phase HPLC using a 10 cm long analytical column (C12 resin, Phenomenex). HPLC eluate was ionized by ESI (Electrospray ionization), followed by MS/MS analysis using collision induced dissociation on an LTQ Orbitrap hybrid MS (Thermo Finnigan, San Jose, CA) with two mass analyzers - Linear ion trap (LTQ), and Orbitrap. One MS scan by Orbitrap was followed by 7 MS/MS scans by LTQ. Other relevant parameters include - spray voltage 2.0 kV; m/z range of 350–1500; isolation width (m/z) of 2.5; and normalized collision energy 35%. MS spectrum data were acquired using XCalibur 2.0 software. MS technical information provided in Table S4. Raw data files are available at https://chorusproject.org/anonymous/download/experiment/-7729244682105805562.
Data analysis was performed using MaxQuant 1.0.12.31 [130]. Experiment design consisted of two sample types, ‘mock’ (Expt1) and ‘infected’ (Expt2). Each sample type had two biological duplicates (A and B) and three technical replicates (1, 2 and 3). Therefore we had 12 MS data files, one for each MS run, namely: M1A, M1B, M2A, M2B, M3A, and M3B for mock samples and corresponding Mr1A, Mr1B, Mr2A, Mr2B, Mr3A, and Mr3B for infected samples. MS/MS peaks were searched against a concatenated forward and reversed version of IPI_mouse_v3.82 [131] database using the Mascot 2.2 search engine [132] via MaxQuant. False discovery rate for identification was less than 1% as estimated by the number of hits to the reversed sequences in the decoy database. Additional technical parameters are provided in Table S4. Thus, we identified a total of 986 proteins at 1% FDR. These included 791 phosphoproteins with 1101/2271 unique phospho (ST) and 38/39 unique phospho (Y) site positions.
Proteins differentially enriched between mock and infected samples were identified by (1) present-absent call based on peptide intensity (zero intensity was considered as absent call) and (2) 1.5-fold increase or decrease in intensity ratio of infected/mock. This analysis was performed using the protein intensities from the MaxQuant output file proteinGroups.txt. The complete quantitated data set is provided in Table S1.
Version 9.05 of the STRING resource [59] was used to generate protein interaction networks for MS-identified proteins. STRING networks are provided in Figures S4 and S5. All interactions are predicted with medium confidence threshold of 0.400, and all active predictive methods were allowed. Interaction networks were processed in Cytoscape 2.8 [60] to assign integer values and color coding to visually depict presence, absence, increase or decrease in protein intensity during infection. Disconnected nodes are not included in the Cytoscape output. Biological Process gene ontology analyses were performed using the BiNGO cytoscape plugin [63]. Overrepresented categories were identified relative to the Mus musculus background gene set using a hypergeometric test with Benjamini and Hochberg false discovery rate correction to define significance. Sorted data provided in Table S2. Clustered proteins in phosphoprotein networks were identified using the MCODE cytoscape plugin [62]. Protein Class gene ontology analyses were performed using PANTHER 7.2 [133] against the Mus musculus background gene set. Kyoto encyclopedia of genes and genomes (KEGG) pathways enrichment was defined through DAVID [134], [135]. For Motif-X [65] analyses, unique phosphopeptides for either mock or infected samples were identified from the global phosphopeptide sequence list. Unique phosphopeptides for either data set were identified using the IPI mouse proteome as background with a minimum of 20 occurrences per motif and a significance threshold of 0.000001. 13 amino acid long motifs were defined where the phosphorylated residue is at position 7. Shorter peptides were extended from mouse IPI database. Prealigned Motif-X output text files were used to generate global unique sequence logos in ICE-LOGO (http://iomics.ugent.be/icelogoserver/logo.html). Group-based Prediction Systems (GPS) 2.1 software [71] was utilized at highest-threshold setting to perform batch identifications of phosphopeptides containing specific kinase target motifs. Sorted data provided in Table S3.
Serum-starved 3T3 fibroblasts were untreated or pretreated with either DMSO (vehicle), U0126 (LC Laboratories), roscovitine (Cayman Chemical), or 5-iodotubercidin (Cayman Chemical) at 10 µM final concentration for 1 h prior to low-volume adsorption with MHV68 at MOI = 5 PFU/cell. Inocula were removed, and cells were cultured in a normal volume of medium. Cells were harvested 24 h post-infection, and progeny virions were liberated by freeze-thaw lysis. Viral titers were determined by MHV68 plaque assay as described [136]. A separate plate was harvested immediately after adsorption and subsequently titered to ensure that drug treatment did not inhibit viral attachment and to determine the 0 h titers for viral yield calculations.
Lentiviral pLKO.1-based shRNA vectors were purchased from Sigma. The following shRNA constructs were used in this study: TRCN0000229528 (c-Jun-1, NM_010591.2-2974s21c1), TRCN0000042693 (c-Jun-2, NM_010591.1-994s1c1), TRCN0000360511 (c-Jun-3, NM_010591.2-2270s21c1), TRCN0000023160 (MAPK1-1, NM_011949.2-490s1c1), TRCN0000054730 (MAPK1-2, NM_011949.2-921s1c1), TRCN0000023186 (MAPK3-1, NM_011952.1-305s1c1), TRCN0000023187 (MAPK3-2, NM_011952.1-662s1c1). Lentiviruses were produced by transfecting 293T cells with shRNA vector plasmid and packaging vectors pSPAX2 and pHCMV-G. Lentiviral supernatants were harvested at 48 and 72 hours post-transfection. Due to inefficient knockdown using single vectors, 3T3 fibroblasts were transduced in 24 hour succession with two distinct lentiviruses each targeting the specified protein. c-Jun (1) stable knockdown cells were transduced with shRNA vectors 1 and 3. c-Jun (2) stable knockdown cells were transduced with shRNA vectors 2 and 3. Transduced cells were selected with puromycin (4 µg/ml) and expanded. Stable knockdown cells were plated and infected with MHV68 at MOI = 0.05 PFU/cell. Cells were harvested at the indicated times and viral titers were determined by plaque assay [136]. Viral yields were determined by dividing output titers at the indicated timepoint by 0 h titers which represent input virus inoculum.
Cells were lysed with alternative RIPA buffer (150 mM NaCl, 20 mM Tris, 2 mM EDTA, 1% NP-40, 0.25% DOC, supplemented with complete mini-EDTA free protease inhibitors (Thermo) and phosphatase inhibitor cocktail 2 (Thermo) and quantified using the Bio-Rad DC or Thermo BCA protein assay prior to resuspending in Laemmli sample buffer, or equivalent numbers of cells (1–2×105) were directly lysed with 100 µl Laemmli sample buffer. Samples were heated to 100°C for 10 min and resolved by SDS-PAGE. Resolved proteins were transferred to nitrocellulose and identified with the indicated antibodies. ORF59, v-cyclin, and MHV68 antisera were previously described [69], [116]. ORF21 mAb was a gift from P.G. Stevenson. MHV68 antiserum was generated as previously describe [137]. P-Ser (AB1603), p-Thr (05-1923), and p-Tyr (05-321) antibodies were purchased from Millipore. P-c-Jun (S73-#9164), c-Jun (#9615), p-ERK1/2 (#4370), ERK1/2 (#4695), p-MARCKS (#2741), and S*PXK/PXS*P motif (#2325) RXXS/T* motif (#9614) antibodies were purchased from Cell Signaling Technology. β-actin mouse monoclonal antibody was purchased from Sigma (A2228). Immobilized antigen and antibody were detected with HRP-conjugated secondary antibodies and SuperSignal Pico West ECL reagent (Thermo Scientific) or Clarity ECL reagent (BioRad) and exposed to film or imaged on a BioRad ChemiDoc MP digital imaging system.
Female C57BL/6 mice 6 to 8 weeks of age were purchased from the Jackson Laboratory. Mice were sterile housed in the animal facility at the University of Arkansas for Medical Sciences in accordance with all federal and university DLAM guidelines. Mice were mock-infected with intraperitoneal injection of 0.2 ml of DMEM or infected intraperitoneally with 106 PFU of H2B-YFP virus diluted into 0.2 ml of DMEM. Four days post-infection mice were sacrificed by isoflurane overexposure and cervical dislocation. Spleens were harvested, homogenized into single-cell suspensions, and erythrocytes were lysed with red blood cell lysis buffer (Sigma) according to manufacturers instructions.
For flow cytometry, splenocytes were fixed and permeabilized with Foxp3/Transcription Factor Staining Buffer Set according to the manufacturers instructions (eBioscience, #00-5523-00). Fixed and permeabilized cells were washed twice with FACS buffer (3 µM BSA, 1 mM EDTA in PBS) and stained with PE-conjugated rabbit anti-c-Jun pS73 (Cell Signaling, #8752) and goat anti-GFP (Rockland, #600-101-215) antibodies diluted in FoxP3 wash buffer (eBioscience) for 30 minutes at room temperature. Stained cells were washed twice with FoxP3 wash buffer and incubated with donkey anti-goat secondary antibody conjugated to Alexa fluor 488 nm (Invitrogen, #A-11055) diluted in FoxP3 wash buffer for 30 minutes at room temperature. Stained splenocytes were washed twice with FoxP3 wash buffer and resuspended in FACS buffer. Antibody-stained cells were analyzed by flow cytometry using a Fortessa (Becton Dickinson) to quantify cellular YFP and c-Jun (p-S73) levels. Splenocytes from mock-infected animals were used to gate for YFP− and YFP+ cell populations, as infected (YFP+) splenocytes are absent from these animals. This gating strategy allowed for detection of YFP+ splenocytes, which expressed YFP at levels exceeding the prior gate.
|
10.1371/journal.pgen.1005727 | Skp1 Independent Function of Cdc53/Cul1 in F-box Protein Homeostasis | Abundance of substrate receptor subunits of Cullin-RING ubiquitin ligases (CRLs) is tightly controlled to maintain the full repertoire of CRLs. Unbalanced levels can lead to sequestration of CRL core components by a few overabundant substrate receptors. Numerous diseases, including cancer, have been associated with misregulation of substrate receptor components, particularly for the largest class of CRLs, the SCF ligases. One relevant mechanism that controls abundance of their substrate receptors, the F-box proteins, is autocatalytic ubiquitylation by intact SCF complex followed by proteasome-mediated degradation. Here we describe an additional pathway for regulation of F-box proteins on the example of yeast Met30. This ubiquitylation and degradation pathway acts on Met30 that is dissociated from Skp1. Unexpectedly, this pathway required the cullin component Cdc53/Cul1 but was independent of the other central SCF component Skp1. We demonstrated that this non-canonical degradation pathway is critical for chromosome stability and effective defense against heavy metal stress. More importantly, our results assign important biological functions to a sub-complex of cullin-RING ligases that comprises Cdc53/Rbx1/Cdc34, but is independent of Skp1.
| Protein ubiquitylation is the covalent attachment of the small protein ubiquitin onto other proteins and is a key regulatory pathway for most biological processes. The central components of the ubiquitylation process are the E3 ligases, which recognize substrate proteins. The best-studied E3 complexes are the SCF ligases, which are composed of 3 core components—Cdc53, Skp1, Rbx1—that assemble to the functional ligase complex by binding to one of the multiple substrate adaptors—the F-box proteins. Maintaining a balanced repertoire of diverse SCF complexes that represent the entire cellular panel of substrate adapters is challenging. Depending on the cell type, hundreds of different F-box proteins can compete for the single binding site on the common SCF core complex. Rapid degradation of F-box proteins helps in maintaining a critical level of unoccupied Cdc53/Skp1/Rbx1 core, complexes and alterations in levels of F-box proteins has been linked to diseases including cancer. Studying the yeast F-box protein Met30 as a model, we have uncovered a novel mechanism for degradation of F-box proteins. This pathway targets free F-box proteins and requires part of the SCF core. These findings add an additional layer to our understanding of regulation of multisubunit E3 ligase.
| Ubiquitin dependent proteolysis controls many cellular processes including signal transduction and cell cycle progression. Ubiquitin is covalently linked to substrates in a multistep process that requires coordinated action of 3 classes of enzymes- E1 ubiquitin activating enzyme, E2 ubiquitin conjugating enzyme, and E3 ubiquitin ligase [1–5]. E3 ubiquitin ligases are the key players in this system as they mediate substrate specific covalent attachment of ubiquitin. Within the E3 ligase family, cullin-RING ligases (CRLs) comprise the largest class, and in this group the SCF ubiquitin ligases are one of the best-understood complexes [2,6]. They are composed of yeast Cdc53 (mammalian cullin-1), Skp1, Rbx1, and one of the multiple F-box proteins, which bind substrates and confer specificity to the complex [7,8].
Amongst the SCF components, F-box proteins are relatively unstable in nature, which contributes to the dynamic assembly of a diverse repertoire of SCF complexes within the cell [9–13]. Accordingly, over expression of a single F-box protein in yeast can change the balanced distribution and diversity of available SCF complexes by sequestering cullin and Skp1 and thus block formation of functional SCF complexes with other F-box proteins [10,11,14]. Many F-box proteins control degradation of critical oncogenes and tumor suppressors and variation in their abundance has been linked to cancer [15,16]. Thus, it is important to understand how cells maintain F-box protein homeostasis. F-box proteins are known to be regulated by autoubiquitylation where their degradation is dependent upon their incorporation into a functional SCF complex [10,11]. The autocatalytic F-box protein degradation pathway is thought to be suppressed by substrate binding resulting in coordination of substrate availability with abundance of the corresponding assembled SCF complex [17,18]. Additional degradation pathways for F-box proteins are likely as it is also important to restrict abundance of unbound F-box proteins to prevent substrate shielding effects that would compete with substrate recognition by fully assembled ligases. Indeed, mammalian Skp2 is targeted for degradation by the anaphase promoting complex or cyclosome [19,20], Fbx5 (Emi1) is degraded by SCFβTrCP/Slimb [21], and the level of the budding yeast F-box protein Dia2, which is required for genomic stability, is restricted by the HECT domain E3 ligase Tom1 [22].
In Saccharomyces cerevisiae, Met30 is one of three essential F-box proteins. SCFMet30 coordinates metabolic pathways of sulfur containing compounds with cell cycle progression. The transcription factor Met4 is a key target of SCFMet30 [23,24]. Low levels of the methyl donor, S-adenosylmethionine cause a block in SCFMet30 dependent ubiquitylation of Met4, activating it and results in cell cycle arrest and transcription of methionine response genes [25]. SCFMet30 also represses expression of enzymes responsible for glutathione synthesis and is thus a key factor in response to heavy metal stress. Cadmium exposure induces active dissociation of Met30 from Skp1, thereby inhibiting SCFMet30 and inducing glutathione production and cell cycle arrest, which together protect cellular integrity [26–28]. Therefore, the SCFMet30 system generates unbound Met30 via heavy metal stress induced dissociation from Skp1. To prevent unwanted effects of excess unbound Met30, such as substrate sequestration, it seemed important, that a degradation mechanism exists in addition to autocatalysis that plays an integral role in maintaining Met30 homeostasis.
Here we report such an additional mechanism for Met30 regulation in addition to autoubiquitylation. This degradation pathway targets Met30 that is detached from Skp1 and involves the SCF core components Cdc53 and Rbx1, as well as the cognate SCF ubiquitin conjugating enzyme Cdc34. Importantly, this ubiquitylation pathway does not require Skp1 (Skp1-independent Cul1-dependent ubiquitylation) and suggests a function of Cdc53/Cul1 that is independent from association with Skp1.
Defense against heavy metal toxicity requires coordinated changes of metabolic pathway flux connected to glutathione synthesis, induction of a cell cycle checkpoint response to avoid continued cell division during stress conditions, and cell protective measures known as sulfur sparing [25]. SCFMet30 is the key regulator of this concerted response [24,26]. Coordinated regulation is achieved because cadmium stress disrupts ubiquitylation of all SCFMet30 substrates by the active and selective disassembly of the SCFMet30 protein complex, but not other SCF ligases [24,26,27]. This is accomplished by Cdc48-mediated dissociation of the F-box component Met30 from Skp1 [24,26,27], which results in generation of a pool of unbound Met30. In addition, Met30 transcription is induced by over seven fold during cadmium stress [29], which further increases the amount of Met30 that is not associated with core SCF components (referred to as unbound or ‘Skp1-free’ Met30). We were interested in how cells cope with the generated excess of Met30 because the well-described autocatalytic F-box protein degradation pathway mechanism [10,11] cannot act on Met30 during these conditions. Interestingly, degradation of the F-box protein Met30 was maintained and even slightly induced in response to cadmium stress (Fig 1A). We next asked whether cadmium stress induces a degradation pathway for unbound Met30, or if dissociation of Met30 from Skp1 might be sufficient to induce its degradation. To this end we examined stability of Met30 that cannot associate with Skp1 even in the absence of cadmium due to mutations in the F-box motif, which forms the Skp1 interaction surface. A mutant form of Met30 lacking its entire F-box domain (Met30ΔFbox) was constitutively unstable even though cells did not experience cadmium stress (Fig 1B). Rapid degradation of Met30ΔFbox was observed both in cycloheximide chase and promoter shut-off experiments, indicating that components of this degradation pathway are constitutively present and that new protein synthesis is not required (S1A Fig). Disruption of the Met30-Skp1 interaction by the single amino acid change L187D [30], rather than complete deletion of the F-box domain, was sufficient to induce Met30 degradation (Figs 1C and S1B) suggesting that Met30 that is not bound to Skp1 is degraded.
To further test this idea and exclude that mutating the F-box region results in conformational changes that target Met30 to a non-physiological degradation pathway, we measured Met30 stability in another condition where Met30 is dissociated from Skp1. To this end we used a yeast strain carrying the temperature sensitive skp1-25 allele, which is inactivated by a temperature shift and disrupts the integrity of SCF ligases at the restrictive temperature [31]. Met30 was efficiently degraded in skp1-25 mutants and cadmium exposure did not further destabilize Met30 (S1C Fig).
Collectively these results demonstrate that disruption of the Met30-Skp1 interaction, by either active signal-induced dissociation (cadmium stress) or by mutations in Met30 or Skp1, induces rapid degradation of the F-box protein Met30. These results suggest a proteolytic pathway that recognizes unbound ‘Skp1-free’ Met30 to avoid accumulation of excess unbound Met30, which could bind substrates and shield them from recognition by fully assembled SCFMet30 complexes.
We sought to further characterize proteolysis of ‘Skp1-free’ or unbound Met30 and asked whether it was dependent upon the ubiquitin-proteasome system. In wild-type cells Met30 may exist either bound or unbound to the SCF core. In order to specifically study the degradation pathway targeting ‘Skp1-free’ Met30, we used Met30ΔFbox as a tool to generate a homogenous population of Met30 free from Skp1. Inhibition of proteasome activity with MG-132 led to stabilization of Met30ΔFbox, suggesting that the ubiquitin proteasome pathway is involved in degradation of ‘Skp1-free’ Met30 (Fig 2A). We next asked what E2 ubiquitin conjugating enzyme might be involved in this degradation process. Surprisingly, degradation of Met30ΔFbox was dependent on the canonical SCF E2, Cdc34 (Fig 2B). The requirement for Cdc34 compelled us to test dependence on SCF core-components even though our previous experiments demonstrated that Skp1 is not involved in this degradation pathway. Unexpectedly, Cdc53 was indispensable for Met30ΔFbox degradation, because inactivation of the temperature sensitive cdc53-1 allele blocked degradation (Fig 2B). Consistent with these results Met30ΔFbox degradation was greatly reduced in a strain expressing a 13Myc-tagged version of the RING finger component Rbx1, which has previously been shown to be a hypomorph allele that results in reduced SCF function at high temperature [32] (Fig 2B). These experiments not only suggest a degradation pathway for Met30 that is independent of the canonical autoubiquitylation mechanism, but more surprisingly, indicate that the cullin-1 (Cdc53) based ligase complex might have functions independent of its adaptor component Skp1. Given these unexpected results we wanted to further test this hypothesis and ensure that skp1-25 and cdc53-1 mutants probe the same degradation pathway and do not induce any secondary effects that might affect interpretation of the degradation results. We thus tested Met30ΔFbox half-life in skp1 single and skp1 cdc53 double mutants. As previously observed, unbound Met30 was rapidly degraded in skp1-25 mutants but, importantly, was stabilized upon inactivation of cdc53-1 in the double mutant (Fig 2C), confirming that degradation was dependent on Cdc53 but not Skp1. In agreement with the Met30ΔFbox degradation data, ubiquitylated Met30ΔFbox was readily detectable in skp1 mutants but absent in cdc53 mutants further supporting the hypothesis of a Skp1- independent degradation function for Cdc53 (Fig 2D).
In addition of being a SCFMet30 substrate itself, the transcription factor Met4 has been shown to function as a substrate receptor in the context of SCFMet30/Met4 to coordinate degradation of its own co-factors [33]. Analysis of Met30ΔFbox degradation in met4Δ mutants showed that it was not involved in degradation of ‘Skp1-free’ Met30 (S2A Fig).
Cullins are regulated by covalent modification with ubiquitin like protein Nedd8 (Rub1 in yeast), which induces a conformational rearrangement of cullin and stimulates ubiquitin transfer by the SCF-bound E2 [6]. Deneddylated cullins interact with Cand1 (Lag2 in yeast), which inhibits binding of Skp1- F-box protein complex and prevents SCF ligase function [12,34,35]. As rubylation is dispensable in yeast [36], we posited that Lag2 bound to Cdc53 may serve as an adaptor for regulating F-box proteins dissociated from Skp1. Deletion of LAG2 failed to stabilize Met30ΔFbox demonstrating that degradation of ‘Skp1-free’ Met30 is independent of Lag2 (S2B Fig).
Together these results suggest existence of a novel mechanism of Met30 regulation that specifically targets Met30 that is displaced from Skp1 and is dependent on the ubiquitin proteasome system and requires the function of Cdc53, Rbx1 and Cdc34, but not Skp1.
Temperature sensitive skp1 alleles have been previously shown to differentially affect SCF ligases depending on the identity of F-box protein subunits [10,37]. Although the skp1-25 allele was specifically selected to represent a complete loss of Skp1 function, including inactivation of SCFCdc4 and SCFMet30, we could not unambiguously exclude that Skp1-25 forms an intact SCF ligase that could ubiquitylate Met30ΔFbox in trans. To address this issue we employed the temperature inducible degron tag (td) strategy to deplete Skp1 protein from cells [38], rather than rely on inactivation of temperature sensitive alleles. A skp1-td strain was constructed by expressing Skp1 fused to the temperature inducible degron under control of the inducible CUP1 promoter. Attenuation of Skp1 induced the expected biological response such as cell cycle arrest with elongated multibudded cell morphology indicative of SCFCdc4 inactivation, and block of SCFMet30 function as assayed by loss of ubiquitylated forms of Met4 (Fig 3A and 3B). Combined CUP1 promoter repression and temperature induced Skp1-td thus efficiently ablated Skp1 function and protein level (Fig 3). Consistent with results using temperature sensitive alleles of skp1, Met30ΔFbox degradation was unaffected when Skp1 function was blocked using the skp1-td strategy (Fig 3C). These results strongly support our hypothesis of Skp1-independent degradation of Met30.
The experiments with skp1 single and skp1 cdc53 double mutants strongly suggested that the ‘Skp1-free’ degradation pathway for Met30 was dependent on Cdc53 and independent of Skp1. We reasoned that in such a scenario, a Cdc53 mutant defective in interacting with Skp1 should be capable to degrade Met30ΔFbox. To examine this, a GAL1 inducible Cdc53Y133R mutant was constructed. Mutation of tyrosine in position 133 to arginine has previously been suggested to disrupt the Cdc53-Skp1 interaction [39]. Immunopurification experiments confirmed that Cdc53Y133R does not interact with Skp1 in vivo (Fig 4A). Accordingly, expression of Cdc53Y133R was unable to rescue the cdc53-1 growth arrest phenotype at restrictive temperature (Fig 4B). Importantly, congruous with our hypothesis, Cdc53Y133R supported degradation of Met30ΔFbox to the same extent as wild type Cdc53 (Fig 4C). These results support the hypothesis of a cullin-1 (Cdc53) function in protein degradation independent of its adaptor component Skp1.
Instability of Met30 in skp1 mutants led us to hypothesize that the ubiquitin ligase responsible for degrading ‘Skp1-free’ Met30 may recognize a domain close to the F-box domain, which is directly or indirectly obstructed by Skp1 binding. In accordance with this idea, a so-called R-motif has been described in the yeast F-box protein Cdc4, which is adjacent to its F-box domain. In addition, F-box—Skp1 interaction acts to suppress R-motif mediated Cdc4 degradation [14]. Thus, to identify the degradation sequences (degron) in Met30 recognized by the ‘Skp1-free’ degradation pathway, we generated deletions in Met30ΔFbox near the F-box domain and compared their stability to that of Met30ΔFbox.
Congruous with our hypothesis, deletion of 100 amino acids proximal and distal to the F-box domain (amino acids 137–277) not only increased steady-state Met30ΔFbox levels (time point 0) but also completely prevented its degradation (Fig 5A). To further narrow down the degron sequence, shorter deletions encompassing either the C terminal or the N terminal 50 amino acids were constructed. Deletion of the N terminal amino acids adjacent to the F-box motif (amino acids 137–187) was sufficient for Met30ΔFbox stabilization (Fig 5A). Interestingly, this region overlapped with a 45 amino acid stretch immediately N-terminal to the F-box domain, which mediates dimerization of WD-40 repeats containing F-box proteins and is conserved from yeast to humans [40]. Overlap of the degron region and the dimerization domain raised the possibility that the ligase for ‘Skp1-free’ Met30 degradation recognized the dimerization motif in Met30. However, mutation of isoleucine 159 and leucine 160, two amino acids crucial for Met30 dimerization [40], in Met30ΔFbox failed to stabilize the protein (S3A Fig), suggesting that different residues within this stretch were being recognized by the ligase.
Smaller deletions within the N terminal 50 amino acids contiguous to the F-box domain suggested that amino acids 170–187 in Met30, corresponding to a region rich in hydrophobic residues, were important for ligase binding because various Met30ΔFbox mutants containing this region were efficiently degraded and mutants lacking this region were stabilized (Figs 5A and S3B). To identify key residues in the degron, we mutated methionine 178 and isoleucine 179, a hydrophobic patch close to the F-box domain and conserved amongst WD-40 repeat containing F-box proteins. Mutation of both residues to glutamate (M178E and I179E) blocked ‘Skp1-free’ Met30ΔFbox degradation (Fig 5B). Introduction of the same mutations into full length Met30 dramatically stabilized Met30 (Fig 5C), suggesting that under normal growth conditions the majority of Met30 in the cell was being targeted for proteolysis via the ‘Skp1-free’ degradation pathway.
To determine the composition of the ligase and identify unknown adaptor components we performed mass spectrometry with purified Met30ΔFbox to analyze its binding partners. In addition, we used the same strategy to profile Cdc53 interacting proteins, with the hope to identify adaptor proteins that function as Skp1 alternatives by searching for commonalities between these two mass spectrometry datasets. We failed to identify any proteins that fit these criteria. Although negative result cannot be conclusive, this result suggested that perhaps Cdc53/Rbx1 might directly bind Met30 without an adaptor protein. To examine binding in absence of Skp1, we used a yeast strain harboring a GAL1-SKP1 allele such that Skp1 depletion could be achieved by growing the strain in media containing dextrose, which efficiently suppresses the GAL1 promoter (Fig 6A, left panel). This yeast strain served as a source of protein extract lacking Skp1. Maltose-binding protein (MBP) fused to the N-terminal region of Met30 was expressed in bacteria and used as substrate. The in vitro-binding assay was performed with immobilized (MBP)-Met30(1–186) and Skp1 depleted yeast lysates. In the absence of Skp1, Cdc53 bound effectively to (MBP)-Met30(1–186) and binding was significantly reduced when (MBP)-Met30(1–186)M178E/I179E was used as the bait (Fig 6A, right panel). Therefore, the same mutations, M178E and I179E that prevent degradation of Met30 by the ‘Skp1-free’ pathway in vivo also reduce Skp1-independent binding of Cdc53 to Met30 in vitro. Because this binding assay contained total yeast lysates, albeit lacking Skp1, it was possible that an unknown yeast protein mediated the interaction between (MBP)-Met30(1–186) and Cdc53. To test whether any other factor apart from Cdc53/Rbx1 were necessary for this interaction, we utilized 6XHisCdc53/GSTRbx1 expressed in bacteria [41] for in vitro binding experiment. Immobilized (MBP)-Met30(1–186) was incubated with Cdc53/Rbx1 expressed in bacteria and binding was analyzed by immunoblotting with anti-Cdc53 antibodies. Similar to the result with yeast lysate, Cdc53/Rbx1 could specifically interact with (MBP)-Met30(1–186) and binding was significantly decreased in the degron mutations that stabilized Met30 in vivo (Fig 6B). Cdc53 expressed in E. coli is spontaneously cleaved at the N-terminal region resulting in a truncated version (residues 267–851)[41]. Importantly, the removed residues 1–266 harbor the Skp1-binding region. Therefore, Met30 can interact with Cdc53/Rbx1 in vitro at a site distinct from the Skp1 binding domain of Cdc53.
As binding studies strongly suggested that Cdc53/Rbx1 could directly interact with Met30 without an adaptor protein, we tested whether Cdc53/Rbx1 could ubiquitylate ‘Skp1-free’ Met30 and was sufficient to function as a ligase. (MBP)-Met30(1–186) was immobilized to amylose resin and incubated with bacterial lysate expressing Cdc53/Rbx1. The substrate-ligase complex was eluted and the ubiquitylation reaction was initiated by addition of the reaction mix containing E1 enzyme, E2 conjugating enzyme, ubiquitin and ATP. (MBP)-Met30(1–186) was ubiquitylated in vitro and the reaction was dependent on Cdc53/Rbx1 complex (S4A Fig). However, the reaction is very inefficient and only a small fraction of Met30(1–186) was ubiquitylated even after 16 h incubation of the reaction. To test whether this weak activity was specific for the ‘Skp1-free’ degradation pathway of Met30, we compared ubiquitylation in (MBP)-Met30(1–186) and (MBP)-Met30(1–186)M178E/I179E mutant. Ubiquitylation of (MBP)-Met30(1–186)M178E/I179E was visibly reduced (S4B Fig). Although these experiments demonstrate ubiquitylation of Met30 by a minimal cullin-1 (Cdc53) complex lacking Skp1 the observed activity is not very robust and requires a long reaction time (~ 16 hours) and yet only a relatively small fraction of the substrate is ubiquitylated. This suggests possible involvement of another factor or a post-translational modification, which is likely not required for binding Cdc53/Rbx1 to Met30 but is required for effective ubiquitylation of the substrate. Alternatively, Cdc53 expressed in E. coli may not efficiently fold into its active structure and thus be only partially active.
Our results strongly supported the hypothesis of Skp1-independent degradation of ‘free’ Met30. We next asked whether this Skp1-independent degradation pathway was unique for Met30 or may be a more general mechanism to limit F-box protein abundance. Such a mechanism for F-box protein homeostasis could be important not only to limit the overall abundance, but also to limit competition for substrates between assembled SCF ligases and the ‘free’ F-box protein subunits, which could lead to substrate shielding and thus prevent substrate ubiquitylation. Although degradation of the F-box protein Cdc4 has been demonstrated to follow the autoubiquitylation pathway [10,11] evidence has also been reported that Cdc4 continues to be degraded in skp1-3 temperature sensitive mutants, and Cdc4 was also shown to be stabilized when Skp1 is overexpressed [14]. In addition, the kinetochore component and F-box protein Ctf13 was reported to be rapidly degraded in a Cdc34 dependent mechanism when Skp1 was inactivated [42]. These examples are consistent with ‘Skp1-free’ Cdc4 and Ctf13 degradation pathways as we suggest here for Met30. To further test this idea, we measured degradation of the F-box proteins Ctf13 and Cdc4 in skp1-td mutants (Fig 7A and 7B). Consistent with a previous report [42] Ctf13 degradation was accelerated in skp1-td mutants (Fig 7A). Cdc4 is mainly regulated via autoubiquitylation. In skp1-td mutants, Cdc4 was slightly more stable compared to wild type cells, but still degraded rapidly even though Skp1 was depleted as evident by deubiquitylated Met4 (Fig 7B) and elongated, multibudded cells. Rapid degradation of Cdc4 even in the absence of Skp1 is inconsistent with previous reports showing that deletion of the F-box region in Cdc4 significantly stabilizes the protein [11]. However, in the Cdc4ΔFbox mutant used in this study the F-box deletion extended into the region corresponding to the degron region in Met30 [43] providing a possible explanation for these conflicting results.
If the autocatalytic pathway was the major mode of degradation responsible for maintaining F-box protein homeostasis then reduction in Skp1 levels should induce complete stabilization of these proteins. Thus, these results provide evidence for existence of an additional, Skp1-independent, degradation pathway for several F-box proteins, which appears to target F-box proteins that are not bound to the SCF core complex. We refer to this degradation pathway as ‘Skp1-Free’ F-box protein degradation pathway to demarcate it from the autoubiquitylation mode of F-box protein regulation.
To explore the significance of the ‘Skp1-free’ F-box protein degradation pathway in normal cellular dynamics, we generated yeast strains bearing either wild type Met30 or the Met30M178E/I179E mutant, each controlled by the native MET30 promoter. As expected, Met30 displayed increased abundance in cells expressing the degron point mutant in comparison to those expressing the wild type allele (Fig 8A). Interestingly, despite the increased protein abundance of Met30M178E/I179E steady-state ubiquitylation of the major SCFMet30 substrate, the transcription factor Met4, [25], was slightly reduced (Fig 8A). The M178E/I179E double mutant specifically blocks Skp1-independent degradation of Met30 and it is thus conceivable that Met30 not bound to Skp1, which is normally rapidly degraded, is particularly increased in the Met30M178E/I179E strain. Consequently, a fraction of Met4 could be protected from ubiquitylation if it interacts with excess Met30M178E/I179E that cannot find a Skp1 binding partner.
We reasoned that the Skp1-independent degradation pathway should be particularly important for the recovery from heavy metal stress. Cadmium exposure leads to dissociation of Met30 from Skp1 [24,26,27] thereby generating a burst of ‘Skp1-free’ Met30, which needs to be controlled by the ‘Skp1-free’ F-box protein degradation pathway. We therefore tested cadmium sensitivity of wild type and MET30M178E/I179E strains. Cells expressing Met30M178E/I179E were significantly more sensitive and exhibited a growth arrest in response to cadmium stress (Fig 8B). We hypothesized that excess dissociated Met30 may bind Met4 and shield it from ubiquitylation by fully assembled SCFMet30 ligase during the recovery phase. The cell cycle checkpoint arrest, initially induced by deubiquitylated Met4 to cope with cadmium stress, may therefore be erroneously maintained in MET30M178E/I179E mutants resulting in apparent cadmium sensitivity. The observed growth defect in the presence of cadmium may not indicate a failure to detoxify cadmium but a defect in reversing cell cycle arrest. We tested this hypothesis by deleting MET32 in MET30M178E/I179E mutants (Fig 8C). Met32 is essential for execution of the Met4-induced cell cycle arrest, but its transcriptional role, which is important for cadmium detoxification, is redundant with Met31 [24,26]. In accordance with our hypothesis, deletion of MET32 suppressed the growth defect of cells expressing Met30M178E/I179E under cadmium stress (Fig 8C), indicating that excess dissociated Met30M178E/I179E interferes with timely inactivation of Met4. Consistent with this idea, two tested Met4 target genes, MET3 and GSH1, were derepressed in Met30M178E/I179E mutants confirming untimely Met4 activation (Fig 8D). Together, these results indicate that the ‘Skp1-free’ F-box protein degradation pathway plays an important role in cellular function to prevent substrate shielding effects by excess unbound F-box proteins.
In addition to enhanced cadmium sensitivity, Met30M178E/I179E mutants displayed increased chromosome loss (Fig 8E). The mechanism for this defect is not known. Protection of an unknown substrate in analogy to Met4 shielding during recovery from cadmium stress is a possible mechanism. However, it is also conceivable that stabilized Met30M178E/I179E interferes locally with Skp1 functions in kinetochore assembly [42,44]
We describe a novel pathway for regulation of F-box protein abundance in addition to autoubiquitylation that specifically targets F-box proteins that are dissociated from Skp1. The architectural theme of SCF ubiquitin ligases employs multiple F-box proteins that bind a common Skp1/Cdc53/Rbx1 core. This arrangement is effective in providing an array of diverse ubiquitin ligases. However, the modular design presents cells with the challenge of balancing the diversity and abundance of different SCF complexes when many F-box proteins, in the case of plants several hundred [45], compete for the shared SCF core components. Cycles of cullin neddylation and CAND1 association maintain a critical level of unoccupied SCF core complexes [9,12,46] while Skp1- F-box protein heterodimers are displaced to bind substrates and recruit them to the Cdc53/Rbx1 complex [47]. This CAND1/Nedd8 cycle maintains dynamic exchange of substrate adapters, but abundance of individual SCF ligases and overall SCF diversity is dictated by the distribution of F-box protein concentrations. It is thus critical to regulate F-box protein levels. In this study we characterize a ‘Skp1-free’ F-box protein degradation pathway that plays an important role in maintaining F-box protein homeostasis. F-box proteins that associate with Skp1 form functional ligases while those that do not, are recognized by the ‘Skp1-free’ F-box protein degradation pathway and degraded by the Cdc53/Rbx1 ligase thereby preventing competition between F-box proteins, limiting substrate shielding effects and ensuring representation even of low abundance F-box proteins in the cellular SCF repertoire. Degradation of F-box proteins that are not bound to Skp1 may also provide an important quality control mechanism to remove damaged F-box proteins. Such a mechanism may be critical for cells because F-box proteins incapable of forming active SCF ligases could maintain an intact substrate binding domain and thus shield their substrates from degradation.
We describe the ‘Skp1-free’ F-box protein degradation pathway in detail for Met30, the substrate adaptor for SCFMet30 ubiquitin ligase, which negatively regulates transcription factor Met4 by proteolysis-independent ubiquitylation [23,48,49]. The ‘Skp1-free’ F-box protein degradation pathway is of particular importance for Met30. First, because Met4 ubiquitylation does not induce its degradation under normal growth conditions, Met4 remains associated with Met30 and therefore prevents the canonical F-box-protein degradation pathway through autoubiquitylation. Accordingly, we observed that the ‘Skp1-free’ degradation pathway is the predominant pathway that ensures turn over of excess Met30 (Fig 5C). Second, heavy metal stress induces active dissociation of Met30 from Skp1 resulting in a burst of ‘Skp1-free’ Met30, which interferes with recovery from cadmium stress when the ‘Skp1-free’ degradation pathway is blocked (Fig 8B and 8C).
Indications for a degradation pathway that targets F-box proteins that are not bound to Skp1 have been reported previously [14,42]. In addition, we show that apart from Met30, the two other essential yeast F-box proteins—Cdc4 and Ctf13 are also degraded in the absence of Skp1 (Fig 7A and 7B), suggesting that this pathway is a common mechanism to restrict F-box protein abundance outside the SCF complex. However, not all F-box proteins are subject to this degradation mechanism, because consistent with other reports [10] we found that deletion of the F-box region in Grr1 or inactivation of Skp1 stabilized Grr1 (S5 Fig). The F-box protein degradation pathway we describe here is thus not universal and is probably functional for only those F-box proteins, which harbor the hydrophobic region adjacent to the F-box domain, which is paramount for Cdc53/Rbx1 ligase binding.
In addition to advancing understanding of SCF ligase regulation, our results also demonstrate a function for Cdc53 (cullin-1) independent from its adaptor Skp1. Not only was Met30 degradation active in the absence of Skp1 in vivo (Fig 3C), but in addition a Cdc53 mutant incapable of binding to Skp1 could fully complement the Met30 degradation defect of cdc53 mutants (Fig 4). In addition, Cdc53/Rbx1 can directly bind Met30 in the absence of Skp1 in vitro and binding depends on the degron region adjacent to the F-box motif (Fig 6). However, Met30 ubiquitylation in this minimal in vitro system with Cdc53/Rbx1, Cdc34, and E1 was very inefficient and required a long reaction time suggesting possible involvement of additional factors, inefficient protein folding in E. coli, or a post-translational modification lacking in this expression system. Further studies need to be conducted to explore these options in detail.
The findings reported here illustrate a Skp1 independent function for the cullin Cdc53 in substrate ubiquitylation. Consistent with these results, a Skp1 independent cullin-1 based ubiquitylation event has been suggested previously in human cells where Rictor, a component of mTORC2 complex associates with Cullin-1 instead of Skp1, to form a functional E3 ubiquitin ligase that promotes ubiquitylation of SGK1 [50].
Together, our findings shed light on the regulation and complexity of E3 ligases and suggest additional diversity in the cullin-RING family of ubiquitin ligases.
Yeast strains used in this study are isogenic to 15DaubΔ, a bar1Δ ura3Δns; a derivative of BF264-15D [51] and are listed in S1 Table. Standard culture media and yeast genetic techniques were employed [52]. Determination of protein degradation rates was done using cycloheximide chase experiments and galactose shut- off experiments. For cycloheximide chase experiments, strains carrying plasmids expressing tagged genes of interest placed under their endogenous promoter were cultured to logarithmic phase and cycloheximide (final concentration 100 μg/ml) was added and cells were collected at time points as indicated. For galactose shut-off experiments, strains expressing genes of interest under the control of GAL1 promoter were cultured in media containing 2% sucrose to logarithmic phase and cultures were transferred to rich media containing 2% galactose (YEPG) for 2 hours or as otherwise indicated. To terminate expression from the GAL1 promoter cells were transferred to YEPD and collected at the indicated time intervals. Quantitation of protein levels was performed using a Fuji LAS-4000 imaging system followed by analyses with the Multi Gauge v3 software.
For immunoblot analysis, yeast whole cell lysates were prepared under denaturing conditions in urea buffer and for immunoprecipitation cells were lysed in Triton X-100 Buffer, as previously described [53]. For purification of HBTH-tagged Met30ΔFbox, cells were lysed and purified under denaturing conditions in binding buffer (8M urea, 300mM NaCl, 0.5% NP-40, 50mM PO4 pH 8, 50mM Tris-HCl pH 8, 20mM imidazole). 1 mg of total protein lysates was used for binding to Ni2+-sepharose (GE Healthcare). Beads were then washed 3 times in binding buffer (without imidazole and pH adjusted to 6.3) and eluted in 150μl elution buffer (8M urea, 200mM NaCl, 50mM PO4, 2% SDS, 10mM EDTA, 100mM Tris-HCl, pH 4.3).
For immunoblot analyses proteins were separated by SDS-PAGE and transferred to a polyvinylidene difluoride membrane. Proteins were detected with the following primary antibodies: anti-Met4 (1:10000; a gift from M. Tyers), anti-Skp1 (1:5000; a gift from R. Deshaies), anti-myc and anti-HA (1:2000; Covance, Princeton, NJ), anti-RGS6H (1:2000; QIAGEN, Germantown, MD), anti-ubiquitin (1:2000; P4G7, #sc-53509, Santa Cruz), anti-Cdc53 (1:1000; yN-18, #sc-6716, Santa Cruz) and anti-MBP (1:2000; N-17, #sc-809, Santa Cruz).
Plasmids expressing MBP- Met30(1–186) and MBP- Met30(1–186)M178E/I179E were cloned by standard techniques into the pET28 vector and transformed into Rosetta cells. Cells were grown at 37°C and induced with 0.5 mM IPTG for 3 hours. Cells were collected, washed once with cold water, pelleted and frozen. Pellets were later suspended in recombinant protein buffer (0.05% Triton-X100, 0.05% NP-40, 150 mM NaCl, 1 mM PMSF, 1 μg/ml each aprotenin, leupeptin and pepstatin), sonicated and cleared by centrifugation at 13,000 rpm at 4°C for 10 minutes. Lysates were bound to prewashed amylose beads (New England Biolabs) for 3 hours at 4°C. Beads were then washed twice with lysis buffer and thrice with buffer U (50 mM Tris pH 8, 50 mM NaCl, 5 mM ATP, 10 mM MgCl2, 0.2 mM DTT). MBP-tagged proteins were eluted in buffer U supplemented with 10 mM maltose. Purified protein was concentrated using Amicon Ultra 50 kDa centrifugal filter and flash frozen.
For in vitro binding assay with yeast lysates, yeast cells containing a GAL1-controlled SKP1 allele and expressing endogenous TAP-tagged Cdc53 were grown in YEP galactose overnight, washed and then transferred to YEP dextrose media for 12 hours to deplete Skp1 protein levels. At this point cells were arrested and showed the characteristic elongated buds. MBP, (MBP)-Met30(1–186) and (MBP)-Met30(1–186)M178E/I179E were lysed and bound to amylose resin as described above. Yeast cell pellets were also lysed in recombinant protein buffer and 3 mg of lysate was incubated with MBP tagged proteins bound to amylose resin, for 2 hours at 4°C. Beads were washed thrice with recombinant protein buffer and bound proteins were eluted by boiling beads in 2x SDS loading buffer.
For the in vitro binding experiment performed with Cdc53/ GSTRbx1 expressed from bacteria, bacterial cell pellets were lysed in recombinant protein buffer, sonicated and 3 mg of lysate was incubated with MBP tagged proteins conjugated to amylose resin, for 2 hours at 4°C. Beads were washed thrice with recombinant protein buffer and bound proteins were eluted by boiling beads in 2x SDS loading buffer.
RNA samples were isolated, and analyzed by real- time Reverse Transcriptase (RT)-PCR as described [53]. Three biological replicates were analyzed for each experiment.
Strains harboring a centromeric plasmid with a URA3 selection marker were grown at 30°C in minimal medium lacking uracil to force cells to maintain the centromeric plasmid. 200 cells were plated on YPD and minimal media (SC-URA) plates. The remaining cells were cultured without selection for 22 hours (~ 7 generations). Cells were counted and 200 cells were plated again on YPD and minimal media plates to measure the number of cells that have lost the centromeric plasmid. Plates were incubated for 2 days at 30°C and chromosome/plasmid loss was determined by difference in number of colonies on SC-URA plates and YPD plates. Experiments were performed in triplicates.
|
10.1371/journal.pcbi.1000829 | Atomic-Resolution Simulations Predict a Transition State for Vesicle Fusion Defined by Contact of a Few Lipid Tails | Membrane fusion is essential to both cellular vesicle trafficking and infection by enveloped viruses. While the fusion protein assemblies that catalyze fusion are readily identifiable, the specific activities of the proteins involved and nature of the membrane changes they induce remain unknown. Here, we use many atomic-resolution simulations of vesicle fusion to examine the molecular mechanisms for fusion in detail. We employ committor analysis for these million-atom vesicle fusion simulations to identify a transition state for fusion stalk formation. In our simulations, this transition state occurs when the bulk properties of each lipid bilayer remain in a lamellar state but a few hydrophobic tails bulge into the hydrophilic interface layer and make contact to nucleate a stalk. Additional simulations of influenza fusion peptides in lipid bilayers show that the peptides promote similar local protrusion of lipid tails. Comparing these two sets of simulations, we obtain a common set of structural changes between the transition state for stalk formation and the local environment of peptides known to catalyze fusion. Our results thus suggest that the specific molecular properties of individual lipids are highly important to vesicle fusion and yield an explicit structural model that could help explain the mechanism of catalysis by fusion proteins.
| Membrane fusion is a common underlying process critical to neurotransmitter release, cellular trafficking, and infection by many viruses. Proteins have been identified that catalyze fusion, and mutations to these proteins have yielded important information on how fusion occurs. However, the precise mechanism by which membrane fusion begins is the subject of active investigation. We have used atomic-resolution simulations to model the process of vesicle fusion and to identify a transition state for the formation of an initial fusion stalk. Doing so required substantial technical advances in combining high-performance simulation and distributed computing to analyze the transition state of a complex reaction in a large system. The transition state we identify in our simulations involves specific structural changes by a few lipid molecules. We also simulate fusion peptides from influenza hemagglutinin and show that they promote the same structural changes as are required for fusion in our model. We therefore hypothesize that these changes to individual lipid molecules may explain a portion of the catalytic activity of fusion proteins such as influenza hemagglutinin.
| Membrane fusion is critical to eukaryotic cell function; cells rely on fusion for vesicle trafficking and secretion, and viruses such as influenza and HIV utilize fusion to infect target cells. This poses a fundamental biophysical question: how do two lipid bilayers merge in a targeted manner without rupture, and how do proteins catalyze this process? Viruses in particular are faced with a host membrane not designed to be permissive to viral entry and must alter host membrane properties to achieve fusion. Simply bringing the viral and cellular membranes together is not sufficient for physiological fusion; mutagenesis experiments in influenza [1], [2] and parainfluenza virus [3] have demonstrated that mutations to either the viral transmembrane anchor or the fusion peptide inserted in the host membrane can block fusion. In some cases [3], these mutations can be rescued by independently altering membrane properties, suggesting a direct connection between fusion peptides and lipid dynamics.
The stalk model for membrane fusion proposes that proteins catalyze the formation of a series of lipidic fusion intermediates: the outer leaflets of each bilayer merge first, followed by opening of a fusion pore and merger of the inner leaflets [4]. There is strong indirect support for this model [4]–[8], and stalk structures have been observed in artificial model systems [9], but direct observation of fusion stalks in physiological membranes is extremely challenging due to their transient nature and small size. Molecular simulations provide an alternative way to study these processes and can also provide atomic detail of the fusion mechanism and transition state, yielding insight into the mechanism of biological catalysis of fusion.
Vesicle fusion has previously been modeled with continuum approaches [8], [10]–[15] or coarse-grained simulation [16]–[19], both of which have made important contributions to refining the stalk hypothesis and outlining fusion mechanisms. One previous high-resolution simulation started from a pre-constructed stalk state, due to computational limitations, and examined a vesicle fusing to itself through a simulation boundary [20]. However, complete simulation of fusion in atomic detail has long been an important goal towards understanding atomic-level effects such as membrane dehydration and bilayer breakup upon stalk formation [21], [22].
In cells, vesicle fusion is typically catalyzed by proteins. To understand the mechanism of this catalysis, we first wish to consider the biophysical nature of fusion, its transition state, and the surrounding molecular events. We have therefore performed atomic-resolution simulations both of complete vesicles fusing and of hemagglutinin fusion peptides interacting with lipid bilayers in order to examine the mechanism of vesicle fusion and especially stalk formation in more detail. The pathway for fusion that we observe in our simulations transits through stalk and hemifused intermediates largely as predicted by the stalk hypothesis, but we observe new high-resolution details important to understanding the transition state for stalk formation and thus how fusion proteins may catalyze the fusion process.
To identify this transition state from simulations, we employ committor analysis [23]–[25], a statistical means to evaluate the transition state (as well as the full reaction pathway) that has been frequently used in the protein folding literature [23], [26], [27]. To the best of our knowledge, this marks the first time such techniques have been applied to systems of this size and complexity, simulating the million-atom vesicle fusion reaction many times over. The transition state we identify is characterized by a hydrophobic nucleation event where lipid tails from opposite vesicles make contact within the intervening hydrophilic layer. This raises the pivotal question of how fusion proteins might accelerate this hydrophobic encounter. From additional simulations of influenza fusion peptides in bilayers, we believe that fusion catalysis may be partially explained by an increased rate of hydrophobic tail protrusion in the presence of fusion peptides.
We have simulated the fusion of vesicles using model membranes composed of binary mixtures of 1-palmitoyl 2-oleoyl phosphatidylcholine (POPC) and 1-palmitoyl 2-oleoyl phosphatidylethanolamine (POPE). These were chosen because they are the two most common non-sterol phospholipids in viral and eukaryotic cell membranes [28]–[31] and form the basis for a number of experimental fusion models, often in combination with cholesterol and sphingomyelin [32], [33]. Recent results suggest that synaptotagmin may induce membrane curvature on the order of 17 nm [34], [35]; we use 15-nm vesicles both for reasons of computational tractability and to approximate that proposed curvature. This corresponds to the small end of the size range for experimentally producible vesicles[36]. Vesicle pairs were placed at 1 nm separation and connected by a single chemical crosslinker per vesicle pair to approximate the appositional effect of fusion proteins. The total system size including solvent was just over a million atoms.
Seven vesicle pairs ranging from 75–100% mole fraction POPE fused in 70–250 ns of simulation each (Figure 1); two pairs at 50% POPE did not fuse in 200 and 700 ns respectively, although we expect them to fuse in longer simulations. Aggregate simulation time totaled 10 microseconds. In all simulations, fusion occurred via initial formation of a small stalk, expansion of the hemifusion diaphragm, and subsequent opening of a fusion pore, consistent with the stalk hypothesis. In successful fusion events, vesicle pairs formed a flattened headgroup-headgroup interface with a thinned water layer prior to fusion stalk formation (Figure 1); this flattened interface greatly increases the contact area between the two vesicles, at the cost of membrane deformation. The 1∶1 POPE:POPC vesicles did not form such an interface, consistent with higher deformation energy and slower stalk formation in POPC-enriched membranes. This flattening is consistent by previous simulations by Stevens et al. [17] but was not shown in other previous coarse-grained simulations [16], [18], [37], and it has not been predicted by continuum models [11], [15]. In addition to a more detailed lipid model, our simulations use an atomic-resolution explicit solvent (the TIP3P water model [38]); desolvation effects may be important to formation of the flattened contact patch.
To further probe the key structural features of the fusion stalk, we have performed committor analysis [23] to quantitatively identify a member of the transition state ensemble. We took simulation snapshots at 5-ns intervals from a fusion simulation and performed 20 simulations each 20 ns in length from each snapshot for a total of 400 20-ns simulations. Analysis of the resulting dataset yields the free-energy profile of a single fusion reaction (Figure 2). The transition state for stalk formation is identified via this committor analysis as the snapshot equally likely to form a stalk or remain as a contact patch. We confirm that the contact patch structure described above is metastable state or local free energy minimum, neither breaking apart rapidly nor rapidly proceeding to stalk formation. We find contact patches to be metastable for the tens to hundreds of nanoseconds, depending in part on the lipid composition. After formation of the contact patch but prior to the transition state, the water layer between vesicles thins substantially (Figure 3). The transition state occurs when a pair of lipid tails from opposing vesicles (Figure 4) make contact in the intervening polar layer. This creates a small hydrophobic region, which either breaks apart and returns to the contact patch structure or grows to form a stable stalk. Contact patch formation and water layer thinning are quantified in Figure 5. Analysis of additional independent fusion simulations confirms this lipid tail contact to be a consistent feature of stalk formation. At the time of contact, lipid tails bulge slightly into the hydrophilic layer and make contact to nucleate a stalk, but they are not grossly flipped, remaining roughly tangent to the vesicle surface (Figure 6). In our simulations, these bulging tails make contact in the polar layer between bilayers rather than inserting into the opposite bilayer. This is a difference from previous coarse-grained simulations [17] and may reflect the increased chain entropy of atomic-resolution lipids compared with coarse-grained simulations.
These simulations suggest that the defining event for fusion occurs when two lipid tails from opposing vesicles make contact through the hydrophilic layer. To first order we can assume these bulging or protrusion events to be independent, making the nucleation probability proportional to the number of contacting lipid pairs, or equivalently the contact area A(t) at time t divided by the area per lipid head group ρ. Since this is a second-order reaction that depends on contact by two lipid tails, the nucleation probability varies with γ2, where γ(t) is the probability of a lipid tail bulging into the hydrophilic layer.
This model would explain why contact patch formation increases stalk formation rates; it also provides a new context to interpret the cooperative activity of fusion proteins and how they may interact with membranes. Engagement of multiple proteins, particularly in the ring arrangement proposed to drive fusion [39], will help promote larger contact patch formation, thus catalyzing stalk formation. Fusion proteins have also been proposed to catalyze fusion by disordering the lipid bilayer [40]. In our formalism, this effect could manifest as a local increase in γ, the probability of lipid tail protrusion, in the vicinity of the fusion peptides.
We use simulations to examine these hypotheses regarding lipid tail protrusion in closer detail. Both tail protrusion probability and lipid tail SCD order parameters in the vesicles are uncorrelated with spatial proximity to the contact interface in our simulations (each individual correlation coefficient <0.2), so the formation of a contact patch does not itself increase protrusion rates. The uniformity of the lipid order parameters across the vesicle surface also argues against a phase change in the contact region; the contact region remains lamellar prior to stalk formation. Average SCD values from 5–15 ns prior to stalk formation are highly similar to those in planar POPC bilayers (Pearson correlation coefficient of 0.94 between values in simulated POPE vesicles and POPC bilayers measured experimentally [41]). In our simulations, therefore, fusion does not occur via a large-scale “disorientation” of lipid tails in closely apposed bilayers as has been previously suggested [42]. Lipid composition does affect protrusion, as POPE vesicles have significantly higher protrusion rates than POPC vesicles (p<0.02, Kolmogorov-Smirnov test).
We also tested the ability of influenza fusion peptides to induce the tail protrusion we observe in the transition state for fusion stalk formation. Hemagglutinin fusion peptides (HA2 residues 1–20) were simulated in POPC bilayers at a peptide:lipid ratio of 3∶500 for 200 ns. Such peptides have previously been studied via molecular dynamics [43]–[46] and predicted to have a disordering effect on bilayers. In our simulations, hemagglutinin significantly increased lipid tail protrusion in nearby lipids but not the bilayer as a whole (Figure 7): lipids within 5 Å of the peptides exhibited significantly increased protrusion frequencies compared to lipids greater than 20 Å away (p<0.02 via Kolmogorov-Smirnov). No such effect was seen in membrane-inserted ion channel used as a negative control, and at >20 Å the protrusion probability was identical within error to protein-free bilayers. This local increase in protrusion does not solely account for the catalytic activity of fusion peptides, but it explains an important contribution to increased stalk formation rates. Most importantly, it provides a explicit lipid structural model for the general disordering effect that fusion peptides are thought to have on lipid bilayers to induce fusion [47].
In our simulations, fusion occurs via the following pattern: formation of a contact patch between the two vesicles precedes stalk formation. In this contact region, we observe thinning of the water layer between vesicles. Stalk formation is nucleated by a stochastic event: hydrophobic contact between a single pair of lipid tails that bulge into this water layer. If we approximate the protrusion of any single lipid tail as a statistically independent event, we derive the following model for the probability of stalk nucleation in any given time interval Δt:where A(t) is the contact patch area at time t, ρ is the area per lipid head group, and γ(t) is the probability of any single tail protruding at time t, and n is the number of contacting tails required to form the transition state. Under our simulation conditions, the transition state contains one tail from each vesicle, so in this case n = 2. As expected for a stochastic encounter in a planar region, most stalks form off-center. This is consistent with previous simulation reports [17] and follows trivially from our model—since P(nucleation) is proportional to the contact area A, it varies with r2, where r is distance to the contact patch center—but this is not how such stalks are intuitively envisioned.
The vesicles simulated here are substantially smaller than either synaptic vesicles or viral particles. This was done for reasons of computational tractability, as smaller vesicles contain fewer atoms and higher membrane curvature increases the rate of fusion pore formation. Our prediction of a flattened pre-stalk intermediate in small, highly-curved vesicles is thus particularly interesting, as flattened structures would be even more favorable in larger vesicles with lower average curvature. Compared to our small simulated vesicles, we expect physiologic fusion from a comparable activated intermediate to proceed more slowly. The kinetics of fusion are difficult to separate experimentally from the generation of an activated complex; physiologic rates have been measured as fast as ∼200 µs from calcium trigger to fusion [48] while reconstituted systems are typically slower, as fast as milliseconds for synaptic fusion [49], [50] or in the milliseconds to seconds range for viral fusion [51], [52].
These atomic-resolution simulations suggest a structural model of the transition state that could explain many aspects of fusion protein activity. To generate this structural model, we have combined parallel simulations on traditional supercomputers with many shorter simulations in a distributed computing environment to apply committor analysis to million-atom systems. In the resulting model, membrane bending by fusion protein assemblies accelerates fusion in part by driving contact patch formation. The tail protrusion induced by influenza peptides in our simulations suggests a mechanism for fusion catalysis by bilayer disordering. Other proteins such as parainfluenza virus F protein [3] and synaptotagmin [53] that are thought to catalyze fusion in part by membrane perturbation near the site of stalk formation might also act in part by catalyzing lipid tail protrusion. This suggests the hypothesis that increased lipid tail protrusion could provide a common physical mechanism of catalysis for structurally diverse proteins: class I viral fusion peptides, membrane-associated loops of class II fusion proteins, and neuronal synaptotagmin.
Each 15-nm vesicle was composed of 877 POPC or POPE phospholipids using the Berger simulation parameters [54]. The crosslinker structure was -CO(CH2)4CO-, connected to POPC lipids via an amide linkage to the headgroup nitrogen. Individual vesicles were first equilibrated in the TIP3P explicit solvent model of water [38]. Pairs of vesicles were then placed at 1 nm separation in a hexagonal box with sides 21 nm and height 32.5 nm and solvated in TIP3P water with or without 150 mM NaCl, leading to a system size of over a million atoms. Simulations were run using Gromacs 4.0 [55] under constant temperature and pressure using Berendsen pressure coupling and the velocity-rescaling thermostat at 310 K [56]. All covalent bond lengths were constrained using LINCS [57], and long-range electrostatics were computed every step using Particle Mesh Ewald (PME) [58]. The amine hydrogen atoms on POPE were converted to virtual interaction sites [55] to enable longer time steps by constraining the only polar hydrogens in the lipid system. The atomic coordinates are constructed every step, and forces acting on them are interpolated back onto the mass centers. This approach has been shown to conserve energy [59], but we also checked the model by testing both 2 fs or 4 fs timesteps, with equivalent results for a pair of full fusion trajectories.
Hemagglutinin fusion peptides were simulated based on PDB structure 1IBN [60] using the AMBER03 force field to model the amino acids [61]. 3 copies of the fusion peptide were placed in a bilayer as reported previously [44] for a total peptide:lipid ratio of 3∶500, solvated with TIP3P water and 150 mM NaCl, and simulated for 200 ns using 2 fs timesteps, PME electrostatics, and constant pressure and temperature conditions with at 300 K with semi-isotropic pressure coupling. Lipid tail protrusion rates were significantly increased for lipids with an average distance of 5 Å to the closest peptide atom (p<0.02, Kolmogorov-Smirnov test). Simulations of a GLIC ion channel based on PDB structure 3EI0 [62] in a POPC bilayer were used as a negative control and showed no such increase. Tail protrusion was defined as any carbon in the lipid tail protruding more than 1 Å beyond the phosphate group.
Each long fusion simulation was run on 128 cores of a cluster using Intel Clovertown or Harpertown CPU's respectively connected by an Infiniband network; the 400 “shooting” simulations were each run on 8–16 cores using the Folding@Home distributed computing network [63]. The aggregate length of vesicle fusion simulations was 10 microseconds.
|
10.1371/journal.pntd.0002372 | Designing Programs for Eliminating Canine Rabies from Islands: Bali, Indonesia as a Case Study | Canine rabies is one of the most important and feared zoonotic diseases in the world. In some regions rabies elimination is being successfully coordinated, whereas in others rabies is endemic and continues to spread to uninfected areas. As epidemics emerge, both accepted and contentious control methods are used, as questions remain over the most effective strategy to eliminate rabies. The Indonesian island of Bali was rabies-free until 2008 when an epidemic in domestic dogs began, resulting in the deaths of over 100 people. Here we analyze data from the epidemic and compare the effectiveness of control methods at eliminating rabies.
Using data from Bali, we estimated the basic reproductive number, R0, of rabies in dogs, to be ∼1·2, almost identical to that obtained in ten–fold less dense dog populations and suggesting rabies will not be effectively controlled by reducing dog density. We then developed a model to compare options for mass dog vaccination. Comprehensive high coverage was the single most important factor for achieving elimination, with omission of even small areas (<0.5% of the dog population) jeopardizing success. Parameterizing the model with data from the 2010 and 2011 vaccination campaigns, we show that a comprehensive high coverage campaign in 2012 would likely result in elimination, saving ∼550 human lives and ∼$15 million in prophylaxis costs over the next ten years.
The elimination of rabies from Bali will not be achieved through achievable reductions in dog density. To ensure elimination, concerted high coverage, repeated, mass dog vaccination campaigns are necessary and the cooperation of all regions of the island is critical. Momentum is building towards development of a strategy for the global elimination of canine rabies, and this study offers valuable new insights about the dynamics and control of this disease, with immediate practical relevance.
| Canine rabies continues to cause tens of thousands of horrific deaths worldwide, primarily in Asia and Africa. Momentum is building towards development of a global elimination strategy for canine rabies, but questions remain over how best to eliminate rabies epidemics. This paper uses data generated from the recent high-profile rabies outbreak in Bali, Indonesia to evaluate different control options. We find that, despite high dog densities, the spread of rabies on the island was remarkably similar to canine rabies spread elsewhere, suggesting that the practice of dog culling is an ineffective control strategy. We then simulate rabies transmission and spread across the island and compare the effectiveness of mass dog vaccination strategies in terms of how many lives are saved and how long it will take for elimination to be achieved. We find that the effectiveness of campaigns is not improved by being more reactive or synchronized but depends almost entirely upon reaching sufficient coverage (70%) across the population in successive campaigns. Even small ‘gaps’ in vaccination coverage can significantly impede the prospects of elimination. The outputs of this study provide the kind of evidence needed by rabies program coordinators to help design effective national control programmes, and to build the evidence-base to drive forward the development and implementation of effective global rabies policy.
| Rabies transmitted by domestic dogs is a re–emerging public health problem in Asia. In recent years incidence has increased dramatically in China [1], [2]; multiple incursions have been reported from Bhutan [3], [4]; and the disease has spread across several previously rabies–free islands in Indonesia (Flores 1997 [5], Maluku 2003, North Maluku 2005, West Kalimantan 2005, Nias 2009 [6]), including the popular tourist destination of Bali [7].
The island province of Bali was historically rabies–free until late 2008, when several local people died in the southernmost peninsula showing signs of the disease. An incursion is thought to have occurred approximately seven months earlier, when a fisherman landed on the peninsula with a dog that was incubating the virus [8]. Initial control efforts by the Balinese government attempted to contain the outbreak to the two administrative districts (Regencies) within the peninsula. However, in August 2009 a human case was diagnosed beyond the outbreak locality, and by July 2010 cases had been confirmed in all nine Regencies of Bali and 62 people had died (Fig. 1). As is common with an unexpected incursion: the island lacked surveillance, medical staff trained in rabies diagnosis, and contingency planning. The ensuing epidemic generated local and international pressure to eradicate rabies and led to plans for island–wide mass vaccination of the dog population (ProMED-mail archive number 20100806.2673).
The government's main concern for the effectiveness of any proposed rabies control programme on Bali was the high density of domestic dogs. Dogs are an important part of Balinese culture; the majority of households own at least one dog [9], though most are unconfined and not easy to restrain for parenteral vaccination. However, a pilot vaccination campaign that used trained dog–catchers equipped with nets showed that more than 80% of dogs could be vaccinated [10], with a team of six vaccinating around 100 dogs per day (Fig. 1A). From initial estimates of the human∶dog ratio (8∶1) the Bali dog population was extrapolated to be 400,000, with densities exceeding 250 km−2 in urban areas [11]. The basic reproductive number, R0, measures the average number of secondary cases arising from a primary infected individual in an otherwise fully susceptible population, and determines the critical level of vaccination coverage needed to protect the population (‘herd immunity’) and bring a disease under control [12]. For directly transmitted diseases such as rabies, R0 is often assumed to depend on population density [12], implying that such high–density dog populations could limit the success of mass vaccination. Estimating R0 for rabies on Bali was therefore a priority for determining whether vaccination would be a feasible control strategy and for setting coverage targets. The relationship between dog rabies incidence and human rabies deaths was a further important consideration for estimating public health impacts of proposed strategies.
Considerable successes have been achieved in the control of rabies in many parts of the world through the mass vaccination of domestic dogs [13], [14], [15], [16] and mounting evidence demonstrates that regional elimination of canine rabies is possible through sustained annual campaigns that attain 70% coverage [17], [18]. However, there are no operational guidelines on how to roll out dog vaccination campaigns strategically in the face of an emerging epidemic. We developed a model to capture the inherent variation in epidemic trajectories, particularly as eradication is approached, to guide strategic choices in planning Bali's first island–wide mass vaccination campaign. We fitted dog rabies incidence to human deaths in order to link the model output to potential human deaths averted. The model addressed concerns over the extremely dense population of dogs and presumed high levels of dog population turnover. We used the model to investigate whether vaccination campaigns that reach 70% of dogs on Bali could provide herd immunity, and how many campaigns would be needed to achieve eradication. We investigated how campaign effectiveness might be affected by use of locally–produced (potentially more affordable and sustainable) vaccines versus longer–acting, imported vaccines, by the speed of delivery and strategic rolling out of the programme across Bali and by the interval between campaigns. Then we examined how robust campaign performance would be to human–mediated movement of dogs around Bali and heterogeneities in coverage arising from political, logistical and operational constraints. Finally we explored the impacts of the vaccination campaigns that have since been implemented on Bali and their prospects for achieving eradication, and provide advice for how these prospects may be enhanced.
Figure 2 provides a visual summary of the model of dog-dog transmission and spread across Bali, as well as the functional form used to predict human rabies cases. We assumed that each infectious dog case causes κ secondary dog cases (‘offspring’), drawn from a negative binomial distribution (κ∼negative binomial(R0, k), Table 1, Fig. 2Ai), with R0 as its mean [20], [21]. Each secondary case was assigned a generation interval selected from a gamma distribution [18] (Table 1, Fig. 2Aii) representing an incubation period plus a period of infection prior to transmission, to determine when new infections were generated. Using an explicit spatial representation of Bali based on 1 km2 grid cells (Fig. 2A), we probabilistically allocated the location of each secondary case. To capture human–mediated transport of dogs across the island, exposed offspring were assigned to a randomly chosen grid cell with probability p so that infected dogs could potentially travel much further distances than a rabid dog is capable of running. To capture the local movement of rabid dogs, secondary cases were displaced from their direct epidemiological predecessors according to a gamma–distributed dispersal kernel [18] (Table 1, Fig. 2Aiii), with probability 1–p.
We estimated the initial epidemic growth rate λ from the monthly time series of confirmed dog rabies cases (Fig. 1A) using a generalized linear model with negative binomial errors [18]. We converted the inferred initial epidemic growth rate to an estimate of R0 using the probability distribution function of the generation interval (Gt) for rabies based on data from natural infections [18], according to Wallinga & Lipstich [22]:
The R0 estimate for Bali was used as the mean of the offspring distribution in the model (Fig. 2Ai).
To estimate the relationship between confirmed dog rabies cases and human deaths, we fitted several functions using maximum likelihood and used AIC to select the best fitting model (Fig. 2Av).
The probability of human-mediated transport of dogs across the island, p, was inferred by incrementally increasing the proportion of rabid and incubating dogs that were moved randomly across the island until the modelled speed of spread matched the observed spread of the epidemic (Fig. 2Aiv, assuming a case detection probability of 0.07 [6]). Other parameters used in the model (Table 1, Fig. 2) were derived from epidemiological data on naturally infected rabid dogs in Tanzania [18].
We modeled vaccination coverage (the proportion of dogs vaccinated, V) in each cell as waning exponentially from the coverage achieved at the time of vaccination, at a rate (Δt = one day) determined by dog population turnover (b = birth rate and death rate, assuming constant population size) and the duration of the vaccine–induced immunity (τ, where v = 1/τ):Parameter estimates are provided in Table 1. We made the conservative assumption that coverage did not accumulate over multiple vaccination campaigns (see Supporting Information for more details). Dog vaccination is represented in the model by reducing the number of secondary cases per primary infection in direct proportion to vaccination coverage at the time of transmission. In effect, each potential secondary case becomes infectious with probability 1−Vt, so in a vaccinated population the number of secondary cases attributed to each case is κv∼binomial(κ,Vt).
The branching process formulation does not account for any effects of depletion of the susceptible population as disease incidence increases. However, since detected incidence on Bali did not exceed 0.2% per annum, depletion of the susceptible population is assumed to play a negligible part. Likewise, we did not include the effects of rabies incidence on the proportion of dogs vaccinated.
The island–wide mass vaccinations on Bali began in October 2010 by which time 477 cases of rabies had been confirmed in dogs. We suspect that samples were retrieved from less than 10% of rabies cases (based on [6], [23] and previous experience during intensive contact tracing studies in northern Tanzania that suggest samples are recovered from around 5–10% of identified cases), therefore we commence vaccination in the model after 7,000 cases had occurred in model realizations. We assume the vaccinations failed to eradicate rabies if 40,000 cases were reached. The expected behavior of the epidemic under alternative scenarios was estimated using two measures: 1) the probability of eradication of rabies from Bali, and 2) the time to eradication from the onset of vaccination. For each scenario we ran 1,000 realizations of the model.
Statistical analyses were carried out in R (version 2.14.2, R Core Team 2012) and the model was built in MATLAB (version 7 release 14, The MathWorks Inc.). Codes are available upon request to the corresponding author.
We explored the sensitivity of performance measures to variation in R0 (between 1 and 2 based on estimates of R0 from rabies outbreaks around the world [18]), vaccination coverage; domestic dog population turnover (assuming constant population size and birth/death rates varying from 0.1 to 2.3 year−1 spanning a range of population replacement from 10% to 90% per year); duration of vaccine–induced immunity; and variation in long-distance dog movement, to investigate the potential impact of restrictions on human–mediated transport of exposed or infected dogs.
The island grid was aggregated into 24 rectangular blocks of similar size (mean 277 km2, range 49–500 km2) to evaluate strategies. We analyzed repeat campaigns (1, 2 and 3 campaigns) under a range of coverage levels (40%, 60% or 80%) and inter–campaign intervals (0, 6 or 12 months). We considered one synchronous campaign vaccinating all 24 blocks in the same month (A in Table 2), four proactive strategies each of six–month duration vaccinating four blocks each month in different sequences (random, rotate, source and furthest, B–E in Table 2) and two reactive strategies of six–month duration (F–G in Table 2). To examine the impact of heterogeneity in vaccination coverage we compared the effect of leaving unvaccinated areas distributed across the island in two ways: either randomly distributed unvaccinated 1 km2 grid cells, or equivalently–sized contiguous blocks of unvaccinated grid cells. Videos of model simulations of a sample of the scenarios we considered are available as Supporting Information.
To estimate vaccination coverages achieved in Bali, data on vaccination dates, numbers of dogs vaccinated and post-vaccination surveys (counts of dogs with or without collars signifying vaccination) were compiled at the banjar (sub-village) level, where possible. Where data were only available at courser resolution, numbers of dogs vaccinated were split between corresponding villages and banjars. Dog population size was calculated from post-vaccination surveys in banjars as: dogs vaccinated/(collared dog count/total dogs counted). If surveys were not available, dog populations were estimated from the human∶dog ratio for the village, district or regency as available. To obtain vaccination coverages by 1 km2 grid cell, banjar centroids were assigned randomly within their village polygon, and coverage averaged from banjar centroids within the grid cell or, if empty, assigned from the nearest banjar centroid. We assumed lakes, reservoirs, forested areas and mountain peaks were not inhabited by dogs. Coverage was assumed to wane as described above, and epidemic trajectories were simulated across the resulting dynamic coverage landscape.
During the course of the Bali outbreak, suspect cases of dogs with rabies that had either bitten people, other animals, or had shown clinical signs of disease were reported to local veterinary laboratories. Where possible such animals were captured and quarantined for observation, though many were culled. Brain samples from animals that had died, been culled or euthanized in quarantine, were tested using the direct fluorescent antibody test to confirm the presence of rabies. These data were collated by month to generate a time series of confirmed dog rabies cases (Fig. 1A). The basic reproductive number (R0) for rabies on Bali was estimated to be 1.2 (95% percentile interval 1.0–1.3) based on the epidemic trajectory until the peak in April 2010 (Fig. 1A, see the Supporting Information for definition of a percentile interval (PI)). Regency–specific R0 estimates varied between 1.0 and 1.5.
We developed a model to capture the variation in biting behavior and movement of infectious dogs. The model was a spatially explicit, stochastic simulation of rabies spread based on a simple branching density–independent process (Fig. 2, videos of simulations are available as Supporting Information). Based on confirmed cases and the estimated date of the index case, it took 26 months, April 2008–June 2010, for rabies to be detected in all nine Regencies of Bali (Fig. 1B). We tuned the probability p of longer distance human–mediated transport of infectious/incubating animals in the model until the modelled and observed speed of epidemic spread matched (Fig. 2Aiv) estimating p to lie between 0.05 and 0.09 during this initial phase of the epidemic (Fig. 2Aiv). The best fit relationship between monthly confirmed dog rabies cases (D) and monthly human deaths (H) was a saturating functional response (Fig. 2Av) with negative binomial errors (k = 3.697):
We observed an exponential relationship between modeled R0 and the median time to rabies eradication (Fig. 3A). Above a threshold value (R0 between 1.3 and 1.4), the probability of eradication fell to below one even for annual campaigns that achieved 70% coverage (Fig. 3A). When R0 was equal to 1.2, vaccination programmes with annual campaigns eventually eradicated rabies if coverage targets of at least 40% were met (Fig. 3B, Fig. 4). If campaigns achieved the WHO–recommended target of 70% coverage, the probability of eradication was largely insensitive to population turnover and duration of vaccine–induced immunity. Only at the highest turnover rates (>70%) and shortest vaccination immunity durations (<1 year) was the time to eradication substantially prolonged (Fig. 3C&D).
The number of consecutive island–wide annual campaigns and coverage achieved strongly influenced the probability of eradicating rabies (Fig. 5A). A single high coverage (80%) campaign did not guarantee eradication, but had a reasonable probability of success (∼0.6), whereas a single 40% or 60% coverage campaign had no prospect of achieving eradication (Fig. 5Ai). Subsequent campaigns greatly increased eradication prospects: two campaigns of 80% coverage or three campaigns of 60% coverage eradicated rabies in more than 90% of model runs, but three consecutive low coverage (40%) campaigns still had a very low prospect of achieving eradication (Fig. 5Aii & iii). Six consecutive low coverage campaigns increased the likelihood of eradication to ∼90% (Fig. 3B).
Thus, a roughly equivalent reasonable chance of eradication (∼90%) can be achieved with a two high coverage (80%), three annual moderate coverage (60%) or six annual low coverage (40%) campaigns. Increasing campaign frequency did not greatly affect the probability of eradication, but annual campaigns of six–month duration with six–month inter–campaign intervals could be slightly more effective than back–to–back campaigns (Fig. 5Aii & iii).
Based on the pilot vaccinations (Fig. 1A), it was estimated that the methods used could be feasibly scaled–up to cover the entire island within a six–month period, but more intensive vaccinations (1–month synchronized) might compromise coverage because of insufficient availability of trained teams. Completing campaigns in six months rather than one month (‘sync’) delayed eradication by a few months, but these delays could be compensated for by a small increase in coverage (Fig. 5C). Therefore on the basis of six-month long campaigns, we compared strategies for how to vaccinate the island, based upon different patterns of rollout under consideration at the time of planning the first campaign (Table 2 B–E). Time to eradication under different strategies varied depending on the spatial evenness of cases and thus was sensitive to potentially long distance, human–mediated transport of dogs (Fig. 5B). When human-mediated dog movement was restricted or at low frequency (p = 0 and 0.02, Fig. 5B) cases were less evenly distributed and the strategy that most rapidly eradicated rabies started vaccinations in the southernmost Regency where the index case occurred (‘source’). In contrast, the strategy that ended in the South (‘rotate’) took longest and the random strategy and the wave–like strategy from West to East (‘furthest’) were intermediate in performance. When human-mediated dog movement was frequent (p = 0.05, Fig. 5B) all four strategies performed similarly. We also compared two six-month reactive strategies (Table 2 F–G): the strategy that vaccinated blocks solely based on incidence (‘reactive’), produced the most variation in eradication times (Fig. 5B). This strategy eradicated rabies more rapidly than all others, including the synchronized campaign, when there was no human-mediated dog movement, but took longest when human–mediated movement was frequent (p = 0.05). The performance of the reactive strategy that did not return to previously vaccinated blocks within the same campaign (‘react w/o repeat’) was more robust to long distance movement (Fig. 5B).
We looked at the probability of eradicating rabies when there were gaps in coverage and under the scenarios of low and high frequency human-mediated dog movement where dogs could potentially be transported to any point on the island. When human-mediated dog movement was relatively low (p = 0.02), and gaps were modelled by excluding randomly distributed 1 km2 grid cells during vaccinations, the effect on the probability of eradication was negligible if the total area omitted was less than ∼10% of the island (Fig. 5D) and declined in a roughly linear fashion, reaching 0.9 when ∼20% of the island was not vaccinated (Fig. 5D). In contrast, when the same proportion of unvaccinated cells were left in contiguous blocks, the probability of eradication dropped rapidly, reaching 0.9 when just 0.4% of the island's area was omitted, which equates to just three neighboring villages of Bali's ∼700 villages (Fig. 5D). In both situations, the probability of eradication reaches zero when ∼50% of the island's area is left unvaccinated, but the decline is exponential when unvaccinated grid cells are aggregated (Fig. 5D). More frequent human-mediated dog movement (p = 0.05) amplifies the effects of gaps in coverage on the probability of eradication, with a greater chance of rabies reaching and persisting in unvaccinated areas (Fig. 5D).
Incorporation of all recorded vaccination efforts on Bali was necessary to generate simulated epidemics that matched the observed epidemic trajectory (Fig. 6). This included initial localized low coverage vaccinations using locally produced vaccines that required 3-month boosters which nevertheless played an important role in building up coverage and slowing the momentum of the epidemic (Fig. 6). Control was subsequently achieved through improving the scale, coverage and orchestration of vaccination, including switching to a longer lasting vaccine (Fig. 1A): in late 2010 and early 2011 the first island-wide campaign achieved target coverages of 70%, although because the campaign took several months to implement, the average island-wide coverage was around 40% (with ongoing turnover and waning immunity continually eroding coverage, Fig. 4). A second campaign was completed later in 2011 building up island-wide coverage to around 60% (Fig. 6). The overall trajectory towards eradication appears very promising especially if gaps are addressed in a third campaign currently underway (Fig. 5Aiv & 6). However, if control measures lapse, there is a more than 30% chance that within three years rabies will re-emerge to an endemic situation (Fig. 5Aiv & 6) with around 55 human deaths per year occurring on the basis of the relationship between confirmed cases and human deaths (Fig. 2Av). Over a ten-year time horizon, under the best-case scenario of rapid eradication from Bali as a result of a 3rd comprehensive coverage vaccination campaign, approximately 550 human rabies deaths would therefore be averted in contrast to the endemic situation. Whereas if control measures are maintained, but not to the level required for eradication, low levels of rabies persistence would avert around 440 human rabies deaths but would require indefinite administration of expensive post-exposure prophylaxes (∼$1.5 million/year). These calculations assume awareness of rabies and the availability of PEP remain the same as over the course of the epidemic to date.
There are strong incentives for carrying out a mass dog vaccination programme to eradicate rabies from Bali. More than 100 human deaths have occurred since the start of the outbreak in 2008 [24]. Costs for the provision of post–exposure vaccine to bite victims in 2010 alone exceeded USD$2 million and would remain high in an endemic situation. If rabies was eradicated by mass dog vaccination, and assuming bite incidence returns to pre-outbreak levels (one tenth those in 2010), then precautionary use of post–exposure vaccine would also be ten–fold lower (∼100,000 USD per year). Following official declaration of freedom from rabies (2 years with no detected cases under effective surveillance [25]) these costs should reduce to zero. Our results suggest that eradicating rabies from Bali through mass dog vaccination is feasible; it would prevent hundreds of human rabies deaths, save millions of dollars, alleviate the trauma and panic that is currently widespread in local communities, and mitigate potential impacts on Bali's tourist industry. We investigated operational aspects of vaccination strategies to determine which are most critical to achieving eradication rapidly.
Our R0 estimate of 1.2 for rabies in Bali is remarkably similar to estimates for canine rabies elsewhere, which range from 1 to 2 [18], despite population densities varying by an order of magnitude. Even under a range of assumptions about the timing and extent of reactive control measures following confirmation of rabies on Bali, R0 remains between 1 and 2. Indeed, improvements in surveillance on Bali during the first year of the epidemic would likely lead to R0 being overestimated rather than underestimated.
The low R0 observed on Bali challenges assumptions that canine rabies transmission depends on population density [12], [17]. The relationship between R0 and density is in many ways parallel to the functional responses in predator prey interactions in population ecology. Borrowing existing concepts from population ecology helps to embed epidemiological phenomena in a different context, and may be helpful in understanding possible mechanisms underlying this relationship. The (much studied) mechanisms underlying Type 2 functional responses in predator prey interactions would be an obvious starting point suggested by the analogy. While further investigation is required to understand this phenomenon, our results suggest that moderate reductions in dog density are unlikely to have any beneficial effects on rabies control. Dog population management is often a common component of rabies control programmes, either exclusively or in combination with dog vaccination. Such programmes should be aware that the mass culling or sterilisation of dogs may not be an effective means of controlling rabies, and that as long as a high proportion of the dog population can be reached with vaccination, rabies should be brought under control.
The sensitivity of vaccination success to R0 (Fig. 3A) highlights the importance of estimating R0 locally and accurately and the need to prioritize surveillance including collection of incidence data. Overall, the low R0 suggests that only 17% of the population would need to be vaccinated to control rabies (Pcrit = 1−1/R0) [12], [17]. However, when realistic operational features are taken into account, particularly the pulsed nature of vaccination campaigns, and the birth of susceptible dogs, we find that coverage of less than 30% may never achieve eradication (Fig. 3B). At least 40% of dogs must be vaccinated to maintain island–wide coverage above 17% at all times (Fig. 4) and consecutive annual campaigns are needed to ensure eradication given the stochastic nature of rabies spread (Figs. 3 & 4). With annual comprehensive vaccinations achieving uniformly high coverage of at least 70% as recommended by WHO [12], [17] we would expect rabies to be eradicated from Bali within 1–3 years of initiating comprehensive vaccinations (Fig. 3B).
While we find that achieving high vaccination coverage is a decisive factor for disease elimination, follow–up campaigns are essential for achieving eradication, especially when achieving high coverage is problematic. At lower coverage, rapid population turnover and use of vaccines that confer only short–lived immunity could cause population–level protection to fall below Pcrit and reduce or preclude the chance of eradication (Fig. 4). Therefore use of long–acting vaccines particularly in populations with high turnover is recommended (Fig. 3C&D). We found a positive effect of six–month intervals between campaigns (Fig. 5Aii) probably because coverage levels were maintained above Pcrit for longer than with equivalent effort in back–to–back campaigns [26]. Our results highlight that a successful vaccination programme requires comprehensive and even coverage. Missing randomly distributed small pockets (totaling <10% of the total area) may not be overly detrimental, but omitting an equivalent contiguous area such as an administrative unit, could jeopardize an entire programme. Hence, mass vaccination programmes which are not perfectly implemented everywhere are of less concern than lack of participation from all communities.
High intensity mass vaccinations conducted over short periods that eradicated rabies from other regions [27] raised concerns about the need to complete campaigns on Bali as rapidly as possible. Our findings suggest taking longer to vaccinate a population (six–months versus one–month) has little impact on the success of otherwise equivalent campaigns, thus easing considerably the otherwise daunting logistical and financial challenges of synchronized mass vaccination campaigns [28]. In practice, increasing the speed with which a campaign is delivered might result in trade–offs if, for example, constraints include availability of personnel. Such logistical considerations are important: for instance, a slower six–month, but higher coverage (70%) campaign takes the same time to eradicate rabies as a one–month synchronized lower coverage (60%) campaign (Fig. 5C). Taking longer to reach more dogs will have a greater impact than achieving low coverage quickly, offering further optimism that eradication is still feasible where resources are limited or hard to synchronize (e.g. community–based).
In terms of spatial roll out, there may be advantages to starting vaccinations where an outbreak began, because this is probably where there are most cases and is the most intuitive starting point for policy makers. However, this may only improve the chances of success if long distance human-mediated dog movement is restricted. The reactive strategy emphasizes this point: with no long distance transport, eradication times were fastest using this strategy because the most infected areas were vaccinated repeatedly. Yet with frequent long distance transport (5%, and as was estimated on Bali) the reactive strategy performed worse than all others. Thus, while in some situations the reactive strategy could pay dividends, it is risky for at least two reasons: first, movement restrictions to slow rabies spread may be difficult to implement; and second, the potential to control an outbreak depends not only on the speed of transmission (R0 and dog movement), but also the quality of surveillance [6] and responsiveness of control measures [29]. In Bali surveillance was not in place before the incursion, which led to delays in initiating a response, and the culling of dogs caused some people to move their dogs to safer areas. Establishing national surveillance and emergency response procedures should be prioritized given the continuing spread of rabies in the region. Further investigation into the potential of reactive strategies is warranted, including contact tracing in focal areas of transmission, and modeling to predict undetected infections [30] and to identify locations posing the greatest risks [26]. Future data collection on the human transport of dogs would be valuable for modeling realistic patterns of spread that may help direct targeted vaccination.
Overall, our analyses strongly support the feasibility of rabies eradication from Bali and our modeling conclusions are borne out by the vaccinations campaigns carried out to date (Fig. 6). Whilst logistical difficulties of mobilization and implementation proved challenging, and heterogeneities in coverage compromised overall effectiveness, the extensive vaccination campaigns conducted have brought the epidemic under control. Further campaigns will be needed to eradicate rabies from Bali, and improving the comprehensiveness of these campaigns should be a high priority to achieve this goal. Once rabies does reach very low levels, then control measures may lapse and the risk of new incursions becomes an obvious danger, which we have not considered here. These risks are being evaluated in on-going field and modelling studies but, in the long term, genetic data could provide valuable information about the frequency and source of incursions. Eradication of rabies from Bali would not only save hundreds of lives, and millions of dollars by mitigating the indefinite need for expensive post-exposure prophylaxis, but would provide a valuable precedent for the feasibility of rabies eradication in very large and dense dog populations through effectively conducted mass vaccinations.
More generally we make the following practical recommendations: 1) There is no evidence that rabies transmission in domestic dogs is density dependent over commonly encountered ranges of dog densities, so controlling rabies in higher density dog populations should not require higher vaccination coverage; 2) Vaccines that provide at least one year of protection should be effective, but the use of vaccines of shorter duration that require a booster could compromise the effectiveness of vaccination campaigns; 3) The advantages to spatially strategic roll-out or intensified synchronous effort for implementing vaccination campaigns are not justified if the increased logistical challenges compromise coverage; 4) Human–mediated transport of dogs expedites the spread of rabies and vaccination performance could be improved by restricting dog movement. However, there is currently no infrastructure to achieve this on Bali and indeed some dog owners in Bali reportedly moved animals to avoid culling or to replace dogs that had been culled, which could jeopardize spatial targeting of vaccination; 5) While achieving high coverage ensures the best possible chance of rabies eradication, repeat campaigns are vital to guarantee this. 6) The greatest concern for eradication programmes would be the lack of participation from any administrative areas, for example in Bali, omission of even the smallest of the nine Regencies that consists of 59 villages or 6% of the island could dramatically reduce the odds of achieving eradication to one third or less (Fig. 5D). Our findings about the impact of omitting contiguous subpopulations may help explain why eradicating disease is so difficult without comprehensive coverage, particularly in landlocked areas with recurrent introductions from neighboring populations [3], [4], [26]. Determining the impact of neighbouring endemic areas on the effort required to eradicate rabies is an important question to address in future studies. Nonetheless, our results further emphasize the need for regional coordination in large–scale control programmes, as evidenced by successful control of rabies in the Americas [31] in contrast to Africa [16].
|
10.1371/journal.pcbi.1003260 | A Dendritic Mechanism for Decoding Traveling Waves: Principles and Applications to Motor Cortex | Traveling waves of neuronal oscillations have been observed in many cortical regions, including the motor and sensory cortex. Such waves are often modulated in a task-dependent fashion although their precise functional role remains a matter of debate. Here we conjecture that the cortex can utilize the direction and wavelength of traveling waves to encode information. We present a novel neural mechanism by which such information may be decoded by the spatial arrangement of receptors within the dendritic receptor field. In particular, we show how the density distributions of excitatory and inhibitory receptors can combine to act as a spatial filter of wave patterns. The proposed dendritic mechanism ensures that the neuron selectively responds to specific wave patterns, thus constituting a neural basis of pattern decoding. We validate this proposal in the descending motor system, where we model the large receptor fields of the pyramidal tract neurons — the principle outputs of the motor cortex — decoding motor commands encoded in the direction of traveling wave patterns in motor cortex. We use an existing model of field oscillations in motor cortex to investigate how the topology of the pyramidal cell receptor field acts to tune the cells responses to specific oscillatory wave patterns, even when those patterns are highly degraded. The model replicates key findings of the descending motor system during simple motor tasks, including variable interspike intervals and weak corticospinal coherence. By additionally showing how the nature of the wave patterns can be controlled by modulating the topology of local intra-cortical connections, we hence propose a novel integrated neuronal model of encoding and decoding motor commands.
| Physiological studies in humans and monkeys have revealed spatially organized waves of neuronal activity that propagate across the cortex during sensory or behavioral tasks. However the functional role of such waves remains elusive. In the present study, we use numerical simulation to investigate whether wave patterns may serve as a basis for neural coding in cortex. Specifically, we propose a theoretical dendritic mechanism which permits neurons to respond selectively to the morphological properties of waves. In this proposal, the arrangement of excitatory and inhibitory receptors within the dendritic receptor field constitutes a spatial filter of the incoming wave patterns. The proposed mechanism allows the neuron to discriminate waves based on wavelength and orientation, thereby providing a basis for neural decoding. We explore this concept in the context of the descending motor system where the pyramidal tract neurons of motor cortex monosynaptically innervate motor neurons in the spinal cord. Pyramidal tract neurons have broad dendritic fields which make them ideal candidates for spatial filters of waves in motor cortex. Our model demonstrates how wave patterns in motor cortex can be transformed into a descending motor drive which replicates some fundamental oscillatory properties of human motor physiology.
| Traveling waves of oscillatory neuronal activity have been observed at many spatial scales although their functional role remains a matter of debate [1]. Waves have been implicated in perception [2]–[7], working memory [8], pathological seizure-like states [9], motor control [10]–[12] and neural computation [13], [14]. Waves also arise readily in neurobiological models of oscillatory activity [15], [16]. We recently proposed that the morphological properties of waves in motor cortex may serve as a neural basis for encoding movement-related information [17]. In the present study we explore how spatially-organized receptors within the dendritic field allow neurons to act as spatial filters of those wave patterns to effectively decode the information contained within their wavelength, coherence and direction. We use numerical simulation to explore this proposal in the context of the human descending motor system where we model the response of the principle output neurons of the motor cortex to simulated waves in cortex (Figure 1). The dendritic receptor field is modeled as a spatial Gabor filter which selectively initiates actions potentials in the neuron whenever it detects a specific wave pattern. Gabor filters have previously been used to characterize the receptor fields of ‘simple cells’ in visual cortex [18], [19] and here we assume that similar structures may likewise be plausible in motor cortex, giving examples of how this could be accomplished. We show that dendritic fields in cortex may serve as biological Gabor filters of internally generated patterns of oscillatory activity. Furthermore, we show how the output neurons of motor cortex may use Gabor filtering to translate those oscillatory patterns into steady motor output in the spine.
The prevailing notion of dendritic computation is credited to McCullough and Pitts [20] who were first to model dendrites as a simple linear summation of synaptic input followed by nonlinear thresholding (see [21]–[24] for reviews). Contemporary accounts have since recognized that dendritic morphology also contributes to transforming synaptic currents prior to their arrival at the cell soma [25]–[28]. Pyramidal neurons, for example, perform coincidence detection between synaptic inputs arriving on the apical and basal dendrites by exploiting transmission delays within the dendrite itself [29]. Yet dendrites are more than just tapped delay lines [30], they are active structures that are sensitive to the spatial patterning of temporal sequences along the dendritic arms [31]–[33]. The timing of spatially organized synaptic inputs is particularly likely to have implications for neural computation in oscillatory neural frameworks where the phase of a signal is paramount.
A conductance-based model of the dendrite is presented that demonstrates how spatially organized inhibitory and excitatory receptors can act in unison as a biological Gabor filter. We then present a two-compartment neural model that combines a model of the dendrite as a Gabor filter coupled with a conductance-based model of the soma. The combined model thus decodes spatial phase patterns (e.g., waves) into realistic action potentials. We apply this model to the case of pyramidal tract neurons (PTNs) which are the principle output neurons of the motor cortex. These neurons have long axons that monosynaptically innervate motor neurons and interneurons in the spine (Figure 1A). The direct corticospinal pathway is known to play a role in skilled reaching and grasping movements in higher species [34] with cell discharge rates that are primarily related to muscle force [35]. PTNs also make extensive lateral connections throughout motor cortex [36], [37] (Figure 1B) and so are ideally placed to broadly sample cortical wave activity (Figure 1C). Specifically, we consider waves in beta band (12–30 Hz) oscillations. Beta oscillations have long been implicated in the execution and planning of movement [38]–[40] but only recently has that activity also been shown to be spatially organized as waves [10]–[12]. Those waves have a spatial scale of approximately 1 cm. The scale of the proposed dendritic mechanism restricts it to wave patterns at smaller spatial scales (e.g., sub-millimeter wavelengths) than those that have thus far been reported.
The efficacy of the proposed mechanism is explored by simulating the full motor pathway from cortex to muscle (Figure 1D) using established models of motor cortex [17], motor neuron (MN) [41] and the surface electromyogram (EMG) of muscle [41]. The full model recapitulates key features of neurophysiological recordings acquired during simple purposeful motor activity, particularly the task-locked modulations in rhythmic coherence between electrocortical and electromuscular activity [42]–[45]. In doing so, we integrate two active fields of research: Traveling oscillatory waves — which encode motor commands — and dendritic computation — which leads to their decoding through spatial filters.
We simulated traveling waves of beta oscillations using a neurobiologically informed model of cortex [17] where coupled oscillators represent the phases of spatially distributed oscillations within a local patch of motor cortex (Figure 2). Modeling neuronal synchronization using a phase-only model is justified by the phase-reduction approximation which has a rich history in theoretical neuroscience [46]–[50]. Oscillators were spatially coupled using an anisotropic form of inhibitory-surround coupling topology to induce traveling waves of synchronization in the cortical sheet [15], [16], [51]–[53]. The resulting wave patterns were sampled by a population of randomly placed PTNs with identical receptor field morphologies. A pool of motor neurons converted the PTN output into a net muscle drive that was quantified by the simulated EMG. It is then shown that the amplitude of the final muscle drive can be controlled by varying the orientation of the cortical wave pattern with respect to the orientation of the PTN receptor fields. The results of each of these levels in our hierarchical model are now presented in sequence.
Cortex was modeled by a 128×128 array of spatially-coupled Kuramoto [54] oscillators (Methods, equations 2–4) where the phase of each oscillator represents the net phase of a localized patch of motor cortex [17]. This model approximates large-scale beta band oscillatory activity in cortex that is thought to be mediated by the long-range lateral connections within the superficial layers [7]. Anisotropic inhibitory-surround coupling (Figures 2A,B) has been previously reported to evoke traveling waves in this model [17] where the topology of the inhibitory surround governs the wavelength and orientation of the emergent traveling waves (Figure 2C). The resulting waves tend to propagate in either direction along the major axis of the coupling kernel. It is common for the waves to be segregated into localized patches which march coherently within each patch but in opposing directions between patches. In such cases the patches appear to be bounded by chains of spiral centers.
A broad distribution of intrinsic oscillator frequencies (M = 20 Hz, SD = 4 Hz; Figure 2D, labeled ‘Osc’) was used to achieve partial synchronization between oscillators. Partial synchronization degrades the wave pattern in a manner that resembles the effects of noise even though the governing equations are entirely deterministic [16]. This injects realistic variability into the cortical model without the need for explicit stochastic terms. That variability is most evident in the simulated LFP (Figure 2E) which exhibits an ongoing waxing and waning that does not appear to repeat periodically. Waxing and waning of oscillatory signals is routinely observed in physiology. Figure 2F shows an example of MEG oscillations in human primary motor cortex recorded during a steady hold task.
To study the effect of the spatial arrangement of dendritic receptors on soma current, we simulated the synaptic currents flowing into the dendrite using the conductance-based model,(1)where is the membrane capacitance, is the membrane potential, is the membrane leak current, and are the net synaptic currents of the excitatory and inhibitory receptor populations respectively. The spatial densities of the receptor populations were chosen so they combined to form a Gabor filter (Figure 3A). The Gabor filter was tuned to respond maximally to waves of length 300 µm (Figure 3B). Many combinations of excitatory and inhibitory densities can satisfy this requirement. Here, we nominated the excitatory density as a Gaussian distribution ( µm; peak density 0.4 synapse/µm2) and solved for the inhibitory distribution given a Gabor function with µm and a peak density of 0.2 synapse/µm2 (see Methods). The resulting distributions have a width of approximately 600 µm which corresponds to the width of PTN receptor fields [36], [37], [55]. The total number of synapses circumscribed by these distributions also fell within physiological estimates of 60,000 to 100,000 synapses per neuron [56]–[58]. Supplementary Figure S1 gives an alternative example where the inhibitory distribution is nominated as Gaussian and we solve the excitatory density. In both cases, the density distributions were randomly sampled to simulate the placement of excitatory and inhibitory receptors within the dendritic field (panel C). In both cases, the estimated frequency response of the combined receptor field (panel D) matched that of the target Gabor filter (panel B), as expected.
The synaptic currents were then studied whilst simulating the bombardment of the receptor field by propagating waves of cortical activity. The waves were approximated by a sinusoidal grating that propagated across the receptor field at 6 mm (20 wavelengths) per second (e.g., Figure 3E). The sinusoidal grating permitted the amplitude of the wave to be computed at exact receptor locations (e.g., Figure 3E). The amplitude of the wave modulated the rate of synaptic bombardment between 0 and 40 spikes/sec with a long-term average of 20 spikes/sec. Synaptic spikes were simulated by a Poisson process that induces an exponential rise ( ms) and fall ( ms) in the post-synaptic conductance of the corresponding receptor (Methods, equation 10). These changes in conductance drive the synaptic currents in the membrane model (equation 1). It was found that the net synaptic current () responded selectively to the orientation of the grating pattern in the receptor field, as predicted by the Gabor filter. The grating with the preferred orientation (Figure 3E) elicits significant modulation of the net balance of excitation and inhibition among the receptor conductances, resulting in large-amplitude oscillations in current (approximately pA) as the grating propagates across the receptor field (Figure 3F). Conversely, the orthogonal grating pattern (Figure 3G) fails to modulate the conductances in any coordinated fashion hence the synaptic current remained near zero (Figure 3H). In all cases, the net inhibitory and excitatory conductances were predominantly balanced at 10 nS each, consistent with physiological observations of balanced excitation and inhibition in spontaneous cortical activity in vivo [59] and patch-clamped cells in vitro [60].
The full motor pathway from cortex to muscle (Figure 1C) was simulated to test the efficacy of translating cortical wave patterns into muscle activity. A pool of identical PTNs (n = 200) were placed randomly in the motor cortex. Following the methods of [41], the outputs of the PTNs were randomly connected to a pool of MNs (m = 100) such that each MN received input from exactly 60 PTNs with the likelihood that any two MNs had 30% of their inputs in common. MNs were modeled as leaky integrate-and-fire neurons with stochastic membrane thresholds (Methods, equations 16–17). The output spikes were convolved with heterogeneous motor unit action potentials (MUAP) to simulate the surface electromyogram (EMG) of muscle (Methods, equations 18–19). Muscle force was not modeled directly but was inferred from the amplitude envelope of the simulated EMG.
Thirty seconds of cortical traveling wave activity was simulated using a fixed cortical coupling kernel that was oriented at 60 degrees from the horizontal (as in Figure 2). This wave sequence was then decoded by PTNs with dendritic filters that were rotated away from the dominant wave orientation by and respectively in each condition. The results are rotationally equivalent to holding the orientation of the dendritic filters fixed while manipulating the orientation of the cortical coupling except in this case there are no confounds with between-trial differences in the self-organized wave patterns. Figure 9 shows various aspects of the descending motor drive for each orientation offset condition. Each column pertains to one condition. The panels in row A show the orientation of each of the dendritic filters in relation to the cortical wave pattern. The panels in row B show the distribution of firing rates exhibited by the 200 PTNs embedded in the cortex. The mean firing rates of the PTN population (22.4 Hz, 18.0 Hz, 9.7 Hz, 2.4 Hz) are seen to diminish as orientation offset increases from to which confirms that PTN responses are selective to wave orientation. The maximum responses occur when the waves are perfectly aligned with the dendritic filter ( offset) whereas the bulk of the PTNs barely fire at all in the case of offset. A persistent spread in the PTN response rates is observed for all orientation offsets. This variation is due to local defects in the wave pattern, as will be discussed later.
Figure 9C shows the distribution of firing rates for the spinal motor neurons (MN). They too exhibit diminished responses as the orientation offset of the dendritic filter is increased. Here the mean firing rates of the motor neuron pool diminishes from 9.2 Hz at zero offset down to virtually no response at offset. The absence of all but a few random spikes in the motor output in the latter case show that the background PTN spikes are insufficient to raise the motor neuron membrane potential above its firing threshold. The simulated EMG traces (Figure 9D) represent the net motor neuron activity as it would be observed at the surface of the muscle. Once again, the response is diminished with increased offset angle between the cortical waves and the dendritic filter. The EMG amplitude may be loosely interpreted as indicative of muscle contractile force.
Lastly, the average coherence between LFP and EMG over 100 randomized trials (Figure 9E; heavy black line) reveals weak but significant 20 Hz corticospinal coherence that also diminishes with offset angle. This reduction of simulated coherence with offset angle is consistent with reduced corticospinal coherence in human MEG/EEG under reduced levels of muscle force (Figure 9E; red). The weakness in the levels of simulated coherence are also consistent with physiological reports of coherence in the range 0.01–0.1 [39], [44], [72]. In our model, this weakness is a direct result of the heterogeneity among the cortical oscillator frequencies (Figure 2D). Eliminating that heterogeneity leads to stronger coherence values (approaching 0.5).
We present a novel solution to the problem of encoding and decoding motor commands in primary motor cortex using spatiotemporal patterns of beta oscillations. In particular, we propose that motor commands encoded in the morphology of traveling waves can be discriminated (decoded) by the dendritic arbors of PTNs to selectively engage spinal motor neurons, thereby orchestrating muscle movement. Our model demonstrates a unique mechanism by which spatiotemporal patterns in cortex exert control over muscle activity while also replicating key aspects of the descending motor system, including variable inter-spike intervals and weak corticospinal coherence during steady motor tasks.
The key aspect of the proposed model is the formulation of the dendritic receptor field as a filter of spatial patterns in the phase of incoming oscillatory signals. In this view, neural information is encoded by the relative timing of the synaptic input. Dendritic computation is thus portrayed as the integration of synaptic phases rather than the integration of synaptic membrane potentials. That is not to say the model confounds phase with membrane potential but rather that it emphasizes the impact of the synaptic phase on the timing of subsequent spikes produced by the neuron. This phase-based approach is consistent with emerging evidence that dendritic integration is sensitive to the relative timing and spatial location of synaptic input on the dendritic arbors [21], [22], [27], [28], [31]–[33]. Phase-only models are justified in studies of neuronal synchronization where the timing of synaptic input is of prime importance [46]–[50], [73]–[76].
We approximated the spatial integration of synaptic input across the dendritic receptor field using a two-dimensional Gabor filter. The spatial bandpass characteristics of Gabor filters are well understood and have previously been used to characterize receptor fields in visual cortex [18], [19]. In those cases, the Gabor filtering refers to the spatial properties the stimulus rather than the spatial properties of the activity patterns in the visual cortex. Nonetheless, the retinotopic map of the visual field is locally preserved in visual cortex (e.g., [77]) so it is reasonable to consider that Gabor filtering may apply at the level of cortical activity patterns. We assume that similar structures may plausibly occur in motor cortex although we are not aware of any direct experimental reports of such. Furthermore, our simulations show that the spatial arrangement of excitatory and inhibitory receptors within the dendritic field is sufficient for the neuron to act as a Gabor filter of spatial patterns of synaptic bombardment. Whilst we suggest that such excitatory and inhibitory inputs arise from local interneurons, it is also possible that such effects reflect restricted corticothalamic circuits, which are known to contribute to the response properties of visual cortical neurons [78]–[80]. We do not propose a specific explanation of how such spatially organized receptor fields may develop, except to recall that a homeostatic balance between excitation and inhibition does appear to be actively maintained by a regulatory push-pull mechanism at the level of the dendrite [60]. We also note that traveling waves have themselves been implicated in guiding the development of neuronal circuits in cerebellar cortex [81]. In such cases, spontaneously generated internal activity is thought to serve as a means of bootstrapping the development of cortical circuitry prior to the onset of sensory experience [82].
Some studies of dendritic morphology in visual cortex have previously dismissed any relationship between the morphology of the dendritic footprint and the functional selectivity of those cells to the orientation [83] or direction [84] of visual stimuli. However, those studies [83], [84] only considered the physical shape of the dendritic field and not the spatial densities of the receptors within it. We emphasize that it is the spatial distribution of excitatory and inhibitory synapses that is key to our findings, not the physical shape of the dendritic footprint. Our findings also show that asymmetrical placement of the dendritic receptors can shift the temporal phase response of cells by up to even though the underlying footprint of the receptor field is unchanged. We suggest this mechanism may be exploited by the brain to fine tune the timing of spikes relative the phase of local oscillations for such purposes as long-range neural coordination [85] or coding by phase-of-firing [70], [71]. We anticipate that the same underlying mechanism would apply to any spatial filter which has periodic modulation on finite support, not just Gabor filters. Ultimately, the veracity of any computational study rests upon the validity of its core assumptions as well as the degree to which such assumptions can be verified or refuted by independent measurement. Here the core assumption is that the spatial distribution of excitatory and inhibitory synapses across a dendritic tree serve as a spatial filter, transforming spatiotemporal patterns of local oscillatory activity in motor cortex into oscillatory changes in the soma potential and thence into periodically-modulated spike sequences in Betz cells. This lends itself to several lines of independent inquiry, including in vivo measurements that couple multi-channel measurements of local field potential to spike activity, as well as morphological characterization. Computational studies such as the present one may hence guide empirical research by providing quantitative predictions that allow differentiating between alternative competing computational frameworks. In the absence of such an approach, the level of detailed description regarding dendritic computation will remain confined to the microscopic scale, leaving macroscopic accounts reliant upon qualitative heuristics.
Although the ability of dendrites to discriminate specific temporal sequences of synaptic inputs has previously been investigated [28], [31], [33] there is relatively little research exploring the potential of dendrites to discriminate spatial patterns of oscillatory inputs. Oscillatory neural signals are key to many cognitive and behavioral processes [86]–[89] and beta oscillations in motor cortex have long been implicated in movement [44], [45], [72]. The spatial organization of those oscillations as traveling waves is only a recent discovery [10]–[12] and the present model demonstrates a plausible neural architecture for transforming small-scale (sub-millimeter) spatiotemporal activity patterns into steady muscle activity. However the present model does not account for decoding the large-scale wavelengths (1 cm) observed in motor cortex since such wavelengths far exceed the spatial resolution of individual PTN receptor fields.
In our model, oscillatory activity in cortex is translated into steady motor output. The long-term motor output remains constant for any given wave pattern, exhibiting only random fluctuations about the mean due to stochastic influences within the motor pool. Despite this overall constancy, echoes of the cortical oscillations are still transmitted through the descending motor pathways where they are observed in the model as weak levels of 20 Hz coherence between the LFP and the EMG. These simulated findings are consistent with the weak but significant levels of corticospinal coherence observed in humans and primates during steady motor tasks [42]–[44]. Motor neurons transmit oscillatory activity to the muscle almost linearly hence the weakness of the neurobiological levels of coherence must be due to degradation of the oscillatory signals in the corticospinal drive[41], [90]. In many computational models, that degradation is replicated using injected noise. In the present model it arises deterministically from the heterogeneity of the cortical oscillators without resort to explicit stochastic terms.
Oscillator heterogeneity plays an important role in the present model. Firstly it demonstrates that pattern formation and discrimination remains feasible even when the intrinsic oscillation frequencies are broadly distributed, as in the beta bandwidth (12–20 Hz). Secondly, it injects significant ongoing variability into the cortical patterns which becomes evident in the irregular PTN inter-spike intervals and the waxing and waning of the simulated LFP. The gradual reduction in simulated inter-spike variability as PTN firing rates increase from 0 Hz to 20 Hz is broadly consistent with observations in motor neurons where variability typically decreases from CV≈0.4 near 7 Hz firing to CV≈0.2 near 20 Hz firing [91], [92]. In the model, that variability arises deterministically from transient synchronizations among the cortical oscillators. In dynamical systems theory, this phenomenon is referred to as metastability because the transient states are not strictly stable but the dwell-time in the vicinity of these states is sufficiently long that they appear stable in the short-term. The richness of brain dynamics is often attributed to metastability [93]–[95] although demonstrations of metastability in neural models with an explicit functional role, such as the present, are rare.
Oscillations in our cortical model also has the effect of entraining the output spikes of individual PTNs into discrete frequency bands. This leads to step-wise increments in PTN firing frequency that would appear to counteract the ability of the PTNs to respond smoothly to changes in the cortical patterns. Nevertheless, a smooth tuning curve is recovered at the population level where the collective responses of all PTNs yields a smooth tuning curve that actually has a sharper cut-off than the constituent dendritic filters. This type of population-level response is consistent with the population code hypothesis proposed by Georgopolous and colleagues whereby specific movements are not encoded by individual neurons in motor cortex but in the collective responses of multiple neurons each with overlapping tuning curves [96].
The simulated waves in the present model only serve as a gross approximation of the traveling waves observed in motor cortex [10]–[12]. While spontaneous and stimulus-evoked waves are both observed during the planning and execution stages of movement, only the phase and amplitude of the stimulus-evoked waves have been successfully correlated with movement. It is possible that movement may also correlate with other wave features that are not are not time-locked to behavioral cues and so are not detected by these experimental techniques [10]. Nonetheless, the present interpretation of wave orientation as encoding specific motor commands is deliberately simplified. Waves in motor cortex can propagate in any direction but predominantly propagate along the rostral-caudal axis in monkeys [10], [11] and the medial-lateral axis in humans [12]. Moreover these traveling waves tend to be solitary waves — perhaps better called wave fronts — rather than the tiled wave patterns presented here. In humans, the medial-lateral propagation direction corresponds with the somatopic progression of the motor map. Consequently, it has been suggested that wave fronts may coordinate the proximal-to-distal sequencing of muscle recruitment that is common to many types of limb movement [97]. Reconciling our model with these recent empirical observations and their heuristic interpretation [97] would suggest that the very long wave front along motor cortex [10]–[12] heralds a sweep through a sequence of movements, whereas each specific movement command nested within this sequence is encoded according to local patches of continuously propagating wavefronts. Such a hierarchy of movement sequences is consistent with other accounts of complex behavior control [98] and indeed general principles of cortical dynamics [99].
Traveling waves are not restricted to motor cortex and the proposed dendritic mechanism may also generalize to traveling waves in other modalities, such as olfactory cortex [3] or visual cortex [7]. At a deeper level, traveling waves are just one specific example of spatially embedded ensemble activity. Greater information capacity could be achieved using more complex spatiotemporal patterns of activity, hence speaking to a broader computational principle, consistent with recent work showing that the spiking behavior can be predicted from its surrounding local field potential [100], [101].
In conclusion, we propose an integrated and novel account for both encoding and decoding motor commands in motor cortex, incorporating basic histological and neurophysiological data into our model. Whilst somewhat speculative — by necessity — our model makes specific predictions regarding the organization of neuronal activity during movement and the fine-grained histology of PTNs, which lend themselves to empirical testing. There exist few other computational accounts of dendritic filtering that explicitly accommodate the oscillatory nature of spatiotemporal neuronal activity. We concede that the exact encoding of motor commands likely diverges somewhat from our present abstraction. Nonetheless, we anticipate that dendritic trees are capable of filtering a broader class of oscillatory spatiotemporal patterns than those we have investigated here. By proposing a formal account that links the information available in spatially organized oscillatory activity to the architecture of dendritic arborization, we suggest a deeper computational principle that may apply more generally in the cortex.
Motor cortex was modeled as a two-dimensional array of spatially coupled Kuramoto [54] oscillators(2)with periodic boundary conditions. The phase of each oscillator represents the oscillatory neural activity of a small patch of motor cortex at spatial position . The frequency of each oscillator was drawn randomly from a normal distribution (M = 20 Hz, SD = 4 Hz) that approximates the beta bandwidth of oscillation frequencies. Center-surround spatial coupling was approximated by an anisotropic kernel,(3)based on the fourth derivative of a Gaussian surface where represents spatial distance. The kernel parameter dictates the strength of the inhibitory surround as shown in Figures 2A and 2B. The inhibitory strength varies radially according to(4)where is the angular position of each oscillator relative to the kernel midpoint and is the orientation of the major axis of the kernel itself. Parameters and thus define the inhibitory strengths along the major and minor axes of the kernel respectively. The kernel is isometric when . We have previously reported waves and uniform synchrony for this type of spatial coupling with parameter values in the range to [17].
In the present study, the strength of the inhibitory surround was fixed at and to produce traveling waves that were spatially aligned with the kernel axes (e.g. Figure 2C). The size of the coupling kernel was fixed at nodes with a Gaussian full-width-half-height of 11 nodes (i.e. ). See [17] for details of the numerical integration method.
The LFP of motor cortex was approximated by treating the cosine of the oscillator phases as analogous to membrane voltage potential and then summing those voltages across space,(5)Spectral density estimates of the LFP were computed using Welch's periodogram method with a Hamming window of 0.5 seconds and 50% window overlap. Sampling frequency was 1000 Hz.
Pyramidal tract neurons were modeled as two passively coupled neural compartments. The first compartment represents the dendritic tree which defines the spatial distribution of incoming connections from the motor cortex. The second compartment represents the soma which defines the spiking output of the neuron in response to the dendritic current. The dendritic current was simulated in two ways. The first method demonstrates the principle of Gabor filtering by dendritic receptors. This is achieved by a conductance-based model of the post-synaptic currents produced by spatially distributed populations of excitatory and inhibitory synapses. The second method applies those findings to the simulation of multiple PTNs in a computationally efficient manner. This is achieved by a phase-only model of the dendritic compartment in which a Gabor filter directly transforms incoming wave patterns into an oscillatory dendritic current.
To allow comparisons between our model of the descending motor system and known physiological properties of spinal motor neurons, we simulated a motor neuron pool that received incoming spikes from n = 200 PTNs that were randomly distributed on the model cortex. The motor neuron pool was modeled as n = 100 leaky integrate-and-fire neurons with stochastic membrane resets using the same method as [41]. The membrane potential of each MN was thus modeled as,(16)where is an all-or-none connection from Betz cell to motor neuron and is the post-synaptic current generated by the incoming spikes. The latter has exponential rise and fall,(17)where denotes the time of the spike and the terms are time constants. All other parameters are described in Table 3.
The connection weights were arranged so that each MN received input from exactly 60 PTNs. These were randomly assigned with the proviso that any two MNs would, on average, share 30% of their inputs with a common set of PTNs [41], [90].
The simulated EMG produced by the motor unit pool was obtained by convolving the motor neuron output spikes with a biologically realistic motor unit action potential (MUAP) and summing the result across all motor neurons. The MUAP was defined as(18)where(19)is a conventional bi-phasic pulse with a time constant of ms and duration of ms [90]. The amplitude of the MUAP for each motor neuron was randomly scaled between 0 and 1 to reflect natural variation in size and location of muscle fibers. Likewise, the polarity of the MUAP was inverted for randomly selected motor neurons. See [41] for the benefits of modeling heterogeneous motor action potentials.
Corticospinal coherence measures the degree by which oscillations in the EMG can be predicted by those in the LFP. It has become an important tool in exploring corticospinal interactions in motor control (see [72]). Weak but significant levels of coherence between 0.01 and 0.1 are typically observed in the beta bandwidth during steady hold tasks (e.g. [44]). We approximated corticospinal coherence by the magnitude squared coherence of the simulated LFP and EMG signals over a 30 sec data window. The coherence spectra were computed using Welch's periodogram method with a Hamming window of 0.5 sec and 50% window overlap. The 95% confidence level for the resulting coherence spectrum is where N = 120 is the total number of data windows [103], [104]. Both the EMG and coherence spectra were averaged over 100 repeat simulations to control for variation in the model parameters and stochasticity in the motor neuron model.
The variability of inter-spike intervals was quantified using both the conventional coefficient of variation (CV) metric and the irregularity (IR) metric,(20)proposed by Davies and colleagues [68]. The latter emphasizes relative changes in consecutive inter-spike intervals () and is more resistant to changes in firing rate than the coefficient of variation. We sought similar levels of inter-spike irregularity (IR≈0.6) to those reported in the PTNs of monkey primary motor cortex during a precision hold task [69].
|
10.1371/journal.ppat.1004402 | Identification and Functional Expression of a Glutamate- and Avermectin-Gated Chloride Channel from Caligus rogercresseyi, a Southern Hemisphere Sea Louse Affecting Farmed Fish | Parasitic sea lice represent a major sanitary threat to marine salmonid aquaculture, an industry accounting for 7% of world fish production. Caligus rogercresseyi is the principal sea louse species infesting farmed salmon and trout in the southern hemisphere. Most effective control of Caligus has been obtained with macrocyclic lactones (MLs) ivermectin and emamectin. These drugs target glutamate-gated chloride channels (GluCl) and act as irreversible non-competitive agonists causing neuronal inhibition, paralysis and death of the parasite. Here we report the cloning of a full-length CrGluClα receptor from Caligus rogercresseyi. Expression in Xenopus oocytes and electrophysiological assays show that CrGluClα is activated by glutamate and mediates chloride currents blocked by the ligand-gated anion channel inhibitor picrotoxin. Both ivermectin and emamectin activate CrGluClα in the absence of glutamate. The effects are irreversible and occur with an EC50 value of around 200 nM, being cooperative (nH = 2) for ivermectin but not for emamectin. Using the three-dimensional structure of a GluClα from Caenorabditis elegans, the only available for any eukaryotic ligand-gated anion channel, we have constructed a homology model for CrGluClα. Docking and molecular dynamics calculations reveal the way in which ivermectin and emamectin interact with CrGluClα. Both drugs intercalate between transmembrane domains M1 and M3 of neighbouring subunits of a pentameric structure. The structure displays three H-bonds involved in this interaction, but despite similarity in structure only of two these are conserved from the C. elegans crystal binding site. Our data strongly suggest that CrGluClα is an important target for avermectins used in the treatment of sea louse infestation in farmed salmonids and open the way for ascertaining a possible mechanism of increasing resistance to MLs in aquaculture industry. Molecular modeling could help in the design of new, more efficient drugs whilst functional expression of the receptor allows a first stage of testing of their efficacy.
| Sea lice are the main parasites affecting farmed salmon and trout in the world. Caligus rogercresseyi is the principal sea louse species infesting farmed fish in the southern hemisphere. Successful control of these parasites has been achieved using macrocyclic lactones (MLs), but resistance has emerged over time. In other invertebrates, MLs target membrane receptors regulating synaptic transmission in the parasite nervous system. Here we identify and study the function of such a receptor from Caligus rogercresseyi, and gain an idea about how two MLs, ivermectin and emamectin, interact with the receptor to produce their effects. Our molecular modeling of the protein in complex with the drugs suggests a novel way in which ivermectin and emamectin exert their effects on CrGluCl due to a lack of conservation at interaction sites identified in the crystal structure of the receptor from C. elegans. We believe that the identification of a ML target in sea louse will aid the study of drug-resistance mechanisms and could help in the design of new, more efficient antiparasitic drugs.
| Sea lice are marine ectoparasite copepods of the Caligidae family (order Siphonostomatoida) that attach to host marine fish and feed on their epidermal tissue, and blood. The Caligidae family is constituted by hundreds of species belonging to the Caligus and Lepeoptherius genera. Increasing interest in these parasites has arisen owing to the ravages they produce on fish aquaculture leading to morbidity and mortality with extremely high economic impact in the industry. Lepeoptherius salmonis is the most important caligid species affecting Northern hemisphere salmon and trout aquaculture, and its biology, sensitivity to chemotherapeutic drugs and distribution have been and presently are very actively studied [1]. A different sea louse, Caligus rogercresseyi is the most important parasite affecting Atlantic salmon and rainbow trout sea water farming in Chile. This Caligus was described as a separate species only in the year 2000 [2] and much remains to be known about its biology. As in the Northern hemisphere, Caligus rogercresseyi infestation within the nationally important Chilean aquaculture industry is associated with increased costs and decreased productivity with high social impact [Alvial et al. cited in 3].
Chemical treatment has been used in fish farms to combat sea lice infestation with variable degree of success. Compounds used include organophosphates, hydrogen peroxide, pyrethroids, chitine synthesis inhibitors and avermectins [3]. Emamectin and ivermectin are macrocyclic lactone avermectins that have been widely used to control parasitic infections in humans and animals. Emamectin benzoate has been the treatment of choice for sea lice treatment given its efficacy and ease of administration. The compound, which is formulated as SLICE (Merck Animal Health), is administered orally with fish feed and provides long-lasting protection against all forms of attached sea lice.
Studies in nematodes have shown that avermectins interfere with synaptic transmission through irreversible activation of glutamate-gated chloride channel receptors leading to eventual paralysis and death of the parasites [4]. Though highly successful as antiparasitic drugs over extended periods of time, resistance to avermectins has emerged and has become a major problem worldwide [5].
Glutamate-gated chloride channels (GluCls) belong to the Cys-loop receptor, or also known as pentameric ligand gated ion channel, family which is widely present in nematodes and insects [6]–[10]. Structurally they are pentamers and can assemble as homopentamers or heteropentamers. Functionally characterized α and β subunits (GluCl-α and GluCl-β subunits), also named GLC-1 and GLC-2, have been reported in the nematodes C. elegans [11], [12] and Hemonchus contortus [13]. The C. elegans GluCl family now extends to six genes: glc-1 to glc-4, and avr-14 and avr-15 [14]. In Drosophila melanogaster only one subunit is present, DmGluClα, which is responsible for the insect sensitivity to avermectin compounds [15]. The same seems to be the case in most insects studied [14]. Recently our understanding of the structural and functional properties of these receptors has made an enormous progress thanks to development of the crystallographic structure of Caenorabditis elegans GluClα subunit (CeGluClα) [16]. The structure revealed details of the agonist binding site in the extracellular domain as well as the ivermectin site. The physiological role of these receptors in invertebrates is to mediate and regulate the inhibitory synapses and cellular excitability making them a very important and effective pharmacological target for insecticides [14].
Emamectin has been used to combat infestation by Caligus rogercresseyi in Chilean farmed salmon systems since 2000 [2], but reports of growing resistance to the drug have emerged [17]. A detailed knowledge of the molecular mechanisms of this resistance is essential in devising strategies to circumvent it. We think that characterizing the pharmacological target of emamectin benzoate from C. rogercresseyi could help to understand the resistance mechanisms and to design new drugs. Given that this molecular entity is not known in this species, our aim in the present work was to identify an emamectin sensitive chloride channel and characterize its electrophysiological properties in oocytes of Xenopus laevis and, taking advantage of the availability of a molecular structure of member of the pentameric ligand gated ion channel family of proteins, to obtain clues about the mode of action of the avermectins.
Caligus rogercresseyi specimens were obtained from three different sources. The first were specimens obtained locally at the Universidad Austral, Valdivia, Chile. The product (see below) obtained from these specimens is hereafter referred to as CrGluCl-Vald. The other sources were salmonid cages located in the Darwin and Errázuriz Channels in Southern Chile, and their products are referred to as CrGluCl-Dw and CrGluCl-Err, respectively. All samples were collected in situ and immediately stored in RNA-later protective solution (Ambion) at −20°C until use. Pools of 6 to 10 adult specimens were pulverized under liquid nitrogen and immediately mixed with Trizol reagent (Invitrogen) to extract total RNA following the supplier instructions. The RNA concentration was quantified based on absorbance at 260 nm and its integrity evaluated by agarose gel electrophoresis.
Two degenerate oligonucleotides were designed based on an amino acid region highly conserved in arthropod glutamate gated chloride channels (GenBank accession number: NP_001171232, NP_001071277, NP_001103244, NP_732447, ABI95855). The forward (PCRdegFor(d64)) and reverse (PCRdegRev(d16)) corresponds to MEYSVQLTFRE and KTNTGEYSC amino acid sequences respectively. The reverse transcription reaction was performed in 5 µg of total RNA primed with oligodT and random primers using Superscrit II kit (Invitrogen). KOD Hot Start DNA polymerase (Novagen) was used for PCR amplification and the amplicon was subcloned in pGEM-T (Promega) vector and analyzed by sequencing. The 5′ cDNA was obtained by 5′RACE and inverse PCR based on the protocol published by Huang, which combines the template-switching effect with inverse PCR [18]. Briefly, the reverse transcription conditions using Superscrit II were basically as supplier recommends except that the mix contained T-S primer for template switching effect. One µl of reverse transcription was used to synthesize the second strand with T-S PCR and PCRdegRev(d16) primers and KOD Hot Start DNA polymerase. The PCR product was purified by column (Wizard, Promega) and quantified. 200 ng of the product were phosphorylated with Polynucleotide kinase (New England Biolabs) and then ligated with T4 DNA ligase (Fermentas). That ligated product was used as template for inverse PCR (PCRi). A PCRi number 1 was performed with PCRiFor1 and PCRiRev1 primers and the product was used as template for PCRi number 2 using PCRiFor2 and PCRiRev2 primers. The 3′ cDNA end was obtained using oligodT with an anchor sequence (cDNA Cloning Primer) in the reverse transcription reaction. The second strand was synthesized with KOD Hot Start DNA Polymerase using 1 µl of transcription reverse product as template and PCRiFor1 and PCR RACE 3′ primers (first PCR). A second PCR reaction was performed using as template the PCR product from the first reaction with the PCRiFor2 and PCR RACE 3′ primers. To obtain the full length open reading frame (ORF) the cDNA was amplified with the PCRfullFor and PCRfullRev primers. The same cDNA was used to amplify 3 partial fragments by conventional PCR using the KOD Hot Start DNA polymerase and the primers PCRfullFor and PCRiRev1, PCRmidFor and PCRmidRev and PCR3′For and PCR 3′Rev. All the PCR products were purified, cloned in the pGEM-T Easy vector (Promega) and sequenced with T7 and SP6 primers. The full ORF was excised from this vector with NotI enzyme and subcloned in the pCR3.1 vector in the same site. The oligonucleotides used in the cloning procedure are listed in Table S1.
To express the channels in Xenopus laevis oocytes, the ORF was cut with BglII and XbaI from crGluCl/pCR3.1 and directionally subcloned in the pTLB and a BamHI site was inserted downstream to the stop codon.
The plasmids crGluCl/pTLB were linearized with BamHI enzyme (Roche) and used as template for synthesis of capped cRNA using mMessage Machine SP6 kit (Ambion, Austin, TX, USA).
Defolliculated oocytes were injected with 5 ng, 10 ng and 20 ng of each cRNA and kept at 16°C in modified Barth's solution for 5 days. Two-electrode voltage clamp recording were performed at room temperature using TURBO TEC-10CX (npi electronic GmbH, Tamm, Germany) and PClamp 9 software (Axon Instruments, USA), from 2 to 4 days after oocyte injection. Oocytes were recorded in a chamber (Model RC-1Z, Warner Instrumenst, USA) under continuous superfusion. Electrodes were pulled to 0.5–2 MΩ and filled with 3 M KCl. The bath reference was 3 M KCl in a 3% agar bridge connected to an Ag–AgCl pellet. Currents were recorded in response to a ramp protocol consisting in a holding potential period of −30 mV for 40 ms followed by a voltage jump to −100 mV for 20 ms. The ramp was from −100 mV to +60 mV with a 360 ms duration. The data were filtered at 1 kHz, digitized using a digidata 1440A analogue-to-digital converter and analyzed using Axon pClamp 9 software. The solution bathing the oocytes during electrophysiological recordings had the following composition (mM): 115 NaCl, 2 KCl, 1.8 CaCl2, 1 MgCl2, 10 HEPES pH 7.5 obtained with NaOH. For the low [Cl−]o solution all NaCl was replaced with the gluconate Na salt. Stock solutions of the drugs in DMSO were: emamectin 10 mM, ivermectin 10 mM, picrotoxin 100 mM. Different concentrations of the drugs were obtained by dilution in bathing solution. The highest DMSO concentration used was seen to have no effect on CrGluCl currents. Analysis of concentration-dependence of glutamate or drug effects was done by fitting a Hill model to the data using the non-linear regression suite of Sigmaplot 12.
In the absence of structural data for the GluCl receptor of Caligus rogercresseyi (CrGluClα), we built a model to help our understanding of experimental results. CrGluClα shares a 53% sequence identity with CeGluClα whose high resolution X-ray diffraction structure has been resolved [16] (comparison of similarly truncated CrGluClα as crystallized CeGluClα). We used these data (PDB ID 3RHW) as a reference structure to build a homology model of CrGluClα using Modeller v 9.10 software [19]. The molecular model was embedded into a 130×130 Å POPC lipid bilayer in a water box. The hydrated system was neutralized with NaCl at a concentration of 150 mM. The system was submitted to a molecular dynamics simulation under periodic bordering conditions (135×135×165 Å3) and isobaric-isothermal ensemble (NPT). The full system was relaxed through molecular dynamics (MD) simulations using NAMD 2.9 software [20] for 1 ns and subsequently equilibrated for 20 ns.
The homology model structure was used in docking simulations to identify ivermectin and emamectin binding sites in CrGluClα. The molecular docking procedure used Glide [21] software of the Schrödinger suite. The drugs, that belong to the macrocyclic lactone group, were designed using Maestro 2D sketcher [22], and prepared with LigPrep [23]. The CrGluClα model was prepared using the Protein Preparation Wizard [24] panel, assigning bond order, adding missing hydrogen atoms, correcting metal ionization states and creating disulphide bonds.
Grid size and position was defined from the position of ivermectin observed in the CeGluClα-ivermectin complex crystal structure. The box centre was defined as the midpoint of a line joining the two most distant atoms of the ligand core. The size of the grid was fixed at 22 Å, considering a distance of 34×34×34 Å in the coordinate axis.
A flexible docking was done using the OPLS-AA force field with standard precision (SP). This approach is appropriate for ligand screening in model-derived structures. To internally generate conformations, the ligand was considered flexible during the coupling process. A postdocking minimization allowed optimization of bond lengths and angles, including torsion angles. Owing the fact that this kind of docking code has not been adjusted to transmembrane conditions some spatial restraints were applied to guide the conformational sampling in the binding site. In order to try and reproduce the H-bonding network observed in the ivermectin binding in the crystallographic data [16], distance restraints were applied between Thr305 (Ser321 in the crystal) and ivermectin (and emamectin) O10 and between Leu263 and ivermectin (and emamectin) O13.
Three systems were considered: CrGluClα, and the CrGluClα-ivermectin and CrGluClα-emamectin complexes. Each system was embedded into a POPC lipid bilayer in a water box (dimensions: 135×135×165 Å3) and a NaCl concentration of 150 M. NAMD software [20] was used to relax each system through molecular dynamics (MD) simulations. MD was performed with the CHARMM force field and the TIP3P water model [25]. The temperature was kept at 300°K with the isobaric-isothermal ensemble using Langevin dynamics with a damping coefficient of 1 ps−1. By means of the Langevin Piston method the pressure was fixed at 1 atm. The equations of motion were integrated employing the Verlet r-RESPA algorithm [26] with a time step of 2 fs. All systems were subjected to energy minimization using periodic boundary conditions.
After a brief stabilization of each system, HOLE algorithm [27] was used to determine channel pore dimensions, by using the coordinates of the ion channel. Pore dimensions obtained considered the last 10 ns of trajectory (total time 100 ns).
The GeneBank database was explored for the presence of sequences for glutamate-gated chloride channels belonging species phylogenetically close to Caligus rogercresseyi. The search produced sequences for Lepeoptheirus salmonis, Drosophila melanogaster, Nasonia vitripennis, Apis mellifera and Tribolium castaneum, which were used to design degenerated primers to attempt cloning a fragment of the receptor from C. rogercresseyi total RNA. The PCR amplification with the degenerated primers (Table S1) gave rise to a product of approximately 450 bp which included the two extracellular domain Cys-loops characteristic of these receptors. The predicted amino acid sequence of this fragment showed a 90% identity with the sequence from L. salmonis and 74% with that of D. melanogaster. Based on this sequence we designed specific primers to amplify the cDNA ends of the channel. In the 5′ and 3′ RACE experiments we obtained PCR products of 700 bp (5′) and 900 bp (3′). The predicted amino acid sequences were 79% (5′) and 83% (3′) identical to those of a putative receptor from L. salmonis and 62% (5′) and 55% (3′) with that of D. melanogaster. In the 5′ RACE sequence we found an ATG with many upstream in frame stop codons. We considered this ATG as the start codon.
After identifying the sequences of the cDNA ends, we designed specific primers to amplify by PCR the full ORF cDNA. RT-PCRs were carried out using RNA purified from C. rogercresseyi from the three geographical sources mentioned above. The amplification yielded products of approximately 1.4 kb (Figure S1) with predicted amino acid sequence showing the general features of the known glutamate gated chloride channels, such as a putative signal peptide preceding the large extracellular N-terminus, two Cys-loops, four transmembrane domains and a shorter C-terminus. In Figure 1, we show a representative clone aligned with the partial sequence available for a putative GluClα from northern hemisphere sea louse Lepeophterius salmonis, and the glutamate-gated chloride channel α-subunits of Drosophila melanogaster and Caenorabditis elegans. The closest to the CrGluCl sequence was that of DmGluClα, which showed a 60% identity and 75% of similarity, with respective figures of 46% and 62% for the comparison with CeGluClα full-length subunit. A more detailed analysis of 33 full-length clones showed some differences in the predicted protein sequence. Predicted sizes ranged from 461 to 464 amino acids and the differences were due to either insertion of one isoleucine at positions 20 (I or Δ) and/or deletion of alanine 376 together with serine 377 (AS or Δ) (see Table S2). In addition, the analysis also revealed 16 amino acid changes (Table S3). To determine if these were genuine differences or whether they originate in PCR errors we amplified smaller, overlapping fragments of the cDNA. Sequencing of the partial clones confirmed the following amino acidic differences, V27I, D73G, R387K and L411Q, plus the above mentioned insertion and deletion. We conclude therefore that they probably represent allelic variants of the protein. Combinations of the precedent variants give rise to 12 different clones whose phylogenetic analysis is shown in Figure 2.
To evaluate electrophysiologically the functional properties of C. rogercresseyi GluCl we utilised clones CrGluCl-Vald13 and CrGluCl-Vald2 for oocyte expression. CrGluCl-Vald13 was the most abundantly represented of the obtained clones, and CrGluCl-Vald2 contains the greatest number of amino acidic differences compared to CrGluCl-Vald13. As shown in Figure 2 these clones are positioned in the most distant branches of the phylogenetic tree of the allelic variants found. No functional difference whatsoever could be detected between the clones. The results shown below pertain to CrGluCl-Vald13, but to calculate averages these values were pooled with those of CrGluCl-Vald2. The receptor is referred to hereafter simply as CrGluClα.
Figure 3A shows current traces obtained using two-electrode voltage-clamp of an oocyte previously injected with CrGluClα RNA. Outward and inward currents were recorded respectively at 60 and −80 mV. Superfusion with increasing concentrations of L-glutamate led to a graded increase in current at both voltages. Figure 3B shows that the glutamate-dependent current could be reversibly inhibited by 100 µM picrotoxin (PTX), a known open ligand-gated chloride channel pore blocker [28], [29]. Average current at 60 mV increased from 0.49±0.043 to 3.09±0.23 µA upon addition of 100 µM glutamate (mean±SEM, n = 18). Addition of PTX decreased the current to 0.48±0.17 µA (n = 14). Current voltage relations taken during addition of 100 µM glutamate before, during and after application of PTX are shown in 3C. The glutamate-dependent current was outwardly rectifying and reversed sign at a depolarized potential compared with the residual current after inhibition with picrotoxin, whose effect was fully reversible. The average reversal potential of the glutamate-induced current was −20±2.2 mV (mean±SEM, n = 6). Figure 3D shows that the current elicited by glutamate was carried by Cl−, as extracellular replacement with an impermeant anion sharply reduced outward current (Cl− influx). Current-voltage relations were also taken during glutamate application under normal and low extracellular [Cl−]. Figure 3E shows these current-voltage relations that have been corrected for the current remaining in 100 µM PTX, a concentration affording maximal effect of the toxin (Figure S2). The outwardly rectified current was markedly reduced in low external Cl− solution and the reversal potential was shifted in a depolarised direction, as expected for a current carried by Cl−. The change in reversal potential ΔErev was 59±4.2 n = 6 (mean±SEM, n = 6), which did not differ significantly (P = 0.087, one-sample t-test) from the ΔErev value of 69 mV predicted for perfect selectivity for Cl− over gluconate. Figure 3F shows dose dependence of the glutamate effect measured at −80 and 60 mV. There was no difference between these measurements and a simultaneous fit of a Hill equation to the data gave an EC50 of 6.89±0.83 µM and nH of 1.33±0.18.
The avermectin family of lactones such as ivermectin are potent and irreversible activators of some ionotropic invertebrate receptors. We have tested the effect of ivermectin on the currents elicited in Xenopus oocytes by expression of CrGluClα. Figure 4A shows a recording of CrGluClα current at 60 mV where addition of a supramaximal activating concentration of 50 µM glutamate produced the expected large increase in outward current. Increasing concentrations of ivermectin also elicited a graded increase in outward current readily inhibited by addition of the channel blocker PTX at 100 µM. Figure 4B shows that ivermectin-dependent current was reversibly inhibited by PTX in addition to being dependent on the presence of Cl− in the bath. In six separate experiments ivermectin at 3 µM increased current from 0.39±0.04 to 2.21±0.43 µA, whilst addition of PTX decreased ivermectin-dependent current to 0.40±0.08 µA (means±SEM). The figure also illustrates the irreversibility of ivermectin effect that remains unabated after removal of the drug. Current-voltage relations taken for ivermectin-activated current before and after PTX inhibition are displayed in Figure 4C. The current-voltage relations in Figure 4D have been corrected for the current remaining in 100 µM PTX, in high and low external Cl− solution. The reversal potential shifted in a depolarised direction upon Cl− reduction. The average change in reversal potential ΔErev was 60±6.7 mV (mean±SEM, n = 4) not significantly different from 69 mV (P = 0.322), consistent with high Cl− selectivity. Emamectin is another member of the avermectin family widely used to combat sea louse infestation in fish aquaculture in its benzoate form. CrGluClα-mediated current was also irreversibly activated by emamectin in a dose-dependent manner (Figure 5A). At 3 µM emamectin, current measured at 60 mV in CrGluClα-expressing oocytes increased from 0.49±0.06 to 1.42±0.13 µA whilst PTX at 100 µM reduced the current to 0.40±0.09 µA (n = 13). Current elicited by a saturating concentration of emamectin was smaller than that elicited by glutamate. Addition of glutamate after activation with emamectin only modestly increased current (Figure S3). Figure 5B shows that the currents stimulated by the avermectins in Xenopus oocytes expressing CrGluCl were outwardly rectified. Reversal potential was −21±2.5 (n = 13) mV for emamectin-elicited current, close to the chloride equilibrium potential of Xenopus oocytes [30].
Figures 4E and 5C show respectively the dose-response curves for ivermectin and emamectin derived from a number of experiments at 60 and −80 mV. There was no significant difference between values at these two voltages, and when analysed together they gave EC50 values for ivermectin and emamectin of 181±10 and 202±21 nM, with corresponding nH values of 2.1±0.26 and 1.1±0.11.
CrGluClα shares a high percentage of sequence identity with CeGluClα whose X-ray structure has been recently solved [16]. We have used this structure to build a homology model for the Caligus receptor to be subsequently used in docking of emamectin and ivermectin followed by molecular dynamics simulations. A view of the resulting modeled structure can be seen in Figure S4. All main features of the reference structure such as secondary structure and side chain conformation were retained in the molecular model of CrGluClα, as expected from the high conservation of sequence particularly in the transmembrane domains.
In order to validate the molecular model and protocol to be used we performed a docking assay with the CeGluClα crystal structure (PDB code 3RHW) and ivermectin. The results produced a position, orientation and interactions of ivermectin with the receptor, with a RMSD of 1.1 Å, very close to those reported in the co-crystallized complex. Interactions included H-bonds with Ser321 and Thr346 in transmembrane domains 2 and 3, and Leu279 of M1 in the adjacent subunit, at respective donor-acceptor distances of 1.9, 1.9 y 2.3 Å. Notice that the numbering used refers to the full-length deduced amino acid sequence (Figure 1) and differs from that used for the crystallized protein that contains some deletions [16]. The CeGluClα-ivermectin docking assay had a GlideScore energy of −8.4 kcal/mol and RMSD of 1.1 Å with respect to the crystal structure.
Docking assays of ivermectin and emamectin employing the CrGluClα model gave the Glide scores values reported in Table 1. Both drugs were seen to be inserted in between subunits and three H-bonds were observed between CrGluClα and ivermectin or emamectin (Table 1). Fig. 6A and B show the positioning of ivermectin and emamectin with respect to the receptor, which occurs between M3 on one subunit and M1 on the neighbouring subunit. Both drugs wedge deeply between M3 and M1, with an -OH group in their cyclohexene ring making H-bond contact with Thr305 of M2 (Figure 6C–F, see Figure S5 for drug structures and identity of groups involved in H-bonding). Two other H-bonds were observed: with the backbone carbonyl oxygen of Leu263 in M1 and with Thr318 at the extracellular end of M3. Residues participating in H-bonding in the CeGluClα-ivermectin crystalographic structure [16] were Leu279 in M1, Ser321 in M2 and Thr346 in M3. Of these, only the first is strictly conserved in CrGluClα (Leu263), with the further two positions occupied by a threonine (305) and an isoleucine (330), as shown in Figure 7. H-bonding at Leu263 and Thr305 of CrGluClα take place with the same positions of ivermectin as corresponding H-bonds at Leu279 and Ser321 in the CeGluClα structure. The Thr318 H-link of CrGluClα is not present in the C. elegans structure, where the corresponding position is taken by Ile334. This H-bond of CrGluClα occurs at the extracellular end of M3 and involves a hydroxyl (ivermectin) or an -NH-CH2 group (emamectin) in the disaccharide moieties of the drugs (Figure S5). This necessitates a bending upwards in the structures of the drugs to reach the position of Thr318 (Figs. 6C–F).
Molecular dynamics was used to evaluate the stability of the interactions between the protein and the ligands. Three different configurations were run: CrGluClα on its own, the CrGluClα-ivermectin and CrGluClα-emamectin complexes. Figure 6A and B show the positioning of the transmembrane domains and the fit of the drugs after 140 ns MD. Insertion of the drugs into the protein tended to separate helix M3 in one subunit from M1 in an adjacent subunit. It also appears that the top aspect of helix M2 moves away from the pore. These alterations are similar to those described for the crystal structure [16] and in MD studies of CeGluClα and other pentameric ligand-gated ion channels [31], [32]. Minimal diameters measured at the transmembrane portion of the pore at the end of these runs were in the order CrGluClα-ivermectin>CrGluClα-emamectin>CrGluClα (Figure 8A), and are consistent with an open channel configuration for the drug-receptor complexes. Points of narrowest pore diameter in the receptor without drugs occurred at Leu299 and P288 (Figure 8). These correspond to hydrophobic seal amino acids in CeGluClα that are conserved in CrGluClα A third narrowing of the pore (T308 in C. elegans) is absent in the Caligus receptor where it is replaced by Ala292. Also shown in Figure 8 are structures of the pore of CrGluClα in the absence of drugs at the beginning and the end of the MD simulation. There was a continuous water column at time zero (Figure 8C), that is interrupted by the advance of Leu299 towards the lumen of the pore at the end of the simulation (Figure 8B). The narrowing at P288 did not interrupt the water column.
Identification of the amino acids in the receptor involved in drug interaction considered residues at a distance <3.0 Å during at least 50% of the molecular dynamics time (averaged times of the five interacting drug molecules). According to these criteria 14 residues were involved in ivermectin (15 for emamectin) binding (see Figure 7), including H-bonding Thr305, Thr318, and Leu263, with the rest involved in van der Waals interactions. All five drug molecules remained at their binding sites during the 140 ns MD runs.
The preceding in silico analysis of ivermectin and emamectin interaction with CrGluCl revealed several characteristics also found in the crystal structure of CeGluClα bound to ivermectin but also some differences. The most salient divergence emerged in the H-bonds between the avermectins and the receptor. With only two of the three residues involved in CeGluCl conserved in the Caligus channel a third, not previously described H-bond acceptor emerged in the form of T318 at the extracellular end of M3. To obtain independent evidence that T318 is involved in the interaction of the avermectins with CrGluCl we mutated this residue to H-bonding-incompetent alanine. Figure 9 shows the result of CrGluClα-T318A expression in Xenopus oocytes that exhibited high spontaneous current with addition of glutamate evoking small increases in current and ivermectin inducing partial, irreversible current inhibition (Figure 9A). The current associated to CrGluClα-T318A expression is carried by Cl− ions. This is confirmed by its decrease upon [Cl−]o decrease and its blockade by PTX. Figure 9B shows average spontaneous currents in CrGluClα-T318A-expressing oocytes, as well as the increase in evoked by glutamate, and the current decrease observed upon addition of ivermectin or emamectin. Current-voltage relations for spontaneous CrGluClα-T318A activity and that remaining after reduction of extracellular Cl− concentration, after correction for that remaining in the presence of 100 µM PTX, are shown in Figure 9C. The reversal potential, and therefore the resting potential of CrGluClα-T318A-expressing oocytes was −18±1.9 mV (mean±SEM, n = 9). The effect of lowering [Cl−]o was to shift the reversal potential by an average ΔErev of 62±6.9 mV (mean±SEM, n = 3), not significantly different from the 69 mV expected for perfect Cl−-selectivity (P = 0.481, one-sample t-test). Average responses to increases in glutamate are shown in Figure 9D and compared to the response of WT CrGluClα. Maximal response was reached at lower glutamate concentrations in the CrGluClα-T318A mutant channel.
The data described suggest that CrGluCl-T318A mutant receptor is spontaneously open and is inhibited by emamectin and ivermectin. To study the efficiency of these inhibitory actions we decided to compare the effect of the avermectins in channels having reached what appears to be full opening under the effect of a maximally effective concentration of glutamate. Figure 10A and B show that either emamectin or ivermectin diminished the glutamate-dependent current in receptors activated through addition of 50 µM glutamate in a concentration-dependent manner, reaching a non-zero minimal current at saturating drug (66±9% and 59±8% of the initial current for emamectin and ivermectin respectively, n = 6 for both sets). Similar experiments performed using the CrGluClα-T318A mutant are seen in Figure 10C and D. Maximal effects of emamectin or ivermectin decreased the current activated by glutamate by 66±5 and 60±4% (means±SEM, n = 6 for both experimental sets) respectively. The concentration dependence of the effects of avermectins differed between WT and T318A mutant CrGluClα receptors as shown in the plots of normalised average responses to emamectin and ivermectin in Figure 10E and F. Fits of Hill decay functions to individual experiments gave half maximal inhibitory concentrations for emamectin of 78±22 and 244±31 nM in WT and T318A CrGluClα receptors respectively (means±SEM, n = 6 for both sets of experiments, t-test p = 0.007). Corresponding values for ivermectin inhibition were 30±3.6 and 340±91 nM (means±SEM, n = 6 for both sets of experiments, t-test p = 0.001). Values for nH were not significantly different from unity except for WT ivermectin data that gave an nH value of 1.9±0.4.
Parasitic nematodes, insects and crustaceans are pathogenic or act as disease vectors to man and other vertebrates. The study of the neurobiology of these invertebrates has been spurred by the search of compounds that interfere with their physiological processes and might therefore serve as chemotherapeutic agents for their control. The glutamate-gated chloride channels GluCls are neuronal or muscle receptors that have been described only in invertebrates. This makes them a very interesting pharmacological target for specific agents affecting parasites without interference with the physiology of their vertebrate hosts. Macrocyclic lactones are drugs that activate or modulate the activity of GluCls and have been widely and successfully used as insecticides and antiparasitic agents in human and veterinary medicine and in agriculture [14], [33].
One example of a devastating parasitic problem within an industrial context is the infestation of farmed salmonids with sea lice [3]. MLs ivermectin and, more recently, emamectin have been used successfully to fight these parasites both in the Northern (Norway, Canada, Scotland) and the Southern hemisphere (Chile), where the respective main species involved are Lepeophtherius salmonis and Caligus rogercresseyi. In both regions, however, ML-resistance has developed by underlying mechanisms that remain essentially unknown [17], [33]–[35]. Possible ways in which drug resistance can arise include a change in the pharmacological target causing the failure of the drug to bind or transduce its binding into the molecular effect, changes in drug metabolism or active removal the drug from the parasite, the host or both perhaps by drug-induced differential gene expression [5], [36].
By analogy with other species of nematodes and insects, it is thought that MLs act on sea louse through irreversible binding to γ-aminobutyric acid and GluCl channels, causing paralysis leading to death. Ivermectin resistance has been associated with mutations in GluCl leading reduced ML-sensitivity in C. elegans and D. Melanogaster [37]–[39]. This mechanism has not been explored in sea lice as GluCl channels have not been identified molecularly in these species.
An early attempt to obtain a cDNA for GluCl from L. salmonis yielded a fragment lacking a start codon [40] which, not surprisingly, failed to produce functional glutamate receptors [41]. The GluCl cDNA isolated here is predicted to result in a 461–464 protein with a putative signal peptide preceding the large extracellular N-terminus, two Cys-loops, four transmembrane domains and a shorter C-terminus. Our sequencing data analysis showed 12 different clones. All the differences were concentrated in the amino end and in the loop located between transmembrane domains 3 and 4, both highly variable regions in glutamate gated chloride channels (see blue arrows in Figure 2). These variable residues have not been reported as involved in glutamate or ivermectin response in other glutamate and ivermectin gated chloride channels so these differences were not predicted to cause changes in the possible functional properties of expressed proteins. Indeed clones 13 and 2 (see Figure 2), the most divergent sequences encountered, gave indistinguishable activities when assayed electrophysiologically after expression in Xenopus oocytes.
Closest homology of CrGluClα was with CeGluClα and DmGluClα subunits [11], [42]. Homology with the partial GluCl sequence of Northern hemisphere sea louse Lepeophtherius salmonis was also high [40]. Despite sequence homology there are differences in the functional characteristics of CrGluClα and those CeGluClα and DmGluClα. CrGluClα-expressing Xenopus oocytes show glutamate-dependent ion currents which, unlike those of DmClGluα [42], do not desensitize. No pretreatment was needed to elicit CrGluClα-mediated currents as is the case for those mediated by CeGluClα that does not show a glutamate response unless previously activated by ivermectin [43]. CrGluClα sensitivity to glutamate is similar to that of DmGluClα [42], with EC50 values of 7 and 23 µM respectively, but with the Caligus receptor showing little cooperativity (nH 1.3) compared with DmGluClα (nH 2.0). Concerning the effect of ivermectin, both CeGluClα and DmGluClα show activation with respective EC50 values of 140 and 40 nM [11], [42] compared with the 180 nM ivermectin EC50 of CrGluClα. The effect of ivermectin on CrGluClα was cooperative, with a Hill coefficient of 2.1. This was more reminiscent of that obtained with CeGluClα subunits than with the Drosophila α subunit, which shows little cooperativity. Judging by the potency of the ivermectin effect, the receptor of Caligus studied here falls within a group of Cys-loop anionic receptors showing the highest affinity ivermectin binding sites [44].
Emamectin benzoate is thought to act in a similar way to ivermectin and has become the chemotherapeutic drug of choice in the treatment of sea louse infestation in the salmon industry [33]. We are not aware, however, of any study of the effect of emamectin on GluCl receptors. Here we show that emamectin activates CrGluClα irreversibly with an EC50 of 202 nM and nH of 1.1. The similar potencies for emamectin and ivermectin action together with similarities in structure suggest also a similar mode of action.
Interestingly, the structure of CeGluClα has recently been obtained by X-ray crystallography [16] giving insights into the glutamate and ivermectin binding sites as well as on the site of action of pore inhibitor picrotoxin. The molecular structure of CeGluClα-ivermectin-glutamate reveals nine residues involved in glutamate binding and eight of these are identical in CrGluClα (Figure 1).
The crystal structure of CeGluClα also shows three residues H-bonding with ivermectin, only two of which are conserved in CrGluClα and are seen to H-bond with ivermectin and emamectin in docking assays. A third H-bonding residue in CeGluClα, namely Thr346 that H-links with the spiroketal moiety of ivermectin, is not conserved in CrGluClα. Instead, Thr318 in a region of M3 closer to the extracellular aspect of the channel in CrGluClα H-bonds with the disaccharide end of ivermectin or emamectin. Lack of conservation of CeGluClα Thr346 is also seen in a GluClα receptor from the parasitic nematode Haemonchus contortus [45] where the corresponding position is occupied by an alanine. Interestingly this receptor shares with CrGluClα the equivalent of Thr318 and responds to ivermectin in similar fashion as the Caligus receptor [46]. The H-bonding between T318 of CrGluClα and the disaccharide moiety of ivermectin and emamectin is novel and requires a bending upwards of the disaccharide moieties of the drug molecules that departs markedly from the topology of ivermectin seen in the crystal structure of CeGluClα. Mutation T318A led to receptors that were spontaneously open, were only slightly enhanced by glutamate and partially inhibited by emamectin or ivermectin. As we noticed in the WT receptor, glutamate and the MLs activate macroscopic currents of different amplitude, with the ML-activated currents significantly smaller than those activated by glutamate. Similarly, lower currents were obtained in the presence of MLs than with glutamate using the CrGluClα-T318A mutant. The potency of drug inhibition of open channels maximally activated by glutamate showed an increased affinity of the avermectins for their binding site in this configuration, perhaps suggesting that splaying of the M1–M3 helices facilitating drug access. Comparison of the EC50 values for ivermectin or emamectin inhibition of WT receptors fully activated by glutamate and those activated by mutation T318A shows that drug affinity was decreased by replacement with the H-bonding-incompetent Ala.
We have not yet a testable hypothesis of a possible mechanism for the T318A mutation-induced activation. We speculate that breaking a putative intramolecular H-bond in which T318 plays an acceptor role, either by the mutation or by action of the ML drugs, could lead to opening of CrGluClα receptor. Interestingly a disease-causing mutation at an equivalent site (V280M) in a glycine receptor leads to spontaneously active channels [47]. It is postulated that V280 might interact with residues in M1 to stabilize the closed state of the channel. Our results, although not entirely understood, point to the importance of T318 in the interaction of emamectin and ivermectin with the Caligus receptor. Perhaps future molecular simulations of the T318A receptor could clarify the role of this residue that appears to play an important part in the channel open-closed equilibrium.
Avermectin resistance through mutation in glutamate receptors has been described. A naturally occurring mutation (G323D) at M3 in a Tetranychus urticae GluCl (corresponding to Gly326 in CrGluClα) markedly increases abamectin LD50 [48]. Human GlyRα1 receptor has an alanine at the equivalent position (288) and shows low sensitivity to ivermectin, which becomes 50-fold higher in A288G-HsGlyRα1. The reciprocal mutation in HcGluα3B, G329A, converts this normally high ivermectin-sensitive receptor into a relatively resistant one [39]. This glycine residue, conserved in GluClα channels, lies at the M3 domain facing the drug binding cavity and it is conceivable that its mutation alters M3 flexibility and the interaction of avermectins with the receptor. This residue dubbed, M3-Gly, appears essential for high ivermectin affinity and is highly conserved in GluCl receptors [44].Mutation P299S, corresponding to P313 in CrGluClα, also markedly decreases the ability of ivermectin to activate DmGluClα [38]. The proline residue is located at the M2–M3 loop that is thought to be stabilised by ivermectin in a conformation favouring the open state [16]. Its mutation might interfere with channel opening by avermectins without a direct interaction with the drugs.
Van der Waals interactions were identified with twelve residues of CeGluClα, with nine of those identical or similar in the Caligus receptor (see Figure 7). Although not studied here in detail it appears that the mechanism whereby ivermectin, and emamectin, maintains CeGluClα open is by intercalating between subunits leading to the separation of helices M1 and M3 of neighbouring subunits [16]. As pointed out by Hibbs and Gouaux [16], the site defined by the binding of avermectins in GluClα receptors appears equivalent to that targeted in other pentameric ligand-gated ion channels by modulators such as alcohol and anesthetics [49]. Another structural characteristic of receptors defined has having high ivermectin sensitivity has been proposed to be the possibility of making numerous H-bonding interactions in the vicinity of S321 that was identified as H-binding in the crystal structure of the C. elegans channel [16]. These possible H-bond partners are M2 T318, S321 and N325, and Q320 and Q280 of M2, and M1 in the neighbouring subunit of the C. elegans receptor [44]. Their corresponding residues in the Caligus channel are S302, T305 (identified as H-bond partner in our calculations), N309/Q304/T264, thus conserving the capability for H-binding in this receptor and possibly contributing to explain its affinity for ivermectin and emamectin.
Our docking and MD calculations suggest that despite using the same general binding pocket, ivermectin and emamectin are seen at the Caligus receptor in a different conformation than that revealed by the crystal of the C. elegans receptor-ivermectin complex and form a different three-dimensional network of H-bonds [16]. This observation, in addition to requiring more work for its confirmation and better understanding, identifies a different type of drug binding site in CrGluClα that could admit new types of activity-modifying molecules.
We have presented here the first report of a full-length ionotropic glutamate-gated GluClα from the salmon parasite Caligus rogercresseyi. Our functional expression data strongly suggest that CrGluClα is an important target for avermectins used in the treatment parasitic infestation in salmon and trout and opens the way for ascertaining a possible mechanism of increasing resistance problems dogging aquaculture industry worldwide. Molecular modeling and docking assays as we demonstrate here could help in the design of new, more efficient drugs with functional expression of the receptor allowing a first stage of testing of their efficacy.
|
10.1371/journal.pgen.1002040 | Dynamic Regulation of H3K27 Trimethylation during Arabidopsis Differentiation | During growth of multicellular organisms, identities of stem cells and differentiated cells need to be maintained. Cell fate is epigenetically controlled by the conserved Polycomb-group (Pc-G) proteins that repress their target genes by catalyzing histone H3 lysine 27 trimethylation (H3K27me3). Although H3K27me3 is associated with mitotically stable gene repression, a large fraction of H3K27me3 target genes are tissue-specifically activated during differentiation processes. However, in plants it is currently unclear whether H3K27me3 is already present in undifferentiated cells and dynamically regulated to permit tissue-specific gene repression or activation. We used whole-genome tiling arrays to identify the H3K27me3 target genes in undifferentiated cells of the shoot apical meristem and in differentiated leaf cells. Hundreds of genes gain or lose H3K27me3 upon differentiation, demonstrating dynamic regulation of an epigenetic modification in plants. H3K27me3 is correlated with gene repression, and its release preferentially results in tissue-specific gene activation, both during differentiation and in Pc-G mutants. We further reveal meristem- and leaf-specific targeting of individual gene families including known but also likely novel regulators of differentiation and stem cell regulation. Interestingly, H3K27me3 directly represses only specific transcription factor families, but indirectly activates others through H3K27me3-mediated silencing of microRNA genes. Furthermore, H3K27me3 targeting of genes involved in biosynthesis, transport, perception, and signal transduction of the phytohormone auxin demonstrates control of an entire signaling pathway. Based on these and previous analyses, we propose that H3K27me3 is one of the major determinants of tissue-specific expression patterns in plants, which restricts expression of its direct targets and promotes gene expression indirectly by repressing miRNA genes.
| All organs and differentiated tissues in multicellular organisms are derived from undifferentiated pluripotent stem cells. The evolutionarily conserved Polycomb-group (Pc-G) proteins control stem cell identity and maintenance, likely by repressing genes involved in differentiation processes. Pc-G proteins are epigenetic regulators, thus they maintain stable expression states of their target genes through cell divisions that are not accompanied by changes in their DNA sequence. In this study, we asked whether Pc-G–mediated gene regulation is also dynamically regulated in plant development to confer stable, but flexible gene expression states that may switch in response to developmental or environmental cues. We therefore generated genome-wide maps of Pc-G activity of undifferentiated stem cell and differentiated leaf cell tissues which revealed dynamic regulation of Pc-G activity in plants. Pc-G activity is correlated with gene repression and its tissue-specific release results in local gene activation. Pc-G proteins target specific gene families in the two analyzed tissues, indicating a role for Pc-G proteins in balancing pluripotency and differentiation in plants. Based on our analyses, we propose that Pc-G activity not only permits long-term gene regulation but also has a more basic gene regulatory function in fine-tuning expression patterns of specific gene families during differentiation.
| Throughout their lifecycle, plants produce new organs through a group of undifferentiated cells which are maintained in structures called meristems. These pluripotent cells continuously divide and, in the shoot, cells in the periphery of the apical meristem differentiate to give rise to lateral organs like leaves or flowers. Stem cell maintenance is tightly controlled by numerous and interconnected pathways involving transcriptional regulation, phytohormones, microRNAs and epigenetic gene regulation (reviewed in [1]).
Epigenetic gene regulation is a key mechanism to confer stable, but reversible gene expression states. Epigenetics has been revealed as fundamental mechanism to maintain cell and tissue identity and to regulate stem cells and cancer.
Important epigenetic regulators of developmental processes are the Polycomb-group (Pc-G) and Trithorax-group (Trx-G) proteins which catalyze histone H3 lysine 27 trimethylation (H3K27me3) or H3K4me3, respectively (reviewed in [2], [3]). In Arabidopsis, around 5% of the canonical histone H3.1 is trimethylated at K27 [4]. Pc-G target genes in plants and animals are covered by H3K27me3 [5]–[8], a distribution that likely permits epigenetic inheritance of the modification [9]. Whereas Drosophila melanogaster Polycomb repressive complex 2 (PRC2) was shown to methylate H3K27 in vitro and in vivo via its histone methyltransferase subunit Enhancer of zeste [E(z)] [10]–[13], evidence for the biochemical activity of the conserved plant PRC2 is still lacking. H3K27me3 was shown to depend at least partially on plant PRC2 members as loss of the E(z) homolog CURLY LEAF (CLF) leads to reduced levels of H3K27me3 [8], [14]. In addition, immunofluorescence analyses of plants lacking the redundantly acting E(z) homologs CLF and SWINGER (SWN) revealed a reduction in euchromatic H3K27me3, but also a frequent re-distribution to chromocenters [15].
In plants, numerous developmental pathways are controlled by Pc-G proteins including seed development, flowering time, vernalization and organ identity (reviewed in [16], [17]). Loss of sporophytic Pc-G activity results in plants that show overproliferation and strong defects in organ identity [18], [19]. The severe phenotypes of Pc-G mutants indicate an essential role for Pc-G proteins in plant development which was strongly supported by whole genome analyses of H3K27me3 targets. In seedlings, more than 4000 H3K27me3 target genes were uncovered which are largely protein-coding genes and mostly exclude transposable element genes and heterochromatic regions [5], [20], [21]. A strikingly large number of developmentally important transcription factor genes showed H3K27me3 coverage [5]. However, the fact that key genes involved in the biosynthesis and inactivation of the phytohormone gibberellic acid exhibit enrichment in H3K27me3 [22] indicates an important role for Pc-G proteins in regulating developmental processes beyond the transcription factor level. Numerous genome-wide analyses of H3K27me3 and Pc-G protein binding have been performed for mammals and Drosophila [6], [23]–[29]. These studies revealed a considerably smaller number of target genes in Drosophila and mammals compared to Arabidopsis, but also key developmental transcription factor genes as H3K27me3 targets.
Presence of H3K27me3 is largely correlated with gene silencing in animals and plants, although H3K27me3 is only partially removed upon gene activation in Drosophila [30]. In Arabidopsis, H3K27me3 targets are enriched for genes with tissue-specific expression patterns or are induced by abiotic or biotic stresses suggesting that H3K27me3 is dynamically regulated in response to developmental or environmental cues [5]. Indeed, several thousand genes either lose or gain H3K27me3 when etiolated seedlings were transferred into light [22]. Furthermore, a novel approach of cell-type specific tagging and isolation of nuclei allowing cell type specific analyses of hair and non-hair cells of the Arabidopsis root epidermis revealed hundreds of genes carrying differential H3K4me3 and H3K27me3 in the two cell types and a strong correlation of gene repression with low H3K4me3 and high H3K27me3 [31].
Different cell types can harbor distinct chromatin profiles which is particularly the case for pluripotent mammalian embryonic stem (ES) cells [7], [32]. Many genes in ES cells carry bivalent marks, thus both H3K4me3 and H3K27me3, which are mostly resolved to monovalent states of either H3K4me3 or H3K27me3 during differentiation [7], [33]. In plants, however, it is currently unclear if undifferentiated cells harbor distinct chromatin profiles.
Here, we present genome-wide analyses of H3K27me3 as well as gene expression analyses from meristematic, undifferentiated cells and differentiated leaf cells using the vegetative shoot apical meristem as a model system. We confirm a large fraction of previously identified H3K27me3 targets and reveal several hundred additional, tissue-specifically methylated genes. Genes with strong expression differences in the two tissues are enriched for differential H3K27me3 suggesting that also in plants Pc-G proteins define gene ON/OFF expression states. We find that a large fraction of microRNA genes are differentially methylated in meristems and leaves and that their tissue-specific expression patterns are controlled by Pc-G proteins. In addition, a large number of genes involved in biosynthesis, transport, perception and signaling of the phytohormone auxin are among the identified target genes suggesting that entire gene regulatory networks are controlled and possibly stabilized by Pc-G mediated gene regulation. Collectively, our analyses suggest that Pc-G proteins control differentiation processes by conferring tissue-specific H3K27me3 of hundreds of genes including microRNA genes and genes involved hormonal pathways.
In Arabidopsis seedlings, several thousand genes are covered by H3K27me3 [5], [20].
It was previously suggested that H3K27me3 is required for stable gene repression throughout most of the plants lifecycle and is either reset to allow gene activation or acquired at later developmental stages to confer stable expression states during differentiation processes [5], [34]. In order to uncover dynamic regulation of H3K27me3 during development, undifferentiated meristematic (Me) and differentiated leaf (Le) tissues of clavata3 (clv3) mutant plants were analysed. Meristematic tissue and stem cells can be easily isolated from clv3 mutants by manual dissection as the mutant harbours larger vegetative shoot apical meristems and increased stem cell numbers [35] (Figure 1A). Importantly, leaf development is not affected by loss of CLV3. Both leaf and meristematic tissue samples were subjected to array based genome wide expression (Figure 1) and H3K27me3 (Figure 2) profiling. Therefore, this approach permits the identification of H3K27me3 targets that gain or lose H3K27me3 during differentiation of meristematic cells. In addition, the associated changes in gene expression of H3K27me3 targets can be revealed.
The expression analysis verified both sample identity and specific enrichment of each dissected tissue because typical transcripts characteristic for the meristematic domain or green tissue were detected as highly and differentially expressed (Figure 1C). Meristem and stem cell identity genes like SHOOTMERISTEMLESS (STM), CLV3 and WUSCHEL (WUS) were highly expressed in the meristematic sample whereas photosynthesis-related genes encoding for subunits of RUBISCO (RBCS-1B) or light harvesting complexes (LHCB1.4) were exclusively detected in leaves.
We generated genome-wide H3K27me3 profiles for each tissue by hybridizing sheared genomic DNA that was immunoprecipitated with an antibody detecting H3K27me3 to high density tiling arrays, as was previously described [36], [37]. We identified a total of 9006 H3K27me3 target genes in the two tissues which include more than 7400 protein coding genes (27,6% of all annotated Arabidopsis protein coding genes) and interestingly also 74 miRNA genes (43% of all miRNA genes) (Figure 2). Importantly, more than 80% of protein coding H3K27me3 target genes identified in our study were also revealed in previous analyses of young seedlings [5], [20], demonstrating the validity of our approach. Both former studies were performed with 14 days old wildtype plants, whole seedlings or their aerial parts, whereas the tissue samples analyzed in our study were derived from 9 weeks old clv3 mutant plants. The large H3K27me3 overlap in the different tissues therefore suggests extensive and stable gene silencing by Pc-G mediated H3K27me3 throughout vegetative plant development. In addition to a large number of common H3K27me3 targets, the analysis identified many, presumably tissue-specific protein coding target genes (1001 genes only detected in seedlings (by Oh et al. [20]); 1230 genes only identified in meristem or young leaf tissue in our study) (Figure 2A). Interestingly, although previous studies showed that transposable elements (TE) are largely devoid of H3K27me3 [5], [20], we detected a high number of TEs as meristem-specific H3K27me3 targets (Figure 2B). This suggests an unexpected role for Pc-G proteins in the regulation of TEs specifically in the stem cell harbouring meristem.
To study the role of H3K27me3 in controlling developmentally important genes we focused our analyses on protein coding and miRNA genes and compared the methylation profiles of the individual tissues (Figure 2B, 2C). This uncovered large differences between both samples as nearly 2000 protein coding and 27 miRNA genes were exclusively methylated in one sample (Me or Le) revealing these as tissue-specific H3K27me3 targets (details for all annotations are shown in Table S1). Interestingly, in contrast to the largely meristem-specific H3K27me3 targeting of TEs, a larger set of protein coding and miRNA H3K27me3 targets was identified in the leaf compared to the meristematic sample (Figure 2B). Thus, heterochromatic and euchromatic loci may be differentially targeted by Pc-G proteins in undifferentiated and differentiated tissues.
For further analyses on the comparison of H3K27me3 presence and gene expression and on gene families we isolated genes harbouring defined methylation levels and performed conservative statistical analyses to group these into sets of equally (M = L) and differentially methylated genes [meristem- (dM+) and leaf- (dL+) specific] (see methods). These analyses uncovered 1519 protein coding genes showing equal H3K27me3 levels in both samples, 158 genes specifically methylated in the meristem and 457 genes methylated in leaves (Figure 2C).
Collectively, our analyses uncover dynamic regulation of a large number of H3K27me3 target genes during differentiation which involves acquisition and removal of H3K27me3 at protein coding, miRNA and TE genes.
We sought to confirm the genome-wide, tissue-specific H3K27me3 patterns by independent ChIP experiments on meristems and leaves of clv3 mutants (Figure S1) and wildtype (Figure 3). The transcription factors KNOTTED-LIKE FROM ARABIDOPSIS THALIANA2 (KNAT2) and KNAT6 are specifically expressed in the meristem ([38], (Figure 3)) and were identified as leaf specifically methylated (dL+) (Figure S1A representatively shows the KNAT6 locus). H3K27me3 ChIP-qPCR analysis of independently dissected clv3 tissues confirmed low meristematic and high leaf H3K27me3 levels for KNAT2 and KNAT6 and revealed an antagonistic pattern of the active mark H3K4me3 (Figure S1B, S1C). Thus, H3K4me3 and H3K27me3 distribution are consistent with meristem specific expression of both KNAT genes (Figure 3E). To further prove the significance of differential H3K27me3 identified by ChIP-chip on clv3 mutants (Figure 2) we performed ChIP-qPCR analyses of 2 months old dissected wildtype plants, grown under short day conditions. Consistent with the data on clv3, KNAT2 and KNAT6 showed leaf specific H3K27me3 (Figure 3). In addition, equal methylation for the non-expressed genes AGAMOUS and FUSCA3 and differential methylation of several transcription factor genes, PIN-FORMED (PIN) auxin efflux carriers and microRNA genes were confirmed (Figure 3). For all genes tested by ChIP, moderate to strong differential expression was detected which was highest in the tissues lacking H3K27me3 (Figure 3E). Thus, these analyses strongly suggest that the clv3 ChIP-chip results are transferable to wild type plants.
Since we had identified tissue-specific H3K27me3 target genes we asked whether differential H3K27me3 is correlated with differential gene expression. In total, more than 22000 protein coding genes (85% of the genome) were expressed in at least one of the tissues. A large fraction (nearly 9500 genes) showed equal expression in both samples [X(M = L)], whereas 2.833 were at least four times higher expressed in one sample compared to the other [1.865 higher in the meristem (dXM≥2, Δlog2) and 968 higher in leaves (dXL≤−2, Δlog2)] (Figure 4A, Figure S2). We next binned genes with similar expression differences in the two tissues (Figure 4B, Table S2). These gene sets were analyzed for enrichment of H3K27me3 targets (Figure 4B) and of genes harbouring equal or differential H3K27me3 levels (M = L, dM+, dL+) (Figure 4C). In addition, we performed the reciprocal analyses (Figure 4D). Genes that were equally expressed both in the meristem and the leaves (X(M = L)) showed a significant underrepresentation of H3K27me3 targets (Figure 4B). In contrast, methylated genes were more frequent in the groups of non expressed and differentially expressed genes (Figure 4B). Interestingly, the relative abundance of target genes increased with increasing tissue specific expression in the meristem (Me+) or leaves (Le+) (Figure 4B). We next asked how differential expression relates to differential or equal H3K27me3 (Figure 4C). This detailed tissue specific analysis clearly revealed that the highly specific expression in one tissue (dXM≥2 or dXL≤−2) was correlated with a depletion of tissue specific methylation (dM+ or dL+) in the same sample and a strong enrichment of repressive histone methylation in the other – non expressing – sample. Importantly, the stronger the differences in gene expression levels between the two tissues, the more genes were identified which showed antagonistic, differential H3K27me3 (Figure 4C). Genes showing equal H3K27me3 levels in both tissues were enriched for non-expressed genes which was not the case for differentially methylated genes, revealing that most genes carrying differential methylation are expressed in at least one of the tissues (Figure 4C, 4D). We found the same correlations when the analyses were performed with the different subsets of H3K27me3 covered genes (Figure 4D). Leaf-specifically expressed genes (dXL≤−2) were more frequent in the set of meristem specifically methylated genes (dM+) and vice versa whereas genes showing equal, but detectable methylation (M = L) were depleted of expressed genes.
In whole genome analyses of H3K27me3 in seedlings, Oh et al. [20] identified 1001 H3K27me3 targets which were not revealed in our analyses (Figure 2A). 824 of these are expressed in at least one of the tissues we analyzed (data not shown), suggesting that these targets are seedling-specifically methylated and are subjected to dynamic changes of H3K27me3 in development.
Previous analyses of H3K27me3 target genes in seedlings identified transcription factors as a major class of H3K27me3 target genes [5]. To reveal whether all transcription factor families and other gene families are preferentially targeted by H3K27me3 we analyzed our datasets generated from leaves and meristems.
We first exploited the resources of the Gene Ontology (GO) annotations from TAIR (http://www.arabidopsis.org/) providing functional categorizations of Arabidopsis protein coding genes (Figure S3). As expected the molecular function “transcription factor activity” was enriched in H3K27me3 target genes, whereas general cellular housekeeping functions like “DNA or RNA metabolism” and “electron transport or energy pathways” were depleted. Importantly, also tissue-specific differences were uncovered: a significant depletion of leaf-specific H3K27me3 target genes was observed in the cellular component “plastid”, consistent with the restriction of photosynthesis to leaves. In addition, genes associated with “kinase activity” and “transporter activity” were revealed as preferential H3K27me3 targets in the meristematic sample, suggesting specific repression of these sets of genes in undifferentiated cells.
We further dissected the H3K27me3 target genes by detailed analysis of gene families with a focus on transcription factor, enzyme and transporter gene families as these were enriched in the broad GO term analyses of H3K27me3 targets (Figure 5; Figure S4). Gene families involved in fundamental cellular processes like DNA replication, cell cycle regulators and protein phosphorylation/dephosphorylation were significantly depleted in H3K27me3 target genes (Figure S4; Table S3). Surprisingly, several transcription factor families were largely not regulated by H3K27me3 including the AUXIN RESPONSE FACTOR (ARF), CCAAT-HAP2 (HAP2) and SQUAMOSA-PROMOTER BINDING LIKE (SPL) gene families (Figure 5). Other transcription factor gene families showed a strong bias towards tissue-specific targeting by H3K27me3 (dM+; dL+) suggesting a specific regulation in the meristem or leaves. Leaf-specific H3K27me3 target genes comprise known regulators of meristem function like HOMEOBOX transcription factors and several other gene families (e.g. IAA and DOF transcription factors), while TCP, CONSTANS-like and GRAS-transcription factors were targeted by H3K27me3 specifically in the meristem indicating a role for theses gene families in differentiation processes or leaf development. A surprisingly large set of enzyme and transporter gene families were uncovered as H3K27me3 targets which are involved in differentiation or developmental processes (e.g. cell wall biosynthesis), hormone biosynthesis (YUCCA flavin monooxygenases, CYTOCHROME P450s) or transport of photosynthetic products (sucrose transporters) and hormones (PINFORMED (PIN)-like auxin efflux carriers) (Figure S4, Table S3). Previous analysis had identified the actin regulator formin ARABIDOPSIS FORMIN HOMOLOGUE 5 (AtFH5) as Pc-G target gene whose mis-expression is partially responsible for Pc-G mutant seed phenotypes, revealing that regulation of enzymatic genes is an important function of Pc-G proteins [39].
Lastly, we studied whether specific transposable element gene families were overrepresented in the different tissues. We generally observed a significantly larger amount of retrotransposons compared to DNA transposons as H3K27me3 target genes (Figure S4H, Table S3). Also tissue-specific differences were identified among retrotransposon gene families. Whereas gypsy-like retrotransposons were significantly enriched in the meristematic fraction, copia-like retrotransposons were overrepresented in leaves and athila-like retrotransposons almost completely non-targeted by H3K27me3 in leaves (Figure S4H).
Collectively, these analyses uncovered H3K27me3 targeting of a diverse set of gene families which are involved in biosynthesis and transport of secondary metabolites and transcriptional regulation and ultimately in cellular differentiation. These sets of genes include many known regulators of meristem or leaf development and provide an important resource for the identification of additional, novel factors involved in differentiation or stem cell regulation. In addition, also specific transposable element gene families are tissue-specifically targeted by H3K27me3 suggesting differential regulation of TE genes in the different tissues.
During the analyses of H3K27me3 targets we found that a large number of miRNA and several trans-acting siRNA (tasiRNA) genes carried H3K27me3 which was, to our knowledge, not previously revealed (Figure 2). These regulatory RNAs display differential methylation in the same way as protein coding genes, thus many show leaf-specific H3K27me3, whereas only few carry H3K27me3 exclusively in the meristem (Figure 2C). Seventy four of the annotated 174 miRNA loci were identified in our analysis and we revealed additional 12 that were detected, but not described by previous studies [5], [20]. Thus, nearly 50% of all microRNA genes are covered by H3K27me3 (Figure 2, Table 1). A high number of miRNAs (25 of 103 unique miRNAs) are encoded in multiple loci which constitute 96 of all 174 miRNA loci (Arabidopsis Small RNA Project (http://asrp.cgrb.oregonstate.edu/) [40]). Multiple and single loci miRNAs are regulated by H3K27me3 but miRNAs encoded in multiple loci are heavily targeted (72%) whereas only 22% of miRNAs present at a single locus are H3K27me3 targets.
Currently 206 genes including many developmentally important transcription factors are known which are negatively regulated by miRNAs at a post-transcriptional level. Interestingly, genes that are regulated by a H3K27me3 covered miRNA genes less frequently carry H3K27me3 compared to genes whose expression is controlled by a miRNA gene devoid of H3K27me3 (Table 1). This bias is even more apparent when specific miRNA and tasiRNA gene families and their targets are analyzed (Table 2): whereas most miRNA169, miRNA156/157, miRNA167 and tasiRNA3 loci carry H3K27me3, their target genes (HAP2, SPL or ARF genes) are largely devoid of H3K27me3. The lack of H3K27me3 targeting of these transcription factor genes is in contrast to the generally high enrichment of transcription factors as H3K27me3 targets (Figure 5; compare HAP2, SPL and ARF families to average enrichment in H3K27me3 targets of all transcription factor genes, p<0.05 (χ2-test) for HAP2 and ARF genes (Table S3)). However, H3K27me3-mediated repression of the miRNA genes likely permits an indirect positive regulation of these specific transcription factor families by Pc-G proteins. Interestingly, for other gene families (e.g. TCP and HD-ZIP transcription factors) both miRNA genes and their targets carry H3K27me3, suggesting that expression of these transcription factors is regulated transcriptionally by Pc-G proteins and post-transcriptionally by miRNAs.
Our in-depth gene family analyses revealed enrichment of specific biosynthetic enzymes, transporters and transcription factors (Figure 5). Therefore, we inquired whether H3K27me3 targets entire developmental pathways which involve local biosynthesis of a signaling molecule, its transport, perception and signal transduction. We focused on the phytohormone auxin as we initially realized that the PIN auxin efflux carriers are largely H3K27me3 targets (Figure S4). Several pathways and gene families (YUCCA monooxygenases, CYTOCHROME P450s and TRYPTOPHANE AMINO TRANSFERASES) participate in the biosynthesis of the phytohormone auxin from tryptophan (reviewed in [41], [42]). Auxin is transported by several different carrier proteins in the plant. The AUXIN RESISTANT1 (AUX1)/LIKE AUX1 (LAX) gene family enables auxin influx and PIN- and several ATP-BINDING CASSETTE (ABC) transporters mediate auxin efflux (reviewed in [43]). The analysis of gene families uncovered that most of these gene families are enriched for H3K27me3 targets (Figure 5, Figure S4), including the genes for which a role in auxin regulation was shown (Table 3, Table 4). Interestingly, although the auxin receptor gene TRANSPORT INHIBITOR RESPONSE 1 (TIR1) and its homologues are largely devoid of H3K27me3, miRNA393a which negatively regulates these is targeted by H3K27me3 in leaves (Table 4). Similarly, the ARF transcription factor genes which are involved in repressing auxin inducible genes (reviewed in [44]) are depleted in H3K27me3, however, miRNAs and tasiRNAs controlling ARF expression levels are largely targeted by H3K27me3 (Figure 5, Table 2, Table 4). On the other hand, IAA class of transcription factors which positively regulate auxin responsive genes are highly enriched for H3K27me3 targets, similar to other transcription factor families. Thus, auxin biosynthesis, transport, perception and transcriptional responses are controlled by H3K27me3, both by direct targeting of the regulatory genes or by indirect regulation through miRNAs. The analyses of the auxin pathway therefore reveal that not only specific gene families but also entire pathways involving diverse gene families can be controlled by H3K27me3.
Only selected plant genes expressed in vegetative tissues were previously analyzed for an overlap of H3K27me3 and Pc-G protein binding and dependence of H3K27me3 on Pc-G proteins [8], [14], [45], [46]. Although immunofluorescence analyses on the severe clf/swn mutants revealed a reduction in euchromatic H3K27me3 [15], it is currently unclear whether the H3K27me3 is completely dependent on Pc-G proteins. To reveal whether presence of H3K27me3 is likely corresponding to Pc-G binding and regulation we used immunoblot analyses to study histone methylation levels in various PRC2 mutants. We used mutants that are partially or completely deficient for Arabidopsis homologs of Drosophila E(z) [clf-28 (null allele), swn-7 (null allele)] or Suppressor of zeste 12 [vernalization2-1 (vrn2-1) (null allele), embryonic flower2–10 (emf2–10) (weak allele of emf2)] [18], [47]. We studied single and double mutants of these as they show different degrees of reduction in somatic Pc-G activity: moderate reduction in clf-28 and emf2–10 more severe reduction in vrn2/emf2–10 double mutants and complete loss of somatic Pc-G activity in the clf/swn double mutant (Figure 6) [18]. H3K27me3 was completely lost in the clf/swn mutants, strongly reduced in vrn2/emf2-10 mutants and only mildly affected in clf or emf2–10 mutants (Figure 6). Loss of H3K27me3 in clf/swn was correlated with a strong increase in H3K4me3. H3K27me1, a modification preferentially present in heterochromatin, was not affected in any of the mutants whereas H3K27me2 showed a clear reduction in vrn2/emf2–10, but increase in clf/swn mutants (Figure S5). Thus, our study of histone methylation in Pc-G mutants strongly suggests that all H3K27me3 is catalyzed by PRC2 proteins and that a possible re-distribution to heterochromatic regions [15] involves only marginal amounts of H3K27me3.
Since the global analysis of H3K27me3 abundance revealed a strong reduction even in the vrn2/emf2–10 double mutant showing relatively mild morphological alterations (Figure 6A) we were interested if a H3K27me3 reduction occurs at most genes or gene-specifically. We therefore studied H3K27me3 of selected genes in leaves from wildtype and vrn2/emf2–10 plants by ChIP-qPCR (Figure S6). This analysis identified three different classes: STM and FUS3 displayed no difference in H3K27me3 in the mutants, several genes including KNAT2 or PIN8 had intermediate H3K27me3 levels and most analyzed genes (e.g. PIN1 and PIN6) showed a strong reduction or a complete loss in the mutants. Thus, the observed global reduction in H3K27me3 is rather caused by the loss of H3K27me3 at some loci than by an equal reduction of H3K27me3 at all target loci (Figure S6).
To study possible changes in gene expression resulting from a reduction in H3K27me3, we performed qRT-PCR based expression analyses of clf/swn and vrn2/emf2–10 double mutants (Figure 7, Figure 8). Clf/swn mutants show complete loss of H3K27me3, do not maintain organ identity post-embryonically and are therefore only comparable to wildtype shortly after germination. However, vrn2/emf2–10 mutants show strong leaf serration especially in older leaves, but produce leaves and flowers (Figure 6), thus tissue-specific changes in gene expression can also be analyzed at later developmental stages. In 9 d old clf/swn seedlings, we observed mis-expression of several previously analysed H3K27me3 targets [8], [48], [49], ranging from mild (KNAT2) to strong (KNAT6, STM and AG) mis-expression (Figure 7A). Surprisingly, for TCP2, TCP4, TCP10 and SPL3 we observed a strong down-regulation in clf/swn mutants although they likely lost H3K27me3 (Figure 7B). However, TCP2, TCP4 TCP10 are negatively regulated by miRNA319 [50] which is targeted by H3K27me3 (Figure 3). Consistently, we observed mis-expression of the miRNA319a precursor in clf/swn mutants which is likely responsible for post-transcriptional down-regulation of TCP2, 4 and 10 in the Pc-G mutant. Expression of TCP5 which is not targeted by miRNA319 is not affected in the mutant. Similarly, SPL3 whose abundance is regulated by the H3K27me3 targeted miRNA156 showed reduced expression in clf/swn mutants (Figure 7B).
H3K27me3 targets 6 of the 8 PIN gene members (Table 3), we therefore also analysed their expression in clf/swn and vrn2/emf2–10 mutants. Except for PIN1, none of the PIN genes showed upregulation in the mutants (Figure 7A, Figure S6). PIN3, PIN4 and PIN7 which are all not targeted by H3K27me3 in leaves even showed a strong down-regulation of expression in clf/swn mutants.
As we also revealed H3K27me3 targeting of several TE genes, we studied expression of several H3K27me3 targeted TE genes. We observed upregulation of two TE genes in the clf/swn mutant (Figure 8), suggesting that Pc-G proteins and H3K27me3 can prevent expression of TE genes, which is consistent with a previous study revealing Pc-G mediated silencing of a TE gene in the endosperm [51]. To study whether interference with DNA methylation can further enhance activation of TE genes, we grew wildtype and clf/swn plants on media containing the DNA methyltransferase inhibitor zebularine [52] and analysed TE activation (Figure 8). The tested TE genes were neither activated by zebularine in wildtype nor further transcriptionally activated in clf/swn suggesting DNA methylation independent regulation of the TE genes which are activated in the Pc-G mutants.
Lastly, we investigated whether Pc-G mutants control tissue-specific expression of differentially methylated H3K27me3 targets. Mir167a and 167d genes show higher levels of H3K27me3 in the meristem compared to leaves (Figure 3). Reporter transgenes in which β-GLUCURONIDASE (GUS) was fused to the promoters of mir167a and mir167d [53] revealed preferential expression of the miRNA genes in cotyledons, but exclusion from the meristem, consistent with the H3K27me3 patterns (Figure 7, Figure S8). In clf/swn mutants, the expression domains of mir167a and mir167d reporters were expanded to the shoot apical meristem, the petioles, the main root and the base of the hypocotyl (Figure 7F, Figure S8). The genes ARF6 and ARF8 genes are regulated by mir167 which is in agreement with their mutually exclusive expression domains [53] (Figure 7E, 7G). Consistent with mis-expression of mir167 in clf/swn mutants we observed loss of shoot meristematic expression of a pARF6::ARF6-GUS reporter gene construct that harbours the miRNA binding site (Figure 7H). This further demonstrates the expansion of the miRNA expression domain to the meristem.
To uncover tissue-specific mis-expression of differentially methylated H3K27me3 targets in the vrn2/emf2–10 mutant we dissected plants and analyzed gene expression in meristematic tissue, young leaves (up to 3 mm blade length) and older leaves (3–10 mm). In wildtype, KNAT2, KNAT6, STM and PIN1 are strongly expressed in the meristem and carry H3K27me3 preferentially in leaves (Figure 3, Figure 7). Expression of these genes was not altered in meristems of vrn2/emf2–10 double mutants as expected (Figure 7C). In older leaves of the mutant, however, the KNAT genes were activated which correlates with a reduction in H3K27me3 (Figure 7D, Figure S6). STM displayed only moderate mis-expression in leaves, consistent with similar levels of H3K27me3 in wildtype and vrn2/emf2–10 mutants. Although H3K27me3 of PIN1 is depleted in vrn2/emf2-10 mutant leaves the gene is only slightly activated in the mutant (Figure 7, Figure S6). For several H3K27me3 target genes exhibiting differential methylation we revealed tissue-specific mis-expression in Pc-G mutants indicating that Pc-G proteins restrict expression of their targets by depositing tissue-specific H3K27me3. We also showed that Pc-G proteins transcriptionally regulate miRNA consistent with an up-regulation of miRNA genes and concomitant down-regulation of miRNA target genes in Pc-G mutants. In addition, although no global effect on DNA methylation and TE activation was revealed in clf/swn mutants [15], a subset of TE genes may be regulated by Pc-G proteins (Figure 8 and [51]).
In this study, we analyzed the dynamic regulation of the PRC2-mediated modification H3K27me3 in undifferentiated meristematic and differentiated leaf tissue. Our analyses revealed differential methylation of hundreds of H3K27me3 target genes including protein coding, transposable element and microRNA genes and expand previous observations that plant Pc-G proteins have important roles in controlling tissue-specific expression patterns of gene families and regulatory networks.
To study the dynamics of H3K27me3 in Arabidopsis development and its function in tissue differentiation we focused on the vegetative shoot apical meristem which harbors continuously dividing stem cells and gives rise to terminally differentiated organs (leaves) and also differentiates into the flower producing inflorescence meristem. Manual dissection of meristems and young leaves was performed on clv3 mutants as these have strongly enlarged meristems and increased stem cell numbers [35], but show no obvious defects in leaf development. Independent analyses on dissected wildtype meristems and leaves confirmed differential H3K27me3 of all tested genes (Figure 3). Furthermore, a large overlap of H3K27me3 targets with previously published data on seedlings [5], [20] revealed a similar set of H3K27me3 targets both in wildtype seedlings and clv3 mutant leaves and meristems. Thus, overall the experimental system appeared highly suitable to study the dynamics of H3K27me3 on a whole genome level during differentiation.
Our analyses identified several hundred protein coding genes, transposable elements and microRNA genes exhibiting differential H3K27me3 patterns between meristematic and leaf tissue. Genes that acquired H3K27me3 during differentiation included several meristematic regulators like STM or KNAT2 and 6, but also many previously non-characterized genes which may present novel regulators of meristem regulation and maintenance. Leaf tissue is developmentally derived from continuously dividing cells in the shoot tip which differentiate in the periphery of the meristem. Thus, H3K27me3 at the subset of genes showing meristem-specific H3K27me3 must be actively or passively removed during differentiation. H3K27me3 demethylases have not been identified in plants yet, however, our analyses strongly suggest the contribution of these enzymatic activities to confer dynamic regulation of H3K27me3. Charron et al [22] observed differential H3K27me3 patterns of protein coding and transposable element genes when dark grown seedlings were compared with dark grown seedlings shifted to light. In addition, dynamic changes in H3K27me3 and/or Pc-G protein binding were also unveiled both during Drosophila development and differentiation of mammalian embryonic stem cells suggesting conserved mechanisms which confer developmental dynamics of H3K27me3 [23], [28], [32], [54].
We confirmed that H3K27me3 is largely excluded from heterochomatic regions and transposable element (TE) genes, especially in leaf tissue (only 4% of all targets) [5], [20]. Interestingly, more than 15% of H3K27me3 targets detected in the meristem were TE genes (Figure 2). H3K27me3 detection at TEs was likely not due to cross-reaction of the antibody to H3K27me1 which is found in heterochromatin, because the same anti-H3K27me3 antibody was used for both analyzed tissues. In addition, the antibody was not detecting histones in clf/swn mutants which lacked all H3K27me3 but showed no alterations in H3K27me1 (Figure 6, Figure S5). Transposable elements are usually repressed throughout sporophytic development, thus loss of H3K27me3 in leaves likely does not result in transcriptional activation. The stem cells are required for plant germline formation and therefore must be particularly protected from transposable element activation. Thus, H3K27me3 may represent an additional, meristem-specific silencing mechanism besides the canonical heterochromatic marks H3K9me2 and H3K27me1 [55], [56]. Also in mammalian embryonic stem cells, transposable elements are marked by a specific set of repressive histone modifications, suggesting that pluripotent cells are protected by a large array of modifications [32]. For a few TE genes we observed up-regulation in the clf/swn Pc-G mutant, consistent with a previous report showing up-regulation of a TE gene in Pc-G mutant endosperm [51], suggesting that Pc-G proteins contribute to the regulation of TEs.
Our data revealed that non-expressed genes are preferential H3K27me3 target genes (Figure 4) as previously reported [5], [20], [21], [31]. Interestingly, protein coding genes which showed strong expression differences between meristem and leaf are enriched for H3K27me3 targets. In addition, differential H3K27me3 was anti-correlated with tissue-specific gene expression patterns (Figure 4). Thus, similar to Drosophila, Pc-G proteins likely control ON/OFF expression states of their targets in Arabidopsis [30]. However, tissue-specific H3K27me3 is not the only determinant which generates tissue-specific expression patterns as many genes showing differential expression are not covered by H3K27me3.
Previous analyses of H3K27me3 identified transcription factors as one major class of H3K27me3 targets, but preference for certain families was not unveiled [5]. We confirmed this finding as most transcription factor families including MADS-, WOX-, HOMEOBOX- and YABBY-transcription factors were preferentially targeted by H3K27me3. However, others including SPL-, ARF- and HAP2-transcription factor genes were largely devoid of H3K27me3, indicating that transcription factor genes are not per se enriched for H3K27me3 (see also below). In addition, certain transporter gene families (e.g. sucrose transporters, auxin carriers) and gene families involved in biosynthetic pathways (peroxidases, cytochrome P450 genes) were among the H3K27me3 targets. In Drosophila and mammals, transporter and biosynthetic genes (hydrolases, cytochrome P450 genes) were also revealed as preferential H3K27me3 targets in addition to transcription factor genes [6], [23], [26], [29]. This strongly suggests an unexpected level of conservation not only of Pc-G proteins but also of their target gene families between animals and plants.
We also identified gene families and functional categories showing preferential H3K27me3 targeting in the meristem (e.g. TCP transcription factor genes, Cytochrome P450 genes; functional categories “plastid” and “kinase activity”) or in leaves (e.g. MYB and HOMEOBOX transcription factor genes). In particular the differential methylation of HOMEOBOX transcription factor genes confirmed the validity of our approach as these are repressed in leaves by Pc-G proteins ([48] and this study).
Thus, our in-depth analyses of H3K27me3 targets confirmed specific gene families as preferential H3K27me3 targets, but also identified novel pathways and gene families that are likely controlled by Pc-G proteins and epigenetic gene regulation.
Our study discovered a large fraction of microRNA genes as H3K27me3 targets many of which exhibited tissue-specific differences in H3K27me3 (Figure 2, Figure 3, Table 1, Table 2). In addition, we uncovered mis-expression of several miRNA genes and down-regulation of their target genes in Pc-G mutants (Figure 7). Although miRNAs largely post-transcriptionally regulate their target genes, some miRNAs recruit the DNA methylation machinery to their target genes resulting in transcriptional repression [57]–[59]. Similarly to H3K27me3, miRNA genes target a multitude of transcription factor genes [60], [61], raising the possibility that miRNAs mediate recruitment of Pc-G proteins to some of their targets. Our analyses suggest that this is likely a rare phenomenon as transcription factor genes which are miRNA targets, are largely not targeted by H3K27me3 (Table 1, Table 2), despite the general enrichment of transcription factors as H3K27me3 targets ([5] and this study). Nonetheless, we also identified gene pairs, for which both the miRNA gene (e.g. mir319a) and their targets (e.g. TCP2, TCP10) displayed differential, but antagonistic H3K27me3 patterns. Thus, for a subset of genes miRNA-mediated targeting of H3K27me3 is an attractive possibility.
In summary, our studies provide evidence that Pc-G proteins and H3K27me3 do not only negatively regulate expression of transcription factor genes by direct association, but also positively control their expression by restricting expression of miRNA genes (Figure 9). This regulation appears to be particularly important in meristematic and leaf tissues as a large fraction of miRNA genes is differentially methylated in these two tissues (Figure 2). Regulation of miRNA genes by Pc-G proteins is likely a conserved mechanism. Recent analyses of miRNA gene regulation during mouse and human lymphopoiesis and genome-wide profiling of Pc-G protein binding sites in Drosophila uncovered H3K27me3-mediated control of miRNA gene expression [62], [63]. In addition, more than 30 miRNA loci were discovered as H3K27me3 targets in human embryonic stem cells [26].
Organ-specific auxin maxima are generated by both local auxin biosynthesis and auxin transport, which is largely mediated by the PIN efflux carrier proteins (reviewed in [43], [64], [65]). To date, three different pathways for the synthesis of auxin from tryptophan have been identified or proposed for plants (reviewed in [42]). Our analyses identified most of the key regulators of auxin transport and biosynthesis to be H3K27me3 targets several of which are leaf-specifically methylated (Figure 3, Figure 5, Table 3, Table 4). Additionally, our results suggest that auxin perception is controlled by Pc-G proteins as H3K27me3 targets the mir393a gene which in turn regulates the auxin receptor genes TIR1/AFB (Table 4) [66], [67]. In addition, H3K27me3 regulates the transcriptional responses to auxin as half of the AUX/IAA repressor genes are covered by H3K27me3. Although only one of the 23 ARF transcriptional activator genes carries H3K27me3, all known miRNAs or tasiRNAs that regulate a total of 8 ARF genes are H3K27me3 targets (Table 2). Thus, we propose that auxin responsive genes are largely suppressed in Pc-G mutants as H3K27me3 restricts expression of the AUX/IAA repressors and promotes expression of the ARF activators by controlling the expression of ARF-regulating miRNAs (Figure 7, Figure 9, Table 2). Consistent with this hypothesis, we observed down-regulation of several auxin-inducible PIN genes (PIN3, PIN4, PIN7) [68] in clf/swn mutants (Figure S7). In addition, the Pc-G mutant terminal flower2 was recently shown to have lower expression of auxin responsive genes including the synthetic reporter DR5::GUS [69] and genome-wide expression profiles of embryonic flower 1 (emf1) and emf2 Pc-G mutants identified more down- than upregulated auxin regulatory genes in the mutants [70]. Loss of vegetative Pc-G activity in clf/swn or vrn2/emf2 mutants is accompanied by loss of organ identity, generation of somatic embryos and callus-like appearance [18], [19] and all these phenotypes are associated with deregulated auxin responses. Although we cannot rule out that these phenotypes are auxin-independent, we showed mis-regulation of PIN, ARF and miRNA167 genes in Pc-G mutants revealing that loss of H3K27me3 is indeed associated with mis-regulation of genes involved in auxin responses (Figure 7). Other studies have also linked Pc-G proteins to hormonal control of gene expression: Charron and colleagues [22] uncovered strong H3K27me3 targeting of genes involved in gibberellic acid biosynthesis and inactivation. Global gene expression in emf1 and emf2 mutants revealed mis-regulation of many genes involved in diverse hormonal pathways [70]. However, many of these effects are likely indirect, thus a rigorous comparison of Pc-G protein binding, H3K27me3 targeting and gene expression profiles in Pc-G mutants in defined tissues will be essential to uncover the mis-regulated target genes that determine the Pc-G mutant phenotypes.
This and previous studies uncovered at least 25% of all Arabidopsis protein coding genes as H3K27me3 targets, some of which are tissue-specifically methylated [5], [20]–[22], [31]. Although H3K27me3 is entirely dependent on Pc-G proteins, H3K4me3 is enriched in clf/swn mutants (Figure 6) and H3K27me3 is strongly correlated with gene repression (Figure 4), only a limited number of H3K27me3 targets are mis-expressed in Pc-G mutants or in tissues in which H3K27me3 is lost (this study and [70]). Also in animals lack of Pc-G proteins results in mis-expression of only a small number of target genes [24]. Thus, loss of H3K27me3 may poise genes for activation but gene expression may require additional environmental or developmental cues. Thus, Pc-G proteins and H3K27me3 may confer stability of developmental processes and gene regulatory networks.
Pc-G proteins are considered as epigenetic regulators conferring mitotically stable gene expression states. The large number of H3K27me3 targets implies that either more than 25% of all protein coding genes are epigenetically regulated or that H3K27me3 has a more general gene regulatory role. Certainly, epigenetic phenomena like the vernalization response are controlled by Pc-G proteins and H3K27me3 [8], [71], [72]. In addition, Pc-G target genes are covered with H3K27me3 which may ensure perpetuation of the modification through replication [9]. Interestingly, for small genes like miRNA genes the number of modified nucleosomes may not be sufficient to guarantee high fidelity of inheritance.
This and previous studies unveiled tissue-specific dynamics of H3K27me3 distribution [22], [31], [51], suggesting that if H3K27me3 is generally mitotically heritable it may persist for only limited cell divisions until it is reset again to allow gene activation. The mechanisms of H3K27me3 inheritance and resetting can now be studied with the identified Pc-G target genes which are dynamically regulated during differentiation.
All plants were grown under either long day (16 h light/8 h darkness) or short day conditions (8 h light/16 h darkness). The clavata3–9 mutant in Columbia background carries a point mutation in the CLV3 coding region which disrupts CLV3 function. clf-28 (SALK_139371), swn-7 (SALK_109121), clf-28 swn-7, vrn2-1 [47], emf2-10 [18] and vrn2-1/emf2–10 mutants were used in this study. Pmir167a::GUS, Pmir167d::GUS and pARF6::ARF6-GUS were previously described [53]. For expression and ChIP-chip analyses, clv3 mutant plants were grown for 9 weeks under short day conditions and enriched for meristematic and leaf tissue by manual dissection. Tissues for expression and ChIP-chip analyses were simultaneously harvested. For zebularine treatment, plants were germinated on plates containing 40 µM zebularine (Sigma, Munich) and harvested 9 days after germination.
Specific gene ontologies (GO) were extracted for specific gene sets with the GO analysis tool on TAIR (www.arabidopsis.org/tools/bulk//go/index.jsp) and calculated as enrichment or depletion over the background of all protein coding genes in the genome (based on TAIR8). Significance was determined by a hypergeometric test. Gene families were extracted from TAIR and the plant transcription factor database (http://plntfdb.bio.uni-potsdam.de/v3.0/) [75] and compared to specific subsets of methylated genes. Significance of over- or under-representation of target genes within a gene family was tested with a χ2-test (in comparison with the observed frequency of targets or subsets within the genome).
All qPCR analyses were performed with a CHROMO4 cycler (BioRad, Munich) and a MesaGreen reaction mix (Eurogentec, Liege) with a two step program. The genes At5g60390 (EF-1α), At4g34270 and At1g13320 were used as reference genes in qRT-PCR based expression analyses [76]. For ChIP-qPCR analyses the PCRs were performed on input and immunoprecipitated samples, % of input was calculated and normalized to the reference gene FUSCA3 (FUS3, AT3G26790) which carries H3K27me3 and is not expressed in both meristematic and leaf sample. For H3K4me3 analyses, ACTIN2 (AT5G09810) was used as reference as it carries H3K4me3 and is equally expressed in both samples. Sequences of oligonucleotides used for gene expression and ChIP analyses are listed in Table S6.
Histone enriched plant extracts were prepared from ten day old seedlings as described [77]. Various histone modifications were detected by standard immunoblot analyses using antibodies specific for H3 (Abcam; ab1791), H3K27me1 (Millipore; 07-448), H3K27me2 (Millipore; 07-452), H3K27me2 (Active Motif; 39245), H3K27me3 (Millipore; 07-449), H3K4me3 (Diagenode; pAb-003-050).
Seven days old plants were histochemically assayed as previously described [78]. Photos were taken with a stereomicroscope equipped with AxioCam ICC1 camera and AxioVision4.8 software (Zeiss).
Data files for transcriptome and H3K27me3 analyses can be accessed via GEO accession number GSE24507.
|
10.1371/journal.pcbi.1004968 | Electrophysiology of Heart Failure Using a Rabbit Model: From the Failing Myocyte to Ventricular Fibrillation | Heart failure is a leading cause of death, yet its underlying electrophysiological (EP) mechanisms are not well understood. In this study, we use a multiscale approach to analyze a model of heart failure and connect its results to features of the electrocardiogram (ECG). The heart failure model is derived by modifying a previously validated electrophysiology model for a healthy rabbit heart. Specifically, in accordance with the heart failure literature, we modified the cell EP by changing both membrane currents and calcium handling. At the tissue level, we modeled the increased gap junction lateralization and lower conduction velocity due to downregulation of Connexin 43. At the biventricular level, we reduced the apex-to-base and transmural gradients of action potential duration (APD). The failing cell model was first validated by reproducing the longer action potential, slower and lower calcium transient, and earlier alternans characteristic of heart failure EP. Subsequently, we compared the electrical wave propagation in one dimensional cables of healthy and failing cells. The validated cell model was then used to simulate the EP of heart failure in an anatomically accurate biventricular rabbit model. As pacing cycle length decreases, both the normal and failing heart develop T-wave alternans, but only the failing heart shows QRS alternans (although moderate) at rapid pacing. Moreover, T-wave alternans is significantly more pronounced in the failing heart. At rapid pacing, APD maps show areas of conduction block in the failing heart. Finally, accelerated pacing initiated wave reentry and breakup in the failing heart. Further, the onset of VF was not observed with an upregulation of SERCA, a potential drug therapy, using the same protocol. The changes introduced at the cell and tissue level have increased the failing heart’s susceptibility to dynamic instabilities and arrhythmias under rapid pacing. However, the observed increase in arrhythmogenic potential is not due to a steepening of the restitution curve (not present in our model), but rather to a novel blocking mechanism.
| Ventricular fibrillation (VF) is one of the leading causes of sudden death. During VF, the electrical wave of activation in the heart breaks up chaotically. Consequently, the heart is unable to contract synchronously and pump blood to the rest of the body. In our work we formulate and validate a model of heart failure (HF) that allows us to evaluate the arrhythmogenic potential of individual and combined electrophysiological changes. In diagnostic cardiology, the electrocardiogram (ECG) is one of the most commonly used tools for detecting abnormalities in the heart electrophysiology. One of our goals is to use our numerical model to link changes at the cellular and tissue level in a failing heart to a numerically computed ECG. This allows us to characterize the precursor to and the risk of VF. In order to understand the mechanisms underlying VF in HF, we design a test that simulates a HF patient performing physical exercise. We show that under fast heart rates with changes in pacing, HF patients are more prone to VF due to a new conduction blocking mechanism. In the long term, our mathematical model is suitable for investigating the effect of drug therapies in HF.
| Heart failure is the leading cause of death and one of the most common causes of hospitalization in the United States. However, the mechanisms that lead to heart failure are still poorly understood. Evaluating the underlying cardiac electrophysiology (EP) can help in the treatment of cardiac arrhythmias and other consequences of heart failure. In this regard, computational biventricular models enable us to investigate the effect of changes in EP parameters and, by comparing the results to empirical clinical evidence, to refine the mechanisms of heart failure.
The definition of heart failure encompasses a broad range of conditions each with a compromised cardiac function. Consequently there is not a “true” single model of heart failure, but rather a range of subclasses, each requiring a separate model. Here, to narrow our scope and enable the formulation of a computational model, we focus on congestive heart failure (CHF). Although the EP of patients with congestive heart failure is still not uniquely defined, there are several common features reported in the literature [1]. For example, one of the defining characteristics of CHF is a prolonged action potential duration (APD) in myocytes [2]. This suggests abnormalities in repolarization currents, specifically in the voltage gated potassium channels. Moreover, the calcium transient is longer in duration and lower in amplitude [3]. The prolonged APD and longer and lower calcium transient suggest a potential mechanism in which the heart is compensating for reduced cardiac output by increasing the time of contraction. Under normal heart rates, these changes might not have a drastic effect on the propagation of the electrical wave of activation. However, during elevated heart rates, abnormalities arise in the voltage dynamics. Specifically, when compared to the normal myocyte, action potential (AP) and calcium alternans occur at a longer pacing cycle length in the failing myocyte. At the tissue level, alternans may be spatially concordant, that is all the myocytes alternate longer and shorter action potentials, or they may be spatially discordant if myocytes in different regions show opposite responses. For example, Qu et al. [4] have shown that by pacing a square block of tissue at 200ms concordant alternans arises, whereas spatially discordant alternans is present at pacing cycle length equal to 180ms, when myocytes have a long AP in one region and a short AP in another during the same excitation cycle. In the subsequent cycle, the opposite happens, that is, regions with previously long AP show short AP and vice versa. Together with changes in the myocyte EP, hearts suffering from congestive heart failure show a lower and more isotropic conduction velocity with respect to a healthy heart. In a normal myocyte, a high density of gap junctions is found at its end, while in a failing cell remodeling causes myocytes to revert to a juvenile state where gap junctions move to the crossfiber and sheet normal directions [5]. Due to this remodeling, the likelihood of a cell exciting a neighboring cell aligned in the fiber direction has decreased, and consequently the electrical conduction velocity is reduced [1].
Modeling cardiac tissue EP requires the combination of two underlying physics: (1) cell level ion channel mediated currents that can be described as the solution to a set of ordinary differential equations (ODEs); and (2) cell-to-cell diffusion via gap junctions as the solution to a reaction-diffusion partial differential equation (PDE) [6]. In this regard, our group has previously developed, verified, and validated a multiscale model [7] to simulate the EP of a healthy heart. Specifically we have reproduced the correct activation, electrocardiogram (ECG), and wave dynamics in a healthy rabbit heart, including the generation of ventricular fibrillation by an ectopic beat. In the present study, we modified the model to capture the known characteristic features of a failing myocyte and analyze their effect on ventricular EP.
Other groups have investigated numerically the effect of failing cell electrophysiology and structural remodeling on the heart’s susceptibility to ventricular fibrillation. For example, Gomez et al. [8] have investigated the effect of fibrosis and cellular uncoupling on the safety factor for conduction. In heart failing conditions, this safety factor is reduced in one-dimensional simulations. In a subsequent study, Gomez et al. [9] have also shown how intermediate levels of fibrosis and cellular uncoupling lead to wave reentry in 2D simulations. Other studies have focused on understanding the link between changes in ion channels expression and biomarkers of the action potential and calcium transient. For example, Walmsley et al. [10] investigate the mRNA expression in healthy and failing myocytes to predict the electrophysiology remodeling in heart failure. In doing so, Walmsley et al. consider that population variability is a key factor in constructing robust computational models. Intersubject variability is particularly important when the in silico model is adopted to test drug therapies. In silico testing of drug therapies is indeed one of the goals of current EP models (see, e.g., [11]).
Using our model we aim to investigate the mechanisms leading to ventricular fibrillation (VF) in heart failure and investigate which clinical signs (e.g., features in the ECG) are precursors to VF. Moreover, we isolate myocyte-level-changes responsible for increased arrhythmogenesis. This computational model allows us to determine if both membrane and calcium changes are necessary to generate VF in the failing heart, and if changes in conduction velocity and anisotropy (due to gap junction remodeling) are also key factors. In order to achieve these goals, we focused on the verification and validation of our model from the cellular to the biventricular level, with particular attention to the electrocardiogram in the normal and failing conditions.
In order to simulate the electrophysiology of heart failure, we need to develop and validate a single cell failing EP model and the numerical methods necessary to simulate the failing EP in a biventricular model. We present both the EP model and the numerical methods in the following.
The heart failure literature reviews by Nattel et al. [1] and Gomez et al. [12] provide an extensive summary of the changes observed in HF. In addition to the ion channel remodeling, altered calcium handling, and gap junction remodeling, structural changes (i.e., repolarization heterogeneities and fibrosis) play a major role in HF [12, 13]. Indeed, arrhythmia is initiated by the prolonged APD and longer calcium transient but is sustained by structural changes.
Our first aim is to formulate a cell model that: 1) reflects the ion channel modifications observed in heart failure; and 2) reproduces the characteristic action potential, calcium transient, sodium transient, and restitution curve observed in failing myocytes. We subdivide the myocyte EP changes depending on their direct effect on membrane ion channels or calcium dynamics.
In the following we describe the pacing protocols used at the single cell level, the simulation procedure at the 1D cable level, and the construction of the model for biventricular simulations. Due to the lowered conduction velocity in heart failure, we also perform benchmark tests to ensure accuracy of the propagation of the voltage wave of activation with respect to mesh size.
Using the models and simulation protocols described earlier, we proceed to validate our single myocyte model, analyze the EP of 1D cables of failing versus healthy myocytes, and investigate VF mechanisms in full biventricular models based on the validated failing myocyte.
By modifying the ion channels as described in the section titled “EP model of the failing myocyte”, we are able to reproduce the characteristic EP of a failing myocyte [1, 12, 19, 30]. Specifically, when compared to a normal myocyte, our failing cell model shows:
In Fig 3 we compare the electrophysiology of a normal and failing basal-epicardial myocyte. The same comparison is carried out in the supporting material for all nine cell types included in our biventricular model to simulate the apex-to-base and transmural APD gradients (Table 3). All transmural and apex-to-base cell regions in our model show the same characteristic differences listed above between healthy and failing myocytes (see supporting material.)
As discussed previously, due to both intersubject variability and the broad definition of congestive heart failure, there is not a single parameter set describing the characteristics of a failing myocyte. In order to assess the robustness of the chosen parameters regarding the single cell action potential and calcium transient, we perform preliminary uncertainty quantification (UQ) analyses. In these analyses, we perturbed each of the parameters governing the heart failure cell model by ±10% using a random uniform distribution, prepace the single cells in the nine transmural and apex-to-base regions, and record their states for one beat. We then plot the upper and lower bounds for the action potential and calcium transients in the nine cell regions.
The action potential UQ plots show that a ±10% variation in parameter values leads to an APD90 difference of approximately ±7% (see supporting material). Moreover, the lower bound APD for the failing cell resulting from the UQ analyses in any of the nine regions is higher than the corresponding APD for a normal myocyte. This is significant because, as discussed in the following sections, the onset of VF will rely on the existence of longer APD regions and corresponding functionally refractory tissue.
The upper and lower bounds for the Ca concentration computed in the UQ analyses show that ±10% cell parameter variation maintain the slower, lower, and longer calcium transient with respect to the normal myocyte (see supporting material). This persistent slow calcium recovery is also important in initiating wave propagation instabilities since it leads to calcium driven alternans.
In summary, preliminary UQ single cell analyses show that a modest variation in cell parameters does not alter the key features in cell electrophysiology that play an important role in the onset and propagation of wave instability and VF.
Wave propagation in homogeneous cell cables at PCL of 400ms reveals increased activation times in the simulations performed with the failing cell model (Fig 4) with respect to the simulations performed with the normal cell model. This activation delay is largely due to a decreased diffusion coefficient in the failing cell cable. No alternans is visible at this resting PCL. At a faster PCL equal to 250ms, the normal cable shows concordant alternans, which is visible due to the alternating shades of dark blue corresponding to the resting state. In contrast, discordant alternans is present in the failing cell cable at PCL equal to 250ms. For example, at locations within 2cm from the starting edge of the cable, we notice, in subsequent beats, a long APD followed by a short APD and then a long APD. This pattern switches at a location more distant than 2cm from the cable edge. Finally, a PCL equal to 200ms produces discordant alternans also in the normal cell cable and accentuates the discordant alternans evident in the failing cell cable.
The apex-to-base and transmural cable simulations produce similar results and therefore we address together the common features observed. The cable heterogeneity is visible at normal pacing conditions (PCL = 400 ms): as the wave progresses through the cable, the APD changes because of the different cell types (e.g., Fig 5). At PCL equal to 250ms, the apex-to-base normal cable shows regular activation in subsequent beats whereas the transmural cable has slight concordant alternans at the boundary between the epicardial and M cell. Similar to the homogeneous cable simulations, both the failing apex-to-base (Fig 5) and transmural cables (Fig 6) show discordant alternans at PCL of 250ms. Finally, at PCL of 200ms the normal apex-to-base and transmural cables exhibit, respectively, concordant and discordant alternans, whereas the failing cables show 2:1 complete conduction block.
By implementing the gradients in gto,f (peak fast potassium outward conductance) and gKs (peak potassium delayed rectifier conductance) reported in Table 3 into the biventricular model, we obtain the transmural and apex-to-base APD gradients characteristic of a failing heart, i.e. reduced APD gradients—especially in the transmural direction—when compared to normal heart gradients (Fig 7).
Using the cell model and the biventricular finite element model described earlier, we proceed to stimulate the normal and failing hearts at PCL = 400ms for four beats and compute the corresponding ECG (Fig 8). At this PCL, overall normal QRS waves and QRS wave progression are visible in the ECG obtained for both the normal and the failing hearts. Moreover no fractionations or slurring are present in either the normal or the failing heart ECG. However, the QRS waves in the failing heart ECG are slightly wider than in the normal heart ECG and marked differences between the normal and the failing hearts are present regarding the T-wave. Specifically, the T-wave peaks are lower in all leads and ST-segment depression is present in leads V5 and V6 for the failing heart.
The formulation and validation of the heart failure model presented starts at the single cell level, is investigated at the one dimensional level, and culminates with a geometrically and microstructurally accurate biventricular model. In the following, we highlight the advantages and limitations of the presented approach and discuss our findings. Mirroring the formulation of our model, we examine the single cell model, the one dimensional simulations, and the results obtained at the biventricular level.
In this work, we began by highlighting the ion-channels changes reported by several authors in the literature. We recognize that the reported values characteristic of heart failure (e.g., the peak ion conductances in heart failure) are not unique. Different heart failure patients present different combinations of abnormal ion channels. Moreover, values are reported in the literature only regarding the main peak ionic conductances but, since the implementation of the calcium handling is highly model dependent, no specific values are reported in the literature regarding the strength of calcium reuptake and release for the Mahajan et. al. cell model [22]. In this study, we first modify the membrane ion channels according to the literature and subsequently calibrated the calcium handling changes in order to obtain the calcium transient typical of heart failure. As is true regarding the changes to the membrane ion channels, the changes to calcium handling are also not unique, and separate combinations may lead to similar calcium transients [42].
Although the formulation of our model is not unique—as the myocytes of different failing hearts are not the same—we need to validate our model and reproduce the electrophysiology characteristic of failing myocardial cells. With this aim we have computed action potentials, calcium transients, sodium transients and restitution curves with our cell model in all nine apex-to-base and transmural regions. All these major features agree with the expected electrophysiology of a failing myocyte: longer APD, lower, slower and longer calcium transient, elevated sodium transient, and early alternans onset. Therefore, our cell model represents key phenomena seen in a failing myocyte.
A potential limitation in our cell model regards the limited slope increase of the restitution curve. Indeed, although the onset of alternans occurs early in the failing myocytes, the slope of the restitution curve increased only slightly. We attribute the cause of this potential limitation to the calcium handling formulation. This aspect of the cell model was conceived to well represent rapid pacing and calcium driven alternans in a normal myocyte but further modifications may be needed to best represent the electrophysiology of a failing myocyte. In order to improve this aspect of the single cell model, more experimental data are essential since the restitution curves presented in the heart failure literature show, at times, different features. For example, Glukhov et al. show [43] (slightly) flatter restitution curves in failing human hearts than in normal hearts. On the contrary, Watanabe et al.[44] report steeper restitution curves in the apical myocardium of failing canine hearts. Differences may be due to the specific experimental preparations and protocols or different types of heart failure and a greater understanding of the experimental data would strongly support a more accurate single cell model.
A second limitation of our failing cell model consists in the absence of a late sodium current that instead is replaced by a leak sodium current. This produced the desired effect of having an elevated intracellular sodium in failing cells, but evaluation of a more careful model is warranted.
Finally, recent studies have focused on the importance of SK channel expression [45], and early afterdepolarizations (EADs) [46] and delayed afterdepolarizations (DADs) [15] in heart failure, and on their pro-arrhythmic effect. Currently, these studies are complementary to the one presented here and each aims at understanding the effect of a subset of alterations characterizing heart failure. A future natural extension of the proposed model will incorporate the single cell changes responsible for SK channel expression and EAD/DADs.
One dimensional simulations are instrumental in understanding the electrophysiology of a group of connected myocytes and the mechanisms leading to ventricular fibrillation in the biventricular model. We observe the early onset of spatially discordant alternans in the one dimensional cable of failing myocytes, but not in the cable made of normal myocytes. Since discordant alternans appears in the homogeneous cable (as well as in the transmural and apex-to-base cables), it is not triggered by the presence of heterogeneities or boundaries between different cell types. Rather the observed discordant alternans is due to the failing cell electrophysiology and, in the examples presented here, is calcium driven. In fact, as shown in failing cell model cable simulations at PCL equal to 250ms (Figs 4, 5 and 6), the DI is approximately 80ms and, correspondingly, the DI/APD restitution curve is fairly flat (slope less than 0.4). This slope will not lead to a voltage driven alternans. On the contrary, calcium transients are significantly longer in the failing myocytes, and subsequent pacing takes place before full calcium recovery from the previous beat has occurred. This is therefore a calcium driven alternans.
At PCL ≈ 200ms, discordant alternans is replaced by 2:1 conduction block in both the transmural and apex-to-base failing cables. As discussed in the following, regional blocking plays a fundamental role in initiating the wave break leading to VF in this model.
Linking discordant and concordant APD alternans to a clinical tool like the ECG can provide insights into the detection and diagnosis of heart failure. The delta APD maps corresponding to the failing heart model show the presence of spatially discordant alternans at PCL = 250ms. At the same PCL, the ECG shows marked T-wave alternans with T-wave inversion (Fig 10—failing cell model at PCL = 250ms and Fig 9). A similar situation featuring slightly discordant APD alternans and T-wave alternans (although more modest and without T-wave inversion) is also seen under rapid conditions in the normal heart model (Fig 10—normal cell model at PCL = 250ms and Fig 9). Moreover, in the normal heart, a more marked T-wave alternans corresponds to a larger discordant APD alternans (Fig 10—normal cell model at PCL = 200ms and Fig 9). This suggests that spatially discordant alternans creates marked T-wave alternans in both the normal and, more significantly, in the failing heart model.
As the pacing cycle length is further decreased (PCL = 225 ms), we observe moderate QRS alternans in the ECG obtained using the failing heart model. The amplitude of the QRS wave in subsequent beats is affected by 2:1 blocking of small regions of the myocardium. Indeed, if certain regions of the myocardium are not being activated at every beat, the magnitude of the voltage wave moving past an ECG lead is lower, and consequently the QRS wave for that beat will show a lower amplitude. Finally, as the pacing cycle length is decreased to 200ms, large regional 2:1 blocking occurs at the base of the heart, and this produces the substrate for initiation of wave break and ventricular fibrillation by pacing acceleration.
Linking together the remarks described above at different pacing cycle lengths, we observe that alternans in the T-waves, and subsequently QRS-waves, were precursors to VF induced by pacing acceleration, and may serve as a model to characterize the risk of arrhythmia in patients. We note that Pastore et al. [47] observed similar features in experiments performed on guinea pig hearts at rapid pacing (e.g., Figs. 6 and 8 in [47]). T-wave alternans occurred first, followed by QRS alternans, blocking and finally VF. Similar to Pastore et al., we notice that at rapid pacing, discordant alternans, not concordant alternans, leads to arrhythmia and VF.
There exist several mechanisms to induce VF. Previously, this group has demonstrated the onset of VF using an S1–S2 stimulus protocol in a healthy heart [7]. Cao et al. [48] have shown another mechanism, in which rapid pacing alone, in the presence of non-trivial CV restitution and steep APD restitution, produces spatially discordant alternans leading to wavebreak and VF. However, this is not the mechanism responsible for VF in the current work, since rapid pacing of the failing Mahajan [22] cell model reported here does not steepen the APD restitution curve significantly. Complementing the work of Cao et. al. [48], we aim to explain a new VF mechanism in which dynamic instabilities are triggered by regions of functionally refractory tissue, due to: (1) rapid pacing (four beats at 200ms followed by two beats at 180ms); (2) heart failure cell changes; and (3) apex-to-base APD gradient. The failing cell model shows a longer APD throughout the myocardium and, as a result of the gradients, the APD is further prolonged in the basal region. During subsequent beats, the basal myocardium requires the longest time to repolarize and, at rapid pacing, temporary refractory tissue is present in subsequent beats. As a consequence, the wave of activation during the beats at 180ms encounters a zone of functionally temporarily refractory tissue, circumvents it, and finally breaks and reenters in the mid and basal region once the resting state has been reached. Regions of long and short APDs cause nonuniform wave propagation, which degenerates into wave reentry and sustained chaos due to the high pacing rate and to the lower conduction velocity. A prolonged APD in failing hearts may be a compensatory mechanism with the aim of increasing the time of contraction and offsetting reduced contractile force due to the reduced calcium transient amplitude. Our model suggests that this prolongation in action potential and calcium transient increases the susceptibility to dynamic instabilities under rapid and accelerated heart rhythm.
Several different factors must align to form a favorable substrate to induce VF. In our model, these different factors include changes to the membrane ion currents in the cell model, changes to the cell model calcium handling, and reduction of diffusion anisotropy and magnitude. All these conditions were necessary to induce VF using the proposed rapid pacing protocol, which, on the contrary, was unable to induce VF in the normal heart model or in a failing heart model including only some of these changes. Calcium and membrane changes were essential to induce alternans, and a lower more isotropic diffusion in practice “enlarged” the heart, making it more susceptible to sustained dynamic instabilities.
In this work we have focused on studying the electrophysiology of a failing heart in a mechanically static model and normal anatomy. This allowed us to decouple the pathological electrophysiology from the pathological mechanics and anatomical remodeling due to heart failure. This strategy enabled us to: 1) clearly distinguish the effects of a failing myocyte electrophysiology on ECG, activation maps, and wave of activation; 2) compare directly the new results with our previous work in a healthy heart [7]; and 3) uncover a new EP mechanism that may trigger VF due to rapid heart rates. However, we want to underline that heart failure is a combination of both mechanical and EP changes, and our current model does not include this complex coupling. In adopting this simplification we have not considered, for example, stretch activated channels and increased wall thickness to compensate for decreased contractile forces. Several groups have shown that mechanical changes can cause fluctuations in APDs and even lead to arrhythmias [49]. Therefore, in future work we aim to complement the model proposed here with a mechanical model of contraction and anatomical remodeling to study the risk of ventricular arrhythmia and VF in a fully coupled electromechanical model.
An additional improvement consists in modeling explicitly the presence of fibroblasts following, for example, the work proposed by [9, 50]. Fibroblasts have a higher resting potential with respect to myocytes. Therefore, a large enough group of fibroblasts may activate neighboring myocytes and trigger an ectopic beat that may degenerate in wavebreak and reentry.
We have constructed a multiscale model to study the electrophysiology of heart failure. We have first validated our model of failing cell electrophysiology against many experimental data reported in the literature. Subsequently, through the monodomain reaction-diffusion equation, we have coupled the single cell electrophysiology to the electrophysiology of anatomically accurate rabbit ventricles. Using the finite element method we have studied how changes at the single cell electrophysiology affect the voltage wave propagation at the tissue and full biventricular level, and the resulting ECG.
This model led to the discovery of a novel mechanism to initiate and sustain VF based on prolonged temporary refractoriness in the heart basal region. The increased basal repolarization time is due to heterogenous APD lengthening, a characteristic feature of HF. We have also shown that changes at the membrane, calcium handling, and tissue levels in cell EP are all responsible and necessary to initiate VF with our protocol and mechanism. Before the onset of wavebreak and VF, T-wave and mild QRS alternans are present in the ECG. Similar results are confirmed experimentally by the work of Pastore et al. [47]. In addition, we show in our model that T-wave alternans is linked to spatially discordant alternans.
We conclude by underlining that additional ion channels or mechanisms may be added to our cell model in a straightforward and modular way. These additional features may both improve the model of the failing myocyte and add ion channels relevant to a particular therapy or a specific form of heart failure.
|
10.1371/journal.pgen.1003942 | Demographic Divergence History of Pied Flycatcher and Collared Flycatcher Inferred from Whole-Genome Re-sequencing Data | Profound knowledge of demographic history is a prerequisite for the understanding and inference of processes involved in the evolution of population differentiation and speciation. Together with new coalescent-based methods, the recent availability of genome-wide data enables investigation of differentiation and divergence processes at unprecedented depth. We combined two powerful approaches, full Approximate Bayesian Computation analysis (ABC) and pairwise sequentially Markovian coalescent modeling (PSMC), to reconstruct the demographic history of the split between two avian speciation model species, the pied flycatcher and collared flycatcher. Using whole-genome re-sequencing data from 20 individuals, we investigated 15 demographic models including different levels and patterns of gene flow, and changes in effective population size over time. ABC provided high support for recent (mode 0.3 my, range <0.7 my) species divergence, declines in effective population size of both species since their initial divergence, and unidirectional recent gene flow from pied flycatcher into collared flycatcher. The estimated divergence time and population size changes, supported by PSMC results, suggest that the ancestral species persisted through one of the glacial periods of middle Pleistocene and then split into two large populations that first increased in size before going through severe bottlenecks and expanding into their current ranges. Secondary contact appears to have been established after the last glacial maximum. The severity of the bottlenecks at the last glacial maximum is indicated by the discrepancy between current effective population sizes (20,000–80,000) and census sizes (5–50 million birds) of the two species. The recent divergence time challenges the supposition that avian speciation is a relatively slow process with extended times for intrinsic postzygotic reproductive barriers to evolve. Our study emphasizes the importance of using genome-wide data to unravel tangled demographic histories. Moreover, it constitutes one of the first examples of the inference of divergence history from genome-wide data in non-model species.
| Demographic processes leave specific and detectable signatures within species genomes. Analysis of patterns of variation within and between closely related species can be used to unravel their divergence history and is crucial for understanding evolutionary processes such as speciation. We applied a set of novel population-genomic tools to investigate patterns of natural variation and infer demographic history of two avian speciation model species: pied flycatcher and collared flycatcher. The analysis supported a scenario consistent with allopatric speciation with recent, postglacial secondary contact. Most likely the ancestral species persisted through one of the glacial periods of the middle Pleistocene and then split into two large descendent populations that appear to have increased in size before experiencing severe bottlenecks during expansion into their current ranges. The two species established secondary contact after the last glacial maximum. This resulted in unidirectional gene flow from pied flycatcher to collared flycatcher. The results are consistent with a scenario where pied flycatcher recolonized northern Europe more rapidly than collared flycatcher. Our study increases the knowledge about the dynamics of the speciation process and constitutes one of the first examples of the inference of complex demographic history using information from genome-wide data in non-model species.
| Considerable attention is currently paid to the role of gene flow during speciation [1]–[4]. In the presence of gene flow, strong ecology-driven divergent selection is an important initial prerequisite for the evolution of reproductive isolation. In allopatric speciation, the initial stages of speciation can be facilitated by genetic drift and local adaptation in geographic separation without being countered by gene flow. Still, gene flow may occur in secondary contact via introgressive hybridization and in some cases boundaries may then collapse [5]–[10]. Irrespective of the role of selection and geographic separation, the evolution and maintenance of reproductive isolation in the face of gene flow is expected to generate a genomic mosaic in which regions permeable to gene flow are less differentiated than regions resistant to introgression [11]. Such mosaics are characterized by the presence of ‘genomic islands of speciation’ [12], [13], genome regions which may harbor loci under divergent selection and potentially underlie reproductive incompatibility. However, what processes contribute to these patterns is still a matter of debate. Under the model of divergence hitchhiking [4], [13], [14], such regions can be extensive, with reduced genetic exchange over several megabases (Mb) of linked sequence. However, it is the remaining regions of the genome that harbor information about patterns of gene flow and other demographic processes that, apart from different types of selection, influence species differentiation. Moreover, in order to correctly infer the evolutionary and population processes causing localized elevated differentiation, it is imperative that background levels of gene flow are well characterized.
A wide range of approaches have been developed to estimate demographic history and/or the role of gene flow during (and after) speciation. Particularly relevant recent developments include numerous coalescent-based methods (e.g. [15]–[18]) that estimate ancestral population sizes, historical gene flow, and divergence times. The coalescent offers a powerful theoretical framework for such analyses [19]–[21] and coalescence modeling is increasingly used in the context of speciation research [22]–[25]. The isolation-with-migration (IM) model of Hey and Nielsen [24] has been successfully applied in the past to distinguish ancestral polymorphism from introgression and to estimate divergence history and the role of gene flow during speciation in many species (reviewed in [26]). However, it exclusively considers demographic scenarios with constant migration rates between species, and thus offers no means to investigate more complex patterns of gene flow over time. Moreover, it is computationally demanding (due to likelihood function evaluation) and its use is limited to rather small datasets [26]–[31]. The Approximate Bayesian Computation (ABC; [32]) approach bypasses exact likelihood calculation by using summary statistics to characterize patterns of variation observed in the data. The approach is also very flexible in defining demographic models used to infer demographic parameters [33]–[37]. Since their first implementation in population genetics, ABC methods have been constantly developed and improved [18], resulting in an increasing number of studies inferring demography within an ABC framework [33], [34], [37], [41]–[50].
Though coalescent modeling can handle genome-wide data, its application for genome-wide demographic inference has so far been restricted by the limited access to whole-genome sequence data. Notable exceptions include studies of the demographic history of humans [41], [51]–[53], other primates [17], [42], [54], [55], and Drosophila melanogaster [47]. With the emergence of the field of speciation genomics and the foreseeable increase in the number of non-model genomes sequenced [56], an increase in the number of studies inferring population history of important study organisms from genome-wide data is also to be expected. Here we present one of the first examples in this direction.
We have recently sequenced and de novo assembled the 1.1 Gb genome of the collared flycatcher Ficedula albicollis [57]. Together with its sister species, the pied flycatcher (F. hypoleuca), it forms an important model system in evolutionary ecology and biology (e.g. [58]–[61]), including studies of hybridization and speciation [62], [63], and genetics [64]–[70]. The two flycatchers are small, migratory birds that belong to the order Passeriformes. The pied flycatcher breeding range covers a large part of the western Palearctic (Figure 1) and overlaps with collared flycatcher in two areas (central Europe and Baltic Sea islands). In these regions the species coexist and hybridize occasionally [71]. However, the fitness of hybrid offspring is severely reduced [72], with females apparently being sterile [73]–[76]. This is in accordance with Haldane's rule, as birds have female heterogamety. Previous genetic studies in flycatchers have indicated no or very low levels of gene flow between allopatric populations of pied flycatchers and collared flycatchers, and moderate gene flow in the area of recent sympatry on the Baltic islands [28], [77], [78].
Here we capitalize on data from a whole-genome re-sequencing effort in flycatchers [57]. These data, comprising >10 million single nucleotide polymorphisms (SNPs), allow us to carefully choose genomic regions spread across the flycatcher genome and analyze them in an ABC framework (augmented by PSMC modeling) to infer demography and gene flow during different stages of species divergence in this ecological model system.
After stringent filtering of whole-genome re-sequencing data from 10 collared flycatchers and 10 pied flycatchers, each individual sequenced at an average of 5× coverage, we investigated sequence variation in 267 independent, noncoding loci covering a total of 534 kb (≈0.05% of the genome). At these loci, genotypes could be called at 429,753 sites in at least seven individuals in each species. Sequence diversity was higher in collared flycatcher (mean π = 0.0033±0.0034) than in pied flycatcher (mean π = 0.0020±0.0025). The data contained a substantial fraction of shared polymorphisms (i.e., sites segregating in both species; 0.22±0.02) and some fixed differences (0.03±0.00); note that ‘fixed’ in this context means monomorphic for different alleles in these particular samples of the two species. We observed many more SNPs that were unique to collared flycatcher (0.54±0.02; fraction of all SNPs) than to pied flycatcher (0.21±0.02). The differentiation between species was moderate (mean Fst = 0.21±0.01). Mean values for Tajima's D statistics were positive for both species (0.17±0.28 for collared flycatcher; 0.40±0.43 for pied flycatcher). All summary statistics are in good agreement with genomic background variation recently reported for whole-genome data [57].
We examined 15 demographic models of flycatcher divergence (five scenarios with three models each; Figure 2), and identified eight models for which the likelihood of observed data (calculated under Generalized Linear Model) fell well within the distribution of retained simulated data (Table S1). These included models from four demographic scenarios: three models from a scenario with recent gene flow (‘recent migration constant size’, RMCS; ‘recent migration recent size change’, RMRSC; recent migration ancient size change, RMASC), two models from a scenario with constant migration (‘constant migration constant size’, CMCS; ‘constant migration and recent population size changes’, CMRSC), two models from a scenario with ancient and recent migration with a period of isolation between the two phases of gene flow, either with constant population size (RAMCS), or recent population size changes (RAMRSC) and one model without migration between species (‘isolation ancient size change’, IASC). All models with ancient gene flow yielded very low P-values (most of the simulated datasets having much higher likelihood than the likelihood of the observed data), indicating that they did not fit the observed data.
Model choice conducted within each of the four plausible scenarios suggested four models that fit the observed data best: IASC, CMCS, RMASC and RAMRSC (Figure 3). Of these models RMASC had the highest posterior probability (PP = 0.90); the posterior probabilities for IASC, CMCS and RAMRSC were very low (0.05, 0.05 and 0.00, respectively). The RMASC model was clearly the best model also when we compared the eight models that fit the data well in a single model selection procedure (PP = 0.77) as well as when we used an alternative nesting procedure (migration nested within population size dynamics; PP = 0.98). The RMASC model was also the best model when ‘not-optimized’ prior ranges were used suggesting that the choice of prior ranges had little influence on best model selection (Table S2). The power to correctly predict the models was 0.57, 0.83, 0.67, and 0.74, which is much higher than the expected 25% and indicates that we were able to clearly discriminate the models. RMASC, i.e. the model with recent migration and ancient size change, fitted the data significantly better than all other tested models, and was therefore chosen for parameter estimation (Figure 3).
The Partial Least Squares (PLS) components of the observed summary statistics fell well within the density distribution of the PLS components of the retained simulations, demonstrating that simulations were appropriately exploring the summary statistic space (Figure S1). To verify the coverage properties of the marginal posterior distributions estimated with the chosen estimation approach, we generated 1,000 pseudo-observed data sets and tested the distributions of posterior quantiles for each parameter of the best model (based on 15,000 retained simulations and seven PLS components). Most of the parameters had a uniform distribution (Table 1; Figure S2) and coefficients of variation indicated that we had enough power to estimate most of the parameters (R2>10%; [79]; Table 1). Nevertheless, to reduce complexity we also considered a model assuming no migration from collared flycatcher to pied flycatcher. This was motivated by very low amounts of gene flow in this direction estimated in the RMASC model (mode = 8.33×10−9). We also updated priors for effective population size of pied flycatcher based on posterior distributions. The model with unidirectional migration (model RUMASC) was run for 2×106 simulations, and submitted to careful examination and validation based on 5,100 retained simulations and seven PLS components. The RUMASC model had higher posterior probability (PP = 0.62) than the RMASC model (PP = 0.38), and the distribution of posterior quantiles exhibited limited bias in the posterior distributions (Table S3; Figure S3). However, the power to correctly predict models RMASC and RUMASC was rather small (0.58 for RMASC and 0.63 for RUMASC). This is expected since both of them produced very similar posterior probability distributions. We therefore present parameter estimates for both models (Figure 4, Figure 5, Table 1, Table S3 and Table 2).
Distributions of divergence time (TS) estimates fell within the range of a few hundred thousand years indicating recent origin of the flycatcher species (mode TS≈340,000 years in RMASC and 230,000 years in RUMASC). The estimated population size of the common ancestor (mode Nanc≈600,000 and 550,000, respectively) was much larger than current Ne of both collared flycatcher (mode Ncoll≈80,000 and 65,000, respectively) and pied flycatcher (mode Npied≈31,000 and 23,000, respectively). Both species thus showed a strong signal of population decline since their initial divergence, with the decrease being more severe in the pied flycatcher than in the collared flycatcher. Posterior probability curves of the relative size of post-split and current population size (NPScoll/Ncoll and NPSpied/Npied) encompassed only values larger than one, but the strength of the decline is difficult to estimate due to wide 90% highest posterior density intervals (HPDI). The rate of gene flow from collared flycatcher to pied flycatcher was very low (RMASC, mode = 8.33×10−9). In the opposite direction, mpied→coll, gene flow was estimated 4.55×10−6 in RMASC and 2.42×10−6 in RUMASC. This corresponds to 0.36 and 0.16 migrants per generation, or one migrant about every three and six generations, respectively. Although the exact timing of gene flow between populations was not possible to estimate (very wide and flat posterior probability distributions of Tmcoll→pied and Tmpied→coll), a model with recent (after Last Glacial Maximum, LGM) gene flow was favored.
To investigate changes in Ne over time in more detail we performed pairwise sequentially Markovian coalescent modeling (PSMC) analysis using the diploid sequence of a collared flycatcher male sequenced at 85× coverage. The analysis showed good resolution between 50 ky and 2 my, and rather small variance associated with most of the Ne estimates (Figure 6). The effective size of the population substantially increased from approximately 500,000 individuals 1 my ago (i.e., before the pied flycatcher-collared flycatcher split) to a maximum of 1.6 million individuals 200 ky ago. From approximately 200 ky ago effective size started to decrease and reached about 500,000 individuals several tens of thousands years ago. The ABC estimate of the effective size of the ancestral population (≈700,000) was thus very similar to the PSMC estimate of Ne before species divergence. Due to a limited number of recent coalescent events that can be inferred from a single genome sequence, the estimation of more recent changes in Ne is not possible [15].
We analyzed sequence variation in several hundred intergenic loci (totaling ≈0.5 Mb) to infer demographic parameters of the divergence history of pied flycatcher and collared flycatchers. Stringent filtering of whole-genome re-sequencing data and careful evaluation of ABC analyses enabled us to infer the demographic scenario of species differentiation with high confidence. The divergence time estimate was consistent with a recent, middle Pleistocene split of the common ancestor of the two species. Since their initial divergence Ne of both species declined and unidirectional gene flow from pied flycatcher into collared flycatcher took place at a recent time scale (most likely after the LGM). Some, but not all, demographic parameters were in good agreement with previous estimates [28], [77]. However, in addition to previous studies that were based on limited data and simple demographic models (constant migration, no population size change over time; [28], [77]), our genome-wide approach enabled us to study divergence in much greater detail. We explicitly modeled contrasting patterns of gene flow and population size changes over time and our results consequently reveal new and important demographic aspects of the divergence history of pied flycatcher and collared flycatcher, which contribute to understanding of the genomic landscape of species divergence in this system. Phrased differently, the work can be seen as relevant in the context of genome divergence as well as of species divergence.
The estimated effective size of the ancestral population (≈600,000) was larger than the current Ne of both species, and much larger than the ancestral Ne estimate of 130,000 reported by Backström at al. [28] (see further below). In agreement with observed patterns of intraspecific diversity, current Ne of collared flycatcher (65,000–80,000) was higher than that of pied flycatcher (23,000–31,000), similar to values estimated for other European populations of these species [28], [77]. However, Ne estimates of both species are in sharp contrast to estimated census sizes in Europe. These are two to three orders of magnitude larger, with 4.2–7.2 million for collared flycatcher and 36–60 million for pied flycatcher (http://www.birdlife.org). The remarkable discrepancies between census and effective population sizes indicate successful postglacial expansions from apparently significantly bottlenecked refugial populations in both species. Moreover, the much higher census size of pied flycatcher compared to collared flycatcher coupled with the opposite relationship for Ne suggests a more rapid, and as testified by current breeding ranges, more extensive post-glacial re-colonization of northern habitats by pied flycatcher. This is in line with the estimated relative sizes of post-split and current Ne (NPScoll/Ncoll≈100 and NPSpied/Npied≈1000) of both species, which indicate a much more severe decline for pied flycatcher. As a general caveat to these issues, we note that changes in population structure over time may affect coalescent rate estimates and, as a consequence, influence Ne estimates [21].
Although the ABC-based estimation of the magnitudes of population decline has to be treated with some caution, our analyses confidently evidence significant post-divergence population size decreases in both species. The rank order for Ne (NPScoll and NPSpied>Nanc>Ncoll and Npied) indicates that the ancestral population differentiated into two descendent populations without any sign of bottleneck associated with initial divergence. Both post-split populations appear to first have increased in size before subsequent population decline during glacial periods. While this interpretation would remain speculative based on the ABC analyses alone, it is supported by the PSMC results. The time of population size increase in the PSMC curve for collared flycatcher largely overlaps with divergence times estimated by ABC, indicating an increase in collared flycatcher Ne after initial differentiation from the ancestral population (Figure 6). The mode for TS in RMASC model (340,000) and its 50% HPDI (230,000–480,000) include almost exclusively the epoch before Ne decline indicated by the PSMC curve. Assuming that the maximum Ne from PSMC analysis (1.6 million) approximates NPScoll, the ratio of NPScoll/Ncoll indicates a 20-fold decline in population size in the last 200 ky (NPScoll/Ncoll = 20.17). This value falls well within the 50% HPDI estimated by ABC analysis, lending additional support for the RMASC model and ABC-based estimates. On the other hand, the mode of TS from RUMASC model (230,000) coincides with the peak of the PSMC curve. However, it is important to note that the divergence time estimate in this model can be biased (as indicated by distribution of posterior quantiles, Table S3) and has to be treated with caution. Nevertheless, regardless of the divergence time estimates and consistent with ABC analysis, PSMC estimation of Ne clearly indicates a rapid population decline during the first half of the last glacial period (100,000 - 50,000 years ago). An alternative scenario consistent with PSMC estimates would imply a population split of the post-split collared flycatcher population into two or more subpopulations followed by their admixture after a period of isolation. In this case PSMC as well as other coalescent-based methods may overestimate Ne during the period of population split (Li and Durbin 2011).
In addition to the evaluation of population size changes during species divergence, the ABC approach enabled us to model various scenarios of gene flow over time. A model with gene flow occurring exclusively after the LGM was favored, with an estimate equivalent to one individual per three to six generations introgressing from pied flycatcher to collared flycatcher. In accordance with previous studies [28], [77], the rate of gene flow in the inverse direction, from collared flycatcher to pied flycatcher was estimated as essentially absent. The estimated pattern of gene flow is in line with the expectations for invading and resident populations: unidirectional gene flow from the resident, stable population into the expanding, invading population [80], [81]. Thus, it is most likely that pied flycatcher colonized northern Europe more rapidly than collared flycatcher and collared flycatcher arrived some time later as an invading species. The scenario is supported by the patterns of estimated Ne (discussed above) and also by recent observations form the Baltic Sea islands where collared flycatchers colonized habitats previously inhabited only by pied flycatchers [82].
Interestingly, the estimates of the divergence time between species (modes of 340,000 and 230,000 years ago in RMASC and RUMASC, respectively) indicate much more recent divergence than estimated from mitochondrial DNA (mtDNA; 1–2 my based on ≈3% mtDNA divergence; [64]). Since pied flycatcher and collared flycatcher have already reached an advanced stage of reproductive isolation (female hybrids are sterile, male hybrids have significantly reduced fertility; [63], [72]) this may be seen as surprising given that birds are thought to develop reproductive barriers rather late in the speciation process [83]. However, mtDNA-based estimates of divergence time may be biased for at least two reasons. First, gene divergence often predates species divergence [84]. Second, due to the stochastic nature of the coalescent process and huge variance associated with single-locus estimates of TMRCA, estimates of divergence time based on mtDNA alone might be unreliable [85]. Indeed, a model-based approach applied to mtDNA data would give huge credible intervals. Divergence time estimates based on 24 autosomal loci clearly reduced variation related to the coalescent processes and pointed towards more recent divergence (approximately 0.5 my; [28]). However, the distribution of IMa-based divergence time was still wide with 90% HPDI exceeding 1 my. With the resolution now given by the genome-wide approach, we could further narrow the interval to less than 700,000 years.
Although the flycatcher system may be exceptional when it comes to the rate of formation of reproductive incompatibility, we note that the hypothesis of speciation potentially being a relatively slow process in birds, with extended times for intrinsic postzygotic reproductive barriers to evolve, is mainly based on data from mtDNA studies [83], [86]. If other genome-wide studies of avian speciation models will also come to suggest more recent divergence than estimated by data from mtDNA, and some preliminary data actually point in this direction (e.g. [87], [88]), this hypothesis may have to be revised. We also note that our estimates of divergence time derive from the ability to include population size changes in ABC models, which has not been possible in previous work. Ignoring the detected decline would lead to an upwards biased divergence time estimate.
Differences between our results and the results presented by Backström et al [28] and Hogner et al. [77] seem most likely attributable to the fact that previous work was based on a relatively limited number of intronic loci. The increased amount of data in the present study may thus have contributed to substantially improving the accuracy of demographic parameter estimates [89]. Moreover, the general pattern of variation observed in our genome-wide data differed from the previous intronic datasets. Consistent with Ellegren at al. [57], nucleotide diversity was smaller than estimates based on limited intronic data (mean π = 0.0020 and 0.0033 for pied flycatcher and collared flycatcher in our genome-wide data and 0.0041 and 0.0044 for populations studied in [28]). This explains our lower estimates of Ne for both species. Also, while mean Tajima's D was positive for collared flycatcher in genome-wide data (0.17), it was estimated negative in previous studies (−0.32).
The likelihood of different scenarios of gene flow during speciation is currently a much debated topic in evolutionary biology (e.g. [3], [4], [90]). Besides the mere verification whether speciation can occur in the face of gene flow, a challenging task is to distingiush between scenarios with gene flow already during initial differentiation (sympatric or parapatric speciation), constant migration during divergence (or multiple admixture events), and gene flow occurring only after a long period of allopatric divergence when populations come into secondary contact. By explicit modeling of different patterns of gene flow over time, we were able to infer a demographic history consistent with allopatric speciation followed by secondary and recent contact as the most likely scenario of flycatcher differentation (models with ancient gene flow had very low posterior probabilities). The results are important in the context of the overall genomic landscape of species divergence in this system. We have recently shown the genomic landscape is highly heterogeneous with one or a few regions per chromosome showing highly elevated differentiation (divergence peaks, potentially representing “genomic islands of speciation”; [57]). These regions, which are low in shared polymorphisms between species and high in private polymorphisms (relative to other regions of the genome), are candidates to have evolved under the strong influence of selection. With the relatively recent divergence suggested by our analyses (i.e. mode of 340,000 and 230,000 my, respectively), elevated divergence in these islands must have been built up rapidly. The co-localization of divergence peaks and centromeres as well as telomeres fed the hypothesis that meiotic drive may have been involved in generating high divergence [39] and potentially segregation distortion is a process potent enough to rapidly generate genetic incompatibilities. Also, reduced recombination in centromeres may contribute to high divergence, however, recombination at telomeres seems elevated on avian chromosomes [91].
It is possible that gene flow upon secondary contact reinforces a genomic landscape of heterogeneous sequence divergence. Specifically, introgression may lower background levels of divergence, or at least act as to their maintenance, while selection, if it occurs, continues to build up divergence in genomic islands in which gene flow is hindered. Importantly, with the levels and continuance of gene flow observed, gene flow alone cannot have been sufficient to have had a predominant role in the evolution of the differentiation landscape. The general implication of this is that the differentiation islands have to be explained by different mechanisms than the breaking down of differentiation by gene flow in the genomic background. As indicated above, a candidate mechanism is obviously divergent selection, however, locally enhanced lineage sorting due to a heterogeneous recombination landscape cannot be excluded and will require very fine-scale estimates of recombination rates to be addressed.
While our study provides unusually detailed insight into the demographic history and the processes affecting genetic differentiation at a genome-wide scale in a speciation model system, it should be stressed that most posterior estimates still reflect considerable uncertainty. Even with the access to a draft genome assembly, data from whole-genome re-sequencing of population samples and complex and computer intensive methods, there are thus limitations as how far genetic data can perfectly reconstruct demographic history. Some accuracy may have been gained by analyzing additional loci, however, theoretical work has recently shown that the width of credible intervals in ABC analyses rapidly decreases when hundreds of loci are analyzed [89]. Moreover, adding more loci would have required relaxing the criterion of only including loci located at least 500 kb apart, which would have made linkage disequilibrium (LD) a possible issue. For future studies, summary statistics taking into account the structure of LD may represent the most promising avenue in order to further distinguish between scenarios and by improving parameter estimates. However, this can only be obtained with sufficient confidence from higher-coverage re-sequencing data than used herein.
We investigated the demographic history of two closely related bird species using whole-genome re-sequencing data and a full ABC approach supported by additional coalescent-based analysis. By applying stringent filtering, careful ABC evaluation and hierarchical model choice we were able to investigate different demographic scenarios including different patterns of population size change and gene flow over time. The best-supported scenario of flycatcher divergence indicated that the ancestral species survived one of the glacial periods of middle Pleistocene, split into two large populations that both appear to have increased in size during the warm interglacial period before they experienced severe bottlenecks. The species probably came into secondary contact after LGM, which resulted in mostly unidirectional gene flow from pied flycatcher to collared flycatcher. Our study constitutes one of the first examples of detailed modeling of the complex divergence history in an emerging model system for speciation genomics. Indeed, Ficedula flycatchers may be a type example of speciation during Pleistocene, where alternating cycles of glacial and inter-glacial periods have shaped genomic differentiation.
We randomly sampled independent loci distributed across the genome, each comprising 2,000 bp of assembled sequence. Each locus was required to be situated at least 500 kb apart from other sampled loci. This physical distance is well above the lengths of linkage disequilibrium blocks seen in collared flycatcher [68]. First, we randomly sampled 2,000 loci (the maximum amount in an approximately 1 Gb genome theoretically possible when not allowing loci to be closer than 500 kb apart) and were able to collect 1,086 loci fulfilling the density criterion, a reduction following from randomness of sampling and chromosome structure. For further analysis we only kept loci that were found in autosomal, noncoding regions of the genome, and we excluded sequences that exhibited elevated levels of divergence (“divergence islands” identified in [57]). Sequence data were subsequently extracted for 10 pied flycatcher and 10 collared flycatchers (generated as described in [57]) and filtered based on sequence coverage. For every sampled locus we analyzed only those sites that passed the threshold of being covered by at least 3 reads per site in at least seven individuals per species; all sites that did not pass the filter's threshold were masked as missing data. This strategy should have enabled us to filter out most of the sequencing errors. As a next filtering step, we used only loci that consisted of no more than 30% of missing data. This step reduced computational time by not simulating too many sites that would not be used in further inference. To avoid the risk of mistakenly calling a heterozygous site as homozygote we haploidized sampled loci by randomly sampling one allele per site. Our final dataset contained 267 loci, i.e. all loci fulfilling all criteria, at which 80% of the sites (429,753 bp out of a total of 534 kb) were covered by called genotypes for each individual.
We analyzed the data under an Approximate Bayesian Computation framework [32]. ABC methods come in different flavors but all standard approaches share the same general scheme and strategy: 1) The observed data is characterized by a set of summary statistics known to be informative about parameters of interest, 2) millions of datasets are generated under a demographic model, each with different parameter values randomly drawn from given prior distributions, 3) if more than one model is considered, the best fitting model is selected, and 4) datasets for which summary statistics are closest to those obtained from real data are used for estimation of the best model parameters [39], [40]. Although intuitively straightforward, ABC is not a ‘plug and play’ analysis and often requires careful investigation of each step of the protocol and several quality control checkpoints.
We used the ABCtoolbox software designed to perform ABC analysis and facilitate and integrate simulation, summary statistics calculation, and parameter estimation steps into a single pipeline ([92]; kindly updated for us by D. Wegmann). Simulations were performed using msABC [93], a modified version of ms [94]. Since we stringently filtered sequence data, we paid special attention to treat the simulated data in the same way. Thus, in every msABC iteration we simulated 267 loci (for 10 pied flycatchers and 10 collared flycatchers), masked all sites that did not pass the coverage threshold in the original data (see above in Samples and Loci) and, before calculating summary statistics, haplodized the data.
As required in msABC all parameters were scaled by a factor N0 which we set to 10,000. Thus, effective population size (Ne) was simulated as N/N0; time parameters equalled T/4N0 and migration parameters (M) were scaled as 4N0mij, where mij is the fraction of population i which is made up of migrants from population j each generation. We based most of the parameter prior range distributions on the results from previous studies (IM estimates based on 24 nuclear loci; [28]) and kept them wide enough to ensure that a variety of plausible parameter values could be captured. We used a standard ABC approach and sampled parameter values from uniform distributions and for most cases set to a log10 scale (Ne, relative population sizes, migration rates. Recombination rate (r) prior was set based on a high-density recombination map recently developed for the collared flycatcher (unpublished data); for each simulated locus we obtained the local estimate of the recombination rate (mean 5.3×10−8). Based on the distribution of recombination rates we set the recombination rate for each locus to be drawn from a Gamma distribution G (α,α/r), with the shape parameter α drawn from U [1], [12]. The mean mutation rate (μ) prior was chosen based on our previous estimates [57], [95]. The mutation rate for each locus was then drawn from a Gamma distribution G(α,α/μ), with the shape parameter α drawn from U [3], [12].
To start with, we first ran five classes of exploratory simulations including different demographic scenarios: isolation, constant migration over time, recent (after LGM) migration, ancient migration, and ancient as well as recent migration with a period of isolation between two phases of gene flow (Figure 2). For each scenario we investigated three models in which we either assumed 1) constant Ne of descendant populations, 2) exponential change in Ne of descendant populations after the LGM, or 3) exponential change in Ne of descendant populations since their initial divergence. Population size changes were modeled by assuming that the population size at the start of size change was a fraction x of the current Ne (priors were set to capture both population growth and decline). In all models with recent migration we assumed bidirectional migration and the ranges of priors were set to cover very small to moderate levels of gene flow. Since the ability to detect strong signatures of asymmetric gene flow between ancient populations is very low we assumed symmetrical migration in all models with ancestral migration events. The number of exploratory simulation varied from 100,000 to 200,000, enough to judge if the model is able to explain the observed data. For every model we checked the fraction of retained simulations (2.5%) with a smaller or equal likelihood than the likelihood of the observed data (P-value reported by ABCtoolbox). The likelihoods were estimated for truncated models under General Linear Model post-sampling adjustment (ABC-GLM, [38]). We also inspected the posterior probability curves to check if the model fitting could be improved by changing the ranges of priors. In several cases we updated ranges and ran particular scenarios one more time. All models for which the likelihood of observed data fell within the distribution of simulated data were run for 2 million simulations.
ABC inference was based on a set of summary statistics calculated for each species separately and for both species combined. We calculated mean and variance across all 267 investigated loci using msABC for the following summary statistics: nucleotide diversity (π), Tajima's D (D) and Fst. In addition, using in-house perl scripts we calculated the proportions of shared, fixed, private (for pied flycatcher and collared flycatcher, respectively) polymorphisms. Following Wegmann at al. [18], we defined a set of orthogonal linear-combinations of summary statistics that best explained the variance in the model parameter space by transforming the full set of summary statistic via Partial Least Squares [96]. All transformation were done in the R package PLS [97] and the appropriate number of PLS components were defined based on root mean squared error plots (RMSEP plots). PLS transformed statistics were used to calculate the Euclidean distance between observed and simulated datasets and up to 3% of simulations with the smallest distance were retained for parameter estimations via the regression adjustment ABC-GLM [38] implemented in ABCtoolbox.
The model choice procedure was conducted in the ABCtoolbox. We used distances calculated based on PLS components to choose the simulations that were closest to the observed data but untransformed summary statistics dataset (excluding statistics that were highly correlated: mean and variance of number of fixed differences correlated with mean FST and variance of number of private polymorphisms (respectively) in IASC model; mean and variance of nucleotide diversity for both species correlated with pied flycatcher nucleotide diversity estimates in CMRSC model) to perform model selection via Bayes factors (ratios of marginal densities). Following Fagundes at al. [98] we applied a hierarchical model choice procedure. First, we evaluated posterior probabilities of different models within each scenario considered here. Then we compared the best model of each scenario to the best models from other scenarios. In addition, to test the robustness of our conclusions, we also compared all models for which the likelihood of observed data fell well within the distribution of simulated data in a single model selection procedure. Moreover, and for the same reason, we applied an alternative nesting strategy where we nested migration dynamics within population size dynamics. To estimate the power of our procedure to distinguish between selected models we generated 1,000 pseudo-observed datasets for each model and checked how many times the ABC model choice procedure failed to correctly predict the true model [41]. Each pseudo-observed dataset produced by a considered model (the true model in this case) was treated as observed data and used to calculate marginal densities of all compared models. Bayes factors were used to judge if a selected model coincided with the true model. Our demographic model evaluation procedure included slight adjustments of prior ranges for particular model parameters and this adjustment procedure may influence model selection by favouring more optimized models over less optimized once. This is the consequence of Bayes factor calculations that are based on the marginal likelihoods of the models considered: the marginal likelihood of a model will be higher if the selected prior probability distributions are more similar to the true posterior probability distributions. Thus, to validate our model choice analyses we ran additional simulations to evaluate the sensitivity of the model posterior probability distributions to choices of different prior distribution. The best model (RMASC) was run with 4 different ‘sub-optimal’ prior ranges (the ranges of sub-optimal priors corresponded to the adjustment we made during exploratory simulations, Table S4). For each sub-optimal model we followed the same hierarchical model choice procedure as for our original simulations.
We validated the chosen estimation procedure and summary statistics by checking for a potential bias in the posterior distributions [18], [42]. We generated 1,000 pseudo-observed datasets with known parameter values and computed coverage property of the posterior distributions obtained with ABC-GLM regression adjustment. The uniformity of the posterior quantiles for each parameter was checked with a Kolomogorov-Smirnov test and its significance was obtained after Bonferroni correction. To verify if retained simulations were exploring the appropriate space of summary statistics, we plotted PLS components together with observed transformed statistics. To check the power to estimate individual parameters we computed the coefficient of variation (R2) by regressing PLS components against model parameters [79]. In addition, we computed the root mean squared error of the mode (RMSE) for each parameter to check the accuracy of the mode as a point estimate [42].
All simulations were run on linux clusters at Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX). Often we run several hundred simulations in parallel and we used in-house scripts to generate random seed numbers for each simulation to avoid the risk of several simulations being identical.
Changes in effective population size over time were assessed by pairwise sequentially Markovian coalescent model analysis [15]. The model estimates the local time to the most recent common ancestor based on a single whole-genome diploid sequence and uses information from the rates of the coalescent events in a given epoch to infer Ne at a given time [15], [99]. Since the method heavily relies on the distribution of polymorphic sites across the genome, it can only be used when both alleles are called with high confidence (i.e., when per-site coverage is high). Thus, we used the diploid sequence of the male collared flycatcher sequenced for genome assembly (mean coverage 85×; [57]). Data was filtered by excluding sites at which read depth was more than twice or less than half of the average read depth, the root mean squared mapping quality of reads covering the site was below 25, the site was within 10 bp around predicted indels and the inferred consensus quality was below 20. A generation time of 1 year and a mutation rate of 1.4×10−9 year/site were applied (based on our ABC analysis). The settings of the PSMC analysis (-p and –t options) were chosen manually according to suggestions given by Li and Durbin ([15], https://github.com/lh3/psmc). To check for variance in Ne estimates we performed a total of 100 bootstrap tests.
|
10.1371/journal.pntd.0003490 | Vaccination with Leishmania infantum Acidic Ribosomal P0 but Not with Nucleosomal Histones Proteins Controls Leishmania infantum Infection in Hamsters | Several intracellular Leishmania antigens have been identified in order to find a potential vaccine capable of conferring long lasting protection against Leishmania infection. Histones and Acid Ribosomal proteins are already known to induce an effective immune response and have successfully been tested in the cutaneous leishmaniasis mouse model. Here, we investigate the protective ability of L. infantum nucleosomal histones (HIS) and ribosomal acidic protein P0 (LiP0) against L. infantum infection in the hamster model of visceral leishmaniasis using two different strategies: homologous (plasmid DNA only) or heterologous immunization (plasmid DNA plus recombinant protein and adjuvant).
Immunization with both antigens using the heterologous strategy presented a high antibody production level while the homologous strategy immunized group showed predominantly a cellular immune response with parasite load reduction. The pcDNA-LiP0 immunized group showed increased expression ratio of IFN-γ/IL-10 and IFN-γ/TGF-β in the lymph nodes before challenge. Two months after infection hamsters immunized with the empty plasmid presented a pro-inflammatory immune response in the early stages of infection with increased expression ratio of IFN-γ/IL-10 and IFN-γ/TGF-β, whereas hamsters immunized with pcDNA-HIS presented an increase only in the ratio IFN-γ/ TGF-β. On the other hand, hamsters immunized with LiP0 did not present any increase in the IFN-γ/TGF-β and IFN-γ/IL-10 ratio independently of the immunization strategy used. Conversely, five months after infection, hamsters immunized with HIS maintained a pro-inflammatory immune response (ratio IFN-γ/ IL-10) while pcDNA-LiP0 immunized hamsters continued showing a balanced cytokine profile of pro and anti-inflammatory cytokines. Moreover we observed a significant reduction in parasite load in the spleen, liver and lymph node in this group compared with controls.
Our results suggest that vaccination with L. infantum LiP0 antigen administered in a DNA formulation could be considered a potential component in a vaccine formulation against visceral leishmaniasis.
| Visceral leishmaniasis caused by Leishmania infantum is the most severe form of leishmaniasis. The disease is fatal if not treated and there is no vaccine available for human use. In the search for potential antigens, the protective ability of conserved parasite protein families such as L. infantum histones (HIS) and acidic ribosomal (LiP0) antigens were successfully tested in the mouse model of cutaneous leishmaniasis. Here, we evaluate HIS and LiP0 antigens using two different immunization strategies in the hamster model of visceral leishmaniasis. Hamsters are highly susceptible to L. infantum infection and we demonstrate that immunization with LiP0, but not HIS, protects against the fatal outcome of visceral leishmaniasis. Immunization with LiP0 was able to induce an increased expression of IFN-γ in detriment of IL-10 and TGF-β in the draining lymph node before infection creating an inhospitable environment for parasite growth. Following challenge, a reduced parasite load in the lymph node, spleen and liver of LiP0 immunized hamsters was detected five months after challenge. These findings suggest that LiP0 used in a DNA formulation could be considered a potential component in a vaccine formulation against visceral leishmaniasis.
| Leishmaniasis is a parasitic disease caused by protozoan from the genus Leishmania transmitted by the bite of infected sand flies. Leishmaniasis is one of the six major tropical diseases targeted by the World Health Organization [1]. The disease has a broad spectrum of clinical manifestations, from cutaneous, self-limited skin lesions to a visceral form of the disease. Visceral leishmaniasis (VL) caused by Leishmania infantum in the New World is the most severe form of disease characterized by hepatosplenomegaly, fever with a high mortality rate if not treated [2].
Although extensive research has been performed to identify an antigen able to elicit a long lasting protection against infection, there is still no successful vaccine available for human leishmaniasis. The majority of Leishmania vaccine candidates tested are able to induce humoral and/or cellular immune responses. However, the immune response derived is not able to induce protection and may contribute to pathology exacerbation [1].
Several secreted and surface Leishmania antigens have been tested, targeting virulence factors or molecules important for parasite invasion but most of these candidates resulted in a short-lived or partial protection [3]. On the other hand, intracellular house-keeping proteins are able to modulate the host immune response because they do not undergo selective pressure by the immune response [4]. During infection, these molecules are released after the destruction of intracellular amastigotes by activated macrophages but they can be also excreted by non-classical secretion pathways [5,6]. Several intracellular antigens such as heat shock proteins, ribosomal proteins and histones have been investigated as potential vaccine candidates against different species of Leishmania [3]. Histones (HIS) are important structural proteins in the organization and regulation of genes. There are four main classes of histones (H2A, H2B, H3, and H4) that are responsible for the composition of the Leishmania nucleosome [7]. Iborra et al demonstrated that immunization with DNA plasmid coding for nucleosomal histones plus CpG adjuvant was able to significantly reduce lesion size after challenge with L. major in BALB/c mice [8]. More recently, Carneiro et al. observed that mice immunized with either a plasmid DNA cocktail composed of four different Leishmania histones or with a combination of the DNA cocktail followed by the corresponding recombinant proteins, resulted in the absence of infected macrophages at the site of challenge with L. braziliensis in the presence of sand fly saliva. The protective response was associated with increased expression of IFN-γ and down regulation of IL-4 at the infection site [9].
Another immunodominant antigen is the Leishmania infantum acidic ribosomal protein (LiP0), a structural component of the large ribosome subunit that has been recognized by sera from both patients and dogs infected by L. infantum [10,11]. Also, it has been demonstrated that ribosomal protein (P0) was able to stimulate proliferation and IFN-γ secretion of a T-cell clone established from a human donor by stimulation with paraformaldehyde fixed promastigotes [12]. Using an immunization strategy that included LiP0 plasmid DNA vaccination and/or with LiP0 recombinant protein plus CpG, Iborra et al. showed the induction of partial protection against L. major infection in BALB/c mice [13]. However, C57BL/6 mice immunized with the same strategy was able to significantly reduce parasite load controlling lesion development [14]. The induction of IFN-γ was related with protection against Leishmania infection in these models. Although not demonstrated in Leishmania, it has been shown that P0 can be located in the surface of some protozoan parasites from the genus Plasmodium and Toxoplasma [15]. Its location on the cell surface of these organisms underlies the protective responses elicited by different vaccines based on P0 antigen. Thus, the generation of humoral responses to Neospora caninum P0 induces protection against neosporosis and toxoplasmosis [16]. Similarly, an experimental vaccine based on the carboxy-terminal domain of Plasmodium falciparum P0 was able to induce humoral responses that protects mice against malaria [17].
However, the immunization strategy used can be imperative to improve immunogenicity. The heterologous prime-boost strategy is used combining different formulations of the same antigen and has been effective in immunizations against cutaneous and visceral leishmaniasis. Dogs immunized with LACK using the heterologous strategy (DNA/protein) presented protection against VL caused by L. infantum [18]. Similarly, Iborra et al. (2005) used the same strategy with pcDNA3-LiP0 followed by recombinant protein (rLiP0) that protected C5BL/6 mice against challenge with L. major [14]. Moreover, administration of CpG ODN with antigens from different pathogens has been shown to induce a strong Th1 immune response [19].
Therefore, based on the protective potential of antigens belonging to conserved protein families used in the cutaneous leishmaniasis mouse model, we hypothesized that immunization with HIS and LiP0 antigens could also protect hamsters against the fatal outcome of L. infantum infection. In addition, we also compared HIS and LiP0 antigens using two different immunization strategies: plasmid DNA only (homologous) or plasmid DNA and recombinant protein plus CpG ODN (heterologous).
Male Syrian golden hamsters (Mesocricetus auratus) six to eight weeks old were obtained from Centro de Pesquisa Gonçalo Moniz/ Fundação Oswaldo Cruz (FIOCRUZ) animal facility. All animal work was conducted according to the Guidelines for Animal Experimentation of the Colégio Brasileiro de Experimentação Animal and of the Conselho Nacional de Controle de Experimentação Animal. The local Ethics Committee on Animal Care and Utilization (CEUA) approved all procedures involving animals (CEUA—Centro de Pesquisas Gonçalo Muniz—CPqGM/FIOCRUZ—L-IGM-011/09 and 005/2011.
DNA plasmid coding for Leishmania infantum acid ribosomal protein (pcDNALiP0) and the histones pcDNAHIS (pcDNA3LiH2-H3 and pcDNA3LiH2B-H4) were cloned and purified as previously described [13,14].
Immunization experiments were carried out in groups of 15 hamsters. In the homologous immunization experiments, hamsters were inoculated three times intramuscularly (i.m.) with 100 μg of DNA of pcDNA3-LiP0, pcDNA3 HIS (50 μg of each plasmid) or pcDNA3 plus 1nM of CpG ODN 1826 (18–24 pb—5´TCC ATG ACG TTC CTG ACG TT-3´ mol wt 6364,1g/mol) (empty plasmid) in a total volume of 50 μL. In the heterologous strategy, hamsters received two inoculations of DNA followed by one intradermal (i.d.) inoculation of recombinant protein (5 μg of each rHIS or 10 μg rLiP0 protein) plus 1 nM of CpG ODN 1826 in the right ear [20]. In all groups, hamsters were inoculated at 2-week intervals. Two weeks after the last immunization 15 hamsters per group were euthanized to collect draining lymph nodes and sera to evaluate the cellular and humoral response induced by the immunization.
Fifteen days after last immunization and two and 5 months after infection a sample of blood was collected from 6–8 hamsters per group by retrorbital plexus and sera was obtained after centrifugation for 1500 rpm for 5 minutes and stored at-4°C until use. Briefly, to measure specific antibody responses by ELISA, standard ELISA plates were coated overnight at room temperature with 100μL of SLA (10 μg/mL), rLiP0 (2μg/mL) or rHIS (1 μg/mL) in PBS. Serum samples were added at dilutions of 1:100. Following a washing step, a goat anti-hamster IgG alkaline phosphatase conjugate (1:1,000 and 1:2,000, Sigma, Missouri, USA) was added and incubated for one hour. The wells were then re-washed, substrate and chromogen (p-nitrophenyl phosphate; Sigma, USA) were added, and absorbance was recorded at 405 nm on a SpectraMax 190 spectrophotometer (Molecular Devices, USA) automatic micro plate reader.
Leishmania infantum (MCAN/BR/00/BA262) promastigotes isolated from a naturally infected dog (Bahia State, Brazil) were cultured in Schneider’s medium (LGC, Brazil) supplemented with 10% of inactivated FBS (fetal bovine serum) (Gibco, USA), 2 mM L-glutamine, 100 IU/ml penicillin, 1% streptomycin (Gibco, USA).
Fifteen days after the last immunization, hamsters were inoculated by intradermal route, in the left ear with 105 stationary phase promastigotes plus 0.5 pair of sonicated salivary glands (SGH) from female Lutzomyia longipalpis, using a 29-gauge needle (BD Ultra-Fine) in 20uL of saline. Salivary glands were dissected from 5- to 7-day-old females and stored in endotoxin-free PBS at −70°C. Salivary glands were sonicated (Sonifer 450 homogenizer, Branson, Danbury, Connecticut), and afterwards centrifuged at 12,000 g for 5 minutes. Supernatant was collected and used immediately.
Parasite load was determined 2 and 5 months post-infection using the quantitative Limiting Dilution Assay (LDA) as described by Titus et al [21]. Briefly, infected liver, spleen and retromaxillar draining lymph nodes were aseptically removed from individual hamsters. Tissues were homogenized and diluted in Schneider’s Insect Medium (Sigma, St. Louis, MO) supplemented with 10% heat inactivated fetal bovine serum (Gibco, USA), 100 U of penicillin/ml and 100 mg/ml of streptomycin. Homogenate samples were serially diluted into 96-wells plates containing biphasic blood agar (Novy-Nicolle-McNeal) medium and incubated for one week at 23°C when the presence of viable parasites was determined. Parasite burden in the tissue was calculated applying ELIDA software [21].
Using Trizol reagent (Invitrogen, USA), total RNA was extracted from the draining retromaxillar lymph nodes two weeks after the last immunization, and from the spleen obtained two and five weeks after infection. First-strand cDNA synthesis was performed with 1–2 μg of RNA in a total volume of 25 μL using Super Script II (Gibco, Carlsbad, CA, USA). DNA was amplified in the thermocycler (Mastercycler gradient—Eppendorf, USA) with an initial pre- incubation at 72°C for 5 minutes, followed by amplification of the target DNA at 42°C for 50 minutes. A standard curve was generated for each set of primers and efficiency of each reaction was determined. The expression levels of genes were normalized to GAPDH levels. Results are expressed in fold change over control. Oligonucleotide primers used were: GAPDH (reverse 5’- CTGACATGCCGCCCTGGAG-3’ and forward 3’-TCAGTG- TAGCCCAGGATGCC-5’); IFN-γ (reverse 5’-GAAGCTCAC- CAAGATTCCGGTAA-3’ and forward 3’-TTTTCGTGACA- GGTGAGGCAT-5’); IL-10 (reverse 5’-AGACGCCTTTCTC- TTGGAGCTTAT-3’ and forward 39-GGCAACTGCAGCGC- TGTC-5’); and TGF-β (reverse 5’-GCTACCACGCCAACTTC- TGTC-3’ and forward 3’-TGTTGGTAGAGGGCAAGG-5’).
For histology, five months after infection spleen and liver fragments of four hamsters from each group were fixed in 10% phosphate-buffered formalin and embedded in paraffin. Four-micrometer sections were stained with hematoxylin–eosin and studied by optical microscopy.
Experiments were repeated three times with five hamsters per group per time point. Comparisons among immunized and non-immunized control groups were done by one-way ANOVA (Kruskal-Wallis) analysis with Dunn’s post-test. Results were considered statistically significant when p≤0.05. All statistical analysis was done using Graph Pad 5.0 software program.
To evaluate production of anti-HIS or anti-LiP0 IgG antibodies we collected sera from immunized hamsters fifteen days after the last immunization. We observed that only hamsters that were immunized with heterologous strategy showed significantly higher titers of antibodies. There was a significant increase of IgG in the group that received DNA plasmid coding for HIS followed by booster of nucleosomal histone recombinant protein plus CpG adjuvant (pcDNA HIS-rHIS+CpG) compared to non-immunized hamsters (Fig. 1A). Similar results were observed regarding antibody production in hamsters immunized with LiP0, using the same heterologous strategy. There was a significantly higher production of anti-LiP0 IgG in vaccinated hamsters (pcDNA-LiP0/ rLiP0+CpG) compared to the animals that received empty plasmid and saline, the control groups (Fig. 1B).
In order to verify the cellular immune response induced by the immunization with HIS and LiP0, expression of IFN-γ, IL-10, and TGF-β was evaluated in retromaxillar draining lymph nodes 15 days after immunization. We observe a significant increase in the ratio of IFN-γ/IL-10 expression in the lymph nodes of animals immunized with HIS and LiP0 using the homologous immunization strategy with DNA plasmids (Fig. 2A). Conversely, there was no difference in the IFN-γ/TGF-β ratio in the HIS immunized group using both immunization strategies (Fig. 2B). Interestingly, we observed that the IFN-γ/TGF-β ratio was significantly higher in hamsters immunized with DNA plasmid (pcDNA-LiP0) using homologous strategy compared with the control group (Fig. 2B).
In order to verify if the immune responses resulting from immunization with acidic ribosomal protein and nucleosomal histone antigens using homologous and heterologous immunization strategies, are able to protect against L. infantum infection, immunized hamsters were challenged with 105 L. infantum plus SGH, trying to mimic natural transmission [22].
Two months after challenge, hamsters immunized with HIS using the homologous strategy showed a significant parasite load reduction in the liver (Fig. 3C) but not in the retromaxillar lymph nodes or spleen compared to control groups (Fig. 3A and B).
On the other hand, we observed a significant reduction in the lymph node, spleen and liver parasite loads in LiP0 DNA plasmid (pcDNA3-LiP0) immunized group, compared with controls five months after challenge (Fig. 3A, B and C). Interestingly, parasite burden in the spleen (Fig. 3B) (p = 0.0016) and liver (Fig. 3C) (p = 0.0001) of hamsters immunized with the LiP0-based DNA vaccine were lower at 5 months than 2 months after infection, indicating that the vaccine was able to induce a long term leishmanicidal response.
To access the humoral immune response after infection, we measured production of anti-LiP0, HIS and SLA IgG antibodies. For this purpose, we collected sera from infected hamsters at two and five months following challenge with L. infantum plus SGS.
We observed an increase in anti-HIS IgG in all groups immunized with HIS especially at five months after infection (Fig. 4A). Conversely, LiP0 immunized groups did not present any significant levels of anti-LiP0 IgG levels two and five months following infection when compared with controls (Fig. 4B).
Interestingly, animals immunized with LiP0 homologous strategy showed a significantly lower anti-SLA IgG levels at five months after infection compared to HIS immunized and controls groups (Fig. 4C).
To evaluate the cellular immune response of hamsters challenged with L. infantum, we investigated the expression of cytokines in the spleen two and five months following infection.
Two months after challenge hamsters immunized with DNA-HIS (homologous strategy) presented an increased expression ratio of IFN-γ/IL-10 when compared with the non-immunized control group (p = 0.0253) (Fig. 5A). Expression ratio of IFN-γ/TGF-β was increased in the pcDNA-HIS (p = 0.0003), and pcDNA-HIS/rHIS+CpG (p = 0.018) immunized groups compared with saline control suggesting that DNA-HIS immunized hamsters presented a pro-inflammatory immune response in the early stages of infection (Fig. 5B). On the other hand, hamsters immunized with LiP0 using independent of strategy employed did not present any increase in the expression ratio of IFN-γ/IL-10 or IFN-γ/TGF-β (Fig. 5A and B).
Five months after infection, hamsters immunized with HIS using both immunization strategies showed an increased expression ratio of IFN-γ/IL-10 (p<0.05) suggesting a pro-inflammatory immune response (Fig. 5C). Concerning LiP0 immunization, there was an increase in the expression ratio of IFN-γ/TGF-β in the spleen of pcDNA-LiP0/rLiP0+CpG immunized hamsters (heterologous strategy) compared with the control group (p<0.0101) (Fig. 5D) suggesting that this strategy induce a delayed cellular immune response.
Five months after infection spleen and liver samples were collected from four hamsters per group for histopathological analysis. In the spleen, we observed disorganized follicles with or without germinal center, macrophage aggregates with L. infantum amastigotes inclusions, epithelioid cells present in the red and white pulp, as well as granulomas in the groups immunized with HIS or LiP0. However, hamsters immunized with HIS displayed a higher loss of splenic structure… Animals immunized with LiP0 presented smaller macrophage aggregates. Fig. 6 shows the granulomas present in the control (A) and LiP0 (B) immunized hamsters after infection (arrows).
At the same time point, liver from control and HIS immunized groups showed accentuated intrasinusoidal leukocytosis with portal mononuclear infiltrate. Areas of necrosis and granulomas containing giant cells filled with inclusions of amorphous material, few Schaumann bodies were observed (arrows in Fig. 6C). Although HIS and LiP0 immunized groups presented comparable intense inflammatory infiltrate the majority of hamsters immunized with LiP0 antigens presented fewer granulomas (arrows in Fig 6D). These data were summarized in Table 1, showing the number and percentage of animals displaying the histopathological alterations.
Leishmania nucleosomal histones (HIS) and acidic ribosomal protein P0 (LiP0) elicit defined immune responses and have been the focus of investigation as potential vaccines against cutaneous leishmaniasis [3]. Herein, we further explore their antigenic potential against the visceral form of leishmaniasis. Using the susceptible hamster model of VL we show that both molecules are immunogenic but only immunization with LiP0 is able to control disease progression.
Following immunization a notable difference was already observed on the profile of cytokine expression in the draining lymph nodes. The expression ratio of IFN-γ/IL-10 of hamsters immunized with pcDNA-LiP0 and pcDNA-HIS was significantly higher compared with control groups. Interestingly, significant differences in the expression ratio of IFN-γ/TGF-β was observed only in hamsters immunized with LiP0 using the homologous strategy. Similar results were observed previously where production of IFN-γ but not IL-4 was detected in the spleen of mice immunized with nucleosomal HIS antigens [8]. Another study also demonstrated IFN-γ production upon stimulation of splenocytes and lymph node cells in vitro with rLiP0 after immunizations with pcDNA-LiP0 [13].
In the group immunized with empty vector we detected higher ratios of IFN-γ /IL-10 and IFN-γ/TGF-β two months after infection, suggesting that the infection could be responsible for this increase. Additionally, we did not detect the same levels for the control group (saline). The empty vector contains CpG motifs that inespecifically increase production of inflammatory cytokines, such as IFN-γ. It is important to emphasize that evaluation of cytokines in the spleen was done ex vivo that does not comprise specific in vitro restimulation. Therefore, we can speculate that the higher number of parasites present in the spleen could contribute to the inflammatory pattern observed. However, IFN-γ expression was not sustained, five months after infection cytokines ratios returned to basal levels that correlated with a high parasite load in the liver and spleen. The same interpretation could be applied to the group immunized with HIS, concerning the restimulation in vitro, where at two months, immunization with HIS induce a lower IFN-γ/IL-10 ratio. However, five months after infection, the number of parasites was high in the spleen and possibly, in an attempt to control parasite growth, there was an increase in IFN-γ expression as observed in other reports in the literature [22]. Interestingly, the ratio of IFN-γ/TGF-β presented opposite results, where we observed an increase in this ratio at two months post-infection and a decrease five months after infection. We know that IL-10 and TGF-β play similar roles in the inhibitory responses and we can speculate these cytokines might alternate in the balance of pro and anti-inflammatory responses after infection [23,24].
Moreover, the expression ratio of IFN-γ/IL-10 and IFN-γ/TGF-β in the animals immunized with LiP0 antigens did not present any significant increase, independently of the immunization strategy used. Interestingly, these animals displayed a significant reduction in the parasite load in lymph nodes, spleen and liver five months after infection, suggesting that this modulation in the immune response is important for infection control. Hamsters immunized with LiP0 also presented fewer infected granulomas and macrophage aggregates in the spleen and liver and lower anti-SLA IgG levels after infection compared to the HIS immunized group indicating control of parasite replication.
A significant production of IgG antibodies was also detected following immunization with both antigens (LiP0 or HIS). Similarly, Iborra et al. also detected specific levels of IgG and IgG2a in mice immunized with LiP0 and HIS antigens. Antibodies against a LiP0 ribosomal protein epitope were also detected in dogs naturally infected with L. infantum [8,11,14]. Specific total IgG production was only detected when the heterologous immunization strategy was employed. Indeed, vaccination strategies using exclusively DNA are known to elicit low IgG production as previously demonstrated [25]. Interestingly, five months after infection, anti-HIS and anti-LiP0 IgG levels increased in all groups with no significant differences between immunization strategies.
The significant parasite load reduction observed in the pcDNA-LiP0 immunized group after five months could be a result of the increased expression ratio of IFN-γ/IL-10 and IFN-γ/TGF-β in the lymph nodes before challenge. The induction of a predominant Th1 immune response following immunization could have an impact on parasite establishment and disease development. The importance of a Th1 environment established before challenge was clearly demonstrated in hamsters immunized with LJM19, a vector salivary molecule, Protection was associated with a considerably higher expression of IFN-γ/TGF-β ratio in the spleen compared with controls [22].
Using the mouse model, Carrion et al. demonstrated that immunization with a pcDNA-HIS vaccine and subsequently challenged with L. infantum did not result in parasite load reduction [26]. Another study demonstrated different results with parasite load reduction in the spleen and liver of mice immunized with HIS pulsed dendritic cells associated with CpG after L. infantum infection. Protection was associated with a potent polarized Th1 type immune response elicited by the immunostimulatory ability of HIS pulsed dendritic cells [27]. More recently, immunization with recombinant histones proteins from L. donovani was able to confer protection against L. donovani infection [28].
In the cutaneous leishmaniasis model, the protective ability of HIS was demonstrated using both homologous and heterologous immunization strategies. BALB/c mice immunized with HIS, using both experimental approaches and challenged with L. braziliensis, were able to control lesion development, decreased parasite burden in lymph nodes and absence of infected macrophages at site of infection [9]. In another study, DNA-HIS vaccines were able to reduce the number of L. major parasites in the draining lymph node and spleen compared to controls [8]. Contrasting results between different animal models and clinical forms were also observed after immunization with KMP11, a different parasite antigen. Immunization of hamsters with KMP11 was able to confer protection against L. donovani and L. infantum infection. The same protective response was not observed when encapsulated KMP11 was used to immunize BALB/c mice against L. braziliensis infection [29–31]. These contradictory findings concerning visceral and cutaneous leishmaniasis models could be explained by the remarkable differences in disease development and the host’s immune response depending on the animal model used. Visceral leishmaniasis is a systemic chronic disease that affects many organs while cutaneous leishmaniasis is characterized by a restricted skin lesion. Although the mouse model has been extensively used to study cutaneous disease, infection with L. infantum is self-limiting. The hamster model of VL is a well described model of susceptibility, with progressive disease that more closely mimics the severity of infection in humans and dogs [32,33]. Assessment of long term memory is important when evaluating a potential vaccine candidate. Unfortunately, there are no immunological tools available to investigate and characterize memory cells in hamsters.
Vaccines based on combination of different antigen candidates have been shown to improve protection. In fact, a recombinant Q protein formed by genetic fusion of five parasite intracellular antigens has been successfully tested in dogs [34–36]. Although this is imperative when considering a new vaccine, the focus of this study was to initially test HIS and LiP0 antigens independently. Besides that, combination of vector salivary antigens, such as LJM19, and parasite antigens could improve protection compared to immunization with isolated antigens. In order to select the antigens that could be used in combination its necessary to initially evaluate their immunogenicity and protection potential independently. Interestingly, more recent data published by our group have showed that combination of LJM19 plasmid with KMP11, a parasite antigen, was not able to improve protection resulting from immunization with LJM19 or KMP11 alone in the hamster model[37].
There are two licensed vaccines for canine VL available in Brazil. The fucose mannose ligand of Leishmania donovani (Leishmune) [38,39] and the Leishmania amastigote recombinant A2-antigen (Leish-Tec) [40]. In a recent study, no significant differences in the rate of seroconversion, clinical signs, parasitism and parasite transmission to the vector were observed in dogs vaccinated with any of the two vaccines [41]. Although both vaccines showed promising results, large-scale field studies are necessary to validate their inclusion in a mass control strategy for canine VL.
Taken together, the data shown herein indicates that although L. infantum LiP0 and HIS antigens were immunogenic in hamsters, only LiP0 antigen was able to confer a significant degree of protection against L. infantum using the susceptible hamster model of VL. Importantly, we also noted that the immunization strategy used is critical when a potential vaccine candidate is being tested. Indeed, immunization with pcDNA-LiP0 followed by rLiP0 boost (heterologous strategy) resulted only in short-term protection after two months. Our results suggest that L. infantum LiP0 antigen administered in a DNA formulation (homologous strategy) is a potential vaccine candidate and further investigation of the protective mechanism could improve L. infantum LiP0 protective effect against visceral leishmaniasis.
|
10.1371/journal.ppat.1004081 | In Vivo Ligands of MDA5 and RIG-I in Measles Virus-Infected Cells | RIG-I-like receptors (RLRs: RIG-I, MDA5 and LGP2) play a major role in the innate immune response against viral infections and detect patterns on viral RNA molecules that are typically absent from host RNA. Upon RNA binding, RLRs trigger a complex downstream signaling cascade resulting in the expression of type I interferons and proinflammatory cytokines. In the past decade extensive efforts were made to elucidate the nature of putative RLR ligands. In vitro and transfection studies identified 5′-triphosphate containing blunt-ended double-strand RNAs as potent RIG-I inducers and these findings were confirmed by next-generation sequencing of RIG-I associated RNAs from virus-infected cells. The nature of RNA ligands of MDA5 is less clear. Several studies suggest that double-stranded RNAs are the preferred agonists for the protein. However, the exact nature of physiological MDA5 ligands from virus-infected cells needs to be elucidated. In this work, we combine a crosslinking technique with next-generation sequencing in order to shed light on MDA5-associated RNAs from human cells infected with measles virus. Our findings suggest that RIG-I and MDA5 associate with AU-rich RNA species originating from the mRNA of the measles virus L gene. Corresponding sequences are poorer activators of ATP-hydrolysis by MDA5 in vitro, suggesting that they result in more stable MDA5 filaments. These data provide a possible model of how AU-rich sequences could activate type I interferon signaling.
| RIG-I-like receptors (RLRs) are helicase-like molecules that detect cytosolic RNAs that are absent in the non-infected host. Upon binding to specific RNA patterns, RLRs elicit a signaling cascade that leads to host defense via the production of antiviral molecules. To understand how RLRs sense RNA, it is important to characterize the nature and origin of RLR-associated RNA from virus-infected cells. While it is well established that RIG-I binds 5′-triphosphate containing double-stranded RNA, the in vivo occurring ligand for MDA5 is poorly characterized. A major challenge in examining MDA5 agonists is the apparently transient interaction between the protein and its ligand. To improve the stability of interaction, we have used an approach to crosslink MDA5 to RNA in measles virus-infected cells. The virus-infected cells were treated with the photoactivatable nucleoside analog 4-thiouridine, which is incorporated in newly synthesized RNA. Upon 365 nm UV light exposure of living cells, a covalent linkage between the labeled RNA and the receptor protein is induced, resulting in a higher RNA recovery from RLR immunoprecipitates. Based on next generation sequencing, bioinformatics and in vitro approaches, we observed a correlation between the AU-composition of viral RNA and its ability to induce an MDA5-dependent immune response.
| The retinoic acid inducible gene I (RIG-I)-like receptor (RLR) proteins are key players in innate immunity and act by recognizing viral RNA (vRNA) in the cytosol. The RLR family consists of the members retinoic acid inducible gene I (RIG-I), melanoma differentiation associated protein 5 (MDA5), and laboratory of genetics and physiology 2 (LGP2) [1]–[3]. In vitro studies have shown that RIG-I and MDA5 recognize the majority of viruses in a complementary manner. While many negative-strand RNA viruses like rabies and influenza viruses are predominantly sensed by RIG-I, picornaviruses are predominantly recognized by MDA5. The observed preferences are, however, unlikely to be exclusive and the exact role of LGP2 still needs to be investigated [4]–[9]. In case of MDA5, a minor contribution to recognition of measles, rabies, vesicular stomatitis and Sendai virus has been reported [10]–[13].
The RLR proteins belong to the DExD/H-box ATPases sharing a central ATP-dependent helicase domain and a C-terminal regulatory domain (RD) that is responsible for initial RNA binding. In addition, RIG-I and MDA5 possess N-terminal tandem caspase activation and recruitment domains (CARDs) that are responsible for downstream signaling transduction [2], [14], [15]. Several crystal structures of RIG-I have shown that, in the absence of virus, the protein exists in an auto-inhibited state where the RD domain folds back to the CARDs, thereby shielding them from the cytosol. Upon viral infection and initial vRNA binding, the protein undergoes large conformational changes leading to the interaction with the mitochondrial associated signaling protein (MAVS) [16]–[19]. This leads to the activation of a downstream signaling cascade and finally to the induction of type I interferon (IFN) expression and the establishment of an anti-viral state. Although the exact nature of RLR ligands is not yet fully understood, several studies report that RIG-I preferentially binds to relatively short (between 25 to 1000–2000 bp) 5′-triphosphate double-stranded RNAs (5′-triphosphate dsRNA) like those of Sendai virus (SeV) defective interfering (DI) particles [20]–[23]. In contrast, MDA5 seems to have a preference for long (more than 1000–2000 bp) dsRNA stretches [24], [25]. Upon binding to dsRNA, MDA5 is thought to cooperatively form polar helical filaments leading to association with MAVS and activation of the downstream signaling cascade [26]–[28].
Viruses have developed numerous strategies to evade the immune system. For instance, viruses of the paramyxovirus family (e.g. measles, parainfluenza, Sendai and Nipah viruses) encode V inhibitor proteins that specifically bind to MDA5 and LGP2, but not always to RIG-I [29]–[31]. By determining the structure of MDA5 in complex with parainfluenza virus V-protein, we previously showed that the viral protein unfolds the ATPase domain of MDA5. This leads to the disruption of the MDA5 ATP-hydrolysis site and prevents RNA bound MDA5 filament formation [32].
One of the remaining key questions in this field is how RLR proteins are able to distinguish between self and non-self RNA in the cytosol. Recently, several studies showed that 5′-triphosphate RNA is not the only RNA ligand for RIG-I. Specific poly U/C-rich regions within certain viral genomes seem to contribute to efficient recognition by the protein [33], [34]. In case of MDA5, it is not known which features of vRNA are required in order to induce an immune response. Expression of subgenomic and subgenic RNA from parainfluenza virus 5 (PIV5) indicated that MDA5 recognizes a specific region within the L mRNA [35]. For picornaviruses, it is speculated that MDA5 binds to long dsRNA that represents replicative intermediates composed of the positive genome and the negative antigenome [36]. These studies were, however, based on in vitro transfection experiments and it has so far not been possible to isolate a natural RNA ligand for MDA5 directly from virus-infected cells.
In this study we combined different methods, including RNA-protein crosslinking and deep sequencing, to investigate in vivo RNA ligands for RLR proteins from virus-infected cells. Based on the crosslinking we were able to co-purify immunostimulatory RNA in a RIG-I and MDA5 dependent manner from measles virus (MeV)-infected cells. Deep sequencing and bioinformatics analysis revealed that RIG-I and MDA5 bind RNA of positive polarity originating from the L gene of the MeV genome. In addition, RIG-I binds to the 5′ ends of genomic and antigenomic RNAs, which probably represent 5′-triphosphate RNA, and are therefore not recognized by MDA5. Furthermore, we showed that RIG-I, but not MDA5, binds RNA of negative polarity, indicating that MDA5 does not efficiently recognize the MeV genome. Based on bioinformatics analysis, we observed a correlation between MDA5-enriched RNA sequences and the AU content and this was confirmed by in vitro transcription assays. In summary, we report the isolation of MDA5-associated RNA from virus-infected cells and the discovery of in vivo occurring activating viral RNA ligands for MDA5.
Several in vitro studies showed that MDA5 preferably recognizes long dsRNA stretches [24], [25]. However, it is still unclear if the protein has a preference for specific RNA sequences. The main reason for this may lie in the weak interaction between the protein and its ligand resulting in very poor RNA levels that co-purify from MDA5 immunoprecipitates. In order to address this problem, we established an RNA-protein crosslinking approach adapted from the PAR-CLIP (Photoactivatable-Ribonucleoside-Enhanced Crosslinking and Immunoprecipitation) methodology [37]. With this approach, we intended to improve RNA recovery from RLR immunoprecipitates in the context of a viral infection. For validation of the method, we compared the crosslinking approach with a conventional pull-down technique previously used for the identification of SeV DI particles as potent RIG-I inducers [20]. We infected A549 human lung carcinoma cells with SeV at a high multiplicity of infection (MOI) in the presence of 4-thiouridine (4SU) and allowed infection to occur over 24 h. A part of the cells was then exposed to 365 nm UV light and endogenous RIG-I was immunopurified (Figure 1a). The recovered RNA was isolated and subjected to quantitative PCR (qPCR) analysis and immunoactivity experiments. The data indicate that treatment of cells with 4SU and exposure to 365 nm UV light lead to a reduction of immunostimulatory activity of RIG-I-associated RNA to 50% (Figure 1b). However, the results of qPCR analysis showed that the crosslinking approach yields a quantitatively improved RNA recovery, with an increase of 50% in SeV DI particles in comparison to the non-crosslinking approach (Figure 1c and d). Furthermore, we confirmed that treatment of cells with the photoreactive nucleoside does not affect cell viability or virus replication (data not shown). Taken together, our data indicate that the crosslinking technique is a promising tool to study in vivo occurring RNA ligands for RLR proteins.
Next, we validated the crosslinking approach on cells that were infected with a variety of viruses, including negative-stranded (−) RNA viruses (MeV [38] and rabies [39]) and positive-stranded (+) RNA viruses (Encephalomyocarditis virus (EMCV [40]) and Mengo virus [41]). In all cases, we infected A549 cells at an MOI of 1.0 in the presence of 4SU. Cells were crosslinked 24 h post infection (hpi) and RIG-I and MDA5 were immunopurified. The recovered RNA was subjected to immunoactivity experiments. Based on the data, we concluded that immunoactive RNA was co-purified in a RIG-I- and MDA5-dependent manner from MeV-infected cells. This induction was significant in comparison to the negative control (Figure 2). In the case of RIG-I-associated RNA, we obtained an immunostimulatory effect that was 2600-fold higher in comparison to the control. For MDA5, we observed an 800-fold induction. The data show that the approach yields RIG-I- and MDA5-specific immunoactive RNA from MeV-infected cells in a RIG-I- and MDA5-dependent manner. Although we detected significant immunostimulatory activity for RLR-associated RNAs from MeV-infected cells, the experimental set up is currently unsuitable for the isolation of RLR RNA ligands from the other viruses (Figure S1). The reason for this may lie in the heterogeneity and the need for precise timing of viral replication cycles or in the efficiency of 4SU incorporation and crosslinking. Utilization of this technique for other viruses may require adjustment of parameters, such as the time points of 4SU addition, crosslinking and harvesting after infection.
Based on the above-mentioned results, we focused our studies on MeV, which belongs to the order of Paramyxoviridae. MeV has a single-stranded RNA genome of negative polarity consisting of 15,894 nucleotides. It comprises six non-overlapping genes, which are flanked by small terminal non-coding regions known as leader (le) and trailer (tr) sequences. These sequences serve as promoter regions during viral replication and transcription [42], [43]. While the replication of the genome and antigenome is performed in a continuous process, viral transcription is carried out in a sequential manner, giving rise to an mRNA gradient declining in the 3′ to 5′ direction (Figure S2), as previously published [44]. Since (−) RNA virus polymerases eventually fail in transcription termination, they generate, in addition to monocistronic mRNAs, numerous alternative RNA species including read-through transcripts, such as leader-N, bi- or tricistronic mRNAs [45]. Furthermore, replication can give rise to abortive replication products and DI RNA with large internal deletions or copy-back genomes [46]. Due to the complex RNA composition of a virus-infected cell, the analysis of specific RNA ligands for RLR proteins is challenging.
In order to shed light on the exact nature of RIG-I and MDA5-associated RNAs derived from MeV-infected cells, we performed a deep sequencing analysis on isolated RNA species from co-immunopurifications with antibodies against endogenous RIG-I and MDA5. As a control, we used an antibody against GFP (GFP protein was not present). The MeV strain used for the studies presented here was a recombinant measles virus rescued from cDNA with the exact sequence of the Schwarz vaccine strain (Genbank AF266291.1) [38].
Obtained sequences were mapped to the MeV antigenome and the relative abundances of these sequences between RIG-I pull-down, MDA5 pull-down, and GFP pull-down were compared. Analysis of the reads showed that RIG-I and MDA5 bind to similar regions within the L gene-derived RNAs. In addition, RIG-I, but not MDA5, binds to RNAs derived from the 3′ and the 5′ ends of the MeV genome (Figure 3a and b). These regions probably represent le or trRNA generated in the course of replication or transcription. Additionally, internal genomic and antigenomic sequences found in the pull-downs could potentially originate from MeV DI particles [46]–[51]. To address this question, we performed a PCR analysis of RLR libraries in which we specifically amplified copyback DI RNA of MeV [47]–[49]. Indeed, we detected copyback DI particles not only in the RIG-I pull-down but also within RNA recovered from MDA5 immunoprecipitates (Figure S3). We did not find DIs in the GFP control pull-downs.
Consistent with previous work, the higher copy numbers of reads indicate that RIG-I binds MeV RNA with higher affinity than MDA5 [11]. This observation is in good agreement with the increased immunostimulatory activity of isolated RNA from RIG-I pull-down samples in comparison to MDA5. Regarding the immunostimulatory activity, RIG-I-associated RNA gives a 4-fold higher induction in comparison to MDA5-associated RNA (Figure 3d and e).
Based on the protocol used for cDNA library preparation, sequencing reads could be separated according to their strand orientation. During cDNA synthesis, adaptors were specifically ligated to the 3′ or 5′ ends, thereby keeping the information of strand specificity during the deep sequencing run. Separation of sequences revealed remarkable differences between both protein immunoprecipitations. RIG-I associated RNA sequences of positive polarity, which represent either antigenomic RNA or mRNA transcripts, are enriched in regions close to the 5′ end of the viral antigenome (leader) but also in distinct regions within the L gene. In contrast, sequences of negative polarity, representing the viral genome, are exclusively enriched in the 5′ end of the genome (trailer region) and in regions of the L gene (Figure 4a).
Analysis of MDA5-associated RNA revealed that sequences of positive polarity were enriched within the L gene originating from similar regions as (+) RNA from the RIG-I library (Figure 4b). In contrast to RIG-I, however, MDA5 did not bind to RNA sequences comprising the 5′ end of the antigenome or leader RNA. Comparison of (−) RNA from RIG-I and MDA5 libraries further revealed that, in contrast to RIG-I, MDA5 did not enrich sequences of negative polarity, including trailer sequences.
According to the analysis of strand specific enrichment, it appears that MDA5 does not bind vRNA of negative polarity that represents the MeV genome. Furthermore, the data evidently rule out the possibility that MDA5 recognizes RNA duplexes of (+) and (−) RNA that might represent replication intermediates, as previously suggested for a positive-strand RNA virus [36]. In fact, the result suggests that MDA5 binds (+) RNA that could either represent mRNA or the MeV antigenome.
To further validate the specificity of the accumulation of RIG-I and MDA5-associated RNA, we calculated specific read enrichments [52] of the RLR libraries compared to the control library (Figure S4). Enrichment (greater than 2× compared to the control library) of RIG-I-associated RNA of positive polarity can be found across the whole genome, whereas only few reads of negative polarity are enriched within the N and L segment. In contrast, enriched sequences of MDA5-associated RNA are exclusively present within the L segment of positive polarity, whereas no specific enrichment was observed for (−) RNA.
Based on the data, we observed a good correlation between the deep sequencing analysis and enrichment calculations, indicating that distinct regions within the MeV genome are indeed specifically enriched in a RIG-I- and MDA5-dependent manner in comparison to the control.
To independently validate the relative amount of RLR-associated RNA, qPCR amplification was performed. The obtained copy read numbers were normalized to the GFP negative control in order to compare the genomic segments in the RIG-I and MDA5 samples (Figure 5a). Analysis of relative abundances confirmed that RIG-I specifically enriches sequences from the 3′ and 5′ regions of the MeV genome, representing either antigenome or viral mRNA. Interestingly, the analysis showed that RIG-I-associated RNA from the genomic 3′ end most likely represents leader read-through transcripts or abortive replication products and not N mRNA. In MDA5 pull-downs, RNA was enriched in the case of the L mRNAs and partly in the case of H mRNAs, while no relevant copy numbers were obtained at other genomic positions. This is in good agreement with the results of the deep sequencing analysis, indicating that MDA5 indeed recognizes RNA originating from the L gene of the MeV genome. Furthermore, comparison of the relative copy numbers between RIG-I and MDA5 revealed remarkable differences between both proteins. The relative abundances in the RIG-I sample were up to 40-fold higher in comparison to MDA5. This observation again indicates that RIG-I has a higher affinity for MeV RNA sequences in comparison to MDA5. Our conclusion is further supported by immunoactivity experiments, where the relative immunostimulatory activity of RIG-I-associated RNA was 20-fold higher in comparison to MDA5 (Figure 5b and c).
To elucidate the exact nature of sequences enriched by RIG-I and MDA5 immunoprecipitations, we conducted a bioinformatics analysis. For this, the complete genome was divided into fragments of size 201 nt with a shifting window of 5 nt. Each sequence was folded in silico (RNAfold [53]) and several RNA primary and secondary structure features were analyzed. The analyzed parameters were set in relation to the mean coverage of sequencing reads from RIG-I and MDA5 pull-down experiments. Heat scatter plots indicate that sequences rich in AU correlate with a high mean coverage of sequencing reads in both the RIG-I (cor = 0.273, cor = 0.334) and MDA5 (cor = 0.358, cor = 0.348) libraries (Figure 6a and b). These data suggest that RIG-I and MDA5 preferably bind to AU-rich sequences originating from the viral genome. Although we further analyzed a variety of secondary structure parameters, including paired nucleotides and bulges, we did not see any other relevant correlation with the mean coverage of sequencing reads (Figure S5 and Figure S6).
To further confirm the obtained sequencing data, we generated 17 single-stranded, 200 nucleotide long in vitro transcripts (IVTs) covering different regions of the MeV antigenome (Table S1). RNAs were double-dephosphorylated in order to ensure that 5′-triphosphate groups were removed. For immunoactivity experiments IVTs were transfected into 293T ISRE-FF reporter cells. The stimulatory effect revealed a correlation of high read numbers from deep sequencing analysis and high stimulatory activity of the IVT sequences (Figure 7). According to the immunostimulatory experiment, we observed increased immunostimulatory activities for transcripts 8, 9, and 12 (Figure 7a). These transcripts correspond to regions at the 5′ end of the L gene, which is also the region with the highest copy numbers of reads (Figure 3). In general, IVTs representing regions within the L gene have higher immunostimulatory activity in comparison to the upstream genomic segments. This is in good agreement to the deep sequencing analysis. Furthermore, calculated Pearson correlations showed that the best correlation between maximal numbers of sequencing reads and the immunostimulatory activity of RNA transcripts can be found in the MDA5 sequencing data (cor = 0.526), while RIG-I and GFP samples showed less correlation (cor = 0.369 and cor = 0.217) (Figure 7b). In order to find a possible explanation for the different immunostimulatory potentials of IVTs, several characteristics of the transcripts were analyzed in silico. The obtained data revealed that the immunostimulatory potential correlates with the AU content of IVTs (cor = 0.599) (Figure 7d), which is consistent with the results from the deep sequencing analysis. Visualization of transcripts on an Agilent bioanalyzer RNA chip indicates that no higher-order structures due to the sequence composition were formed that might explain differences in immunostimulatory activity (data not shown).
In order to get a more general conclusion about the contribution of the AU content to the immunostimulatory potential of RNAs, in vitro transcripts from Mengo virus (Table S3) were tested for their immunostimulatory activity. The transcripts were generated according to the protocol for MeV RNA sequences. We again observed a correlation (cor = 0.583) of the AU content of the tested sequences and their immunostimulatory potential (Figure S 7a and b). These data are consistent with the in vitro analysis of MeV RNA sequences indicating that the AU composition of RNA might play a general role in activating RLR signaling.
Finally, we asked whether the ATP hydrolysis activity of MDA5 correlates with the immunostimulatory potential of the tested IVTs. We measured the ATP hydrolysis rate of recombinant mouse MDA5 in the presence of RNA transcripts (Figure 7 and Figure S8) and observed a negative correlation between the maximum number of sequencing reads in the MDA5 library and the ATP hydrolysis rate (cor = −0.414, Figure 7c). Analysis of the in vitro data revealed that AU-rich sequences lead to a decrease in ATP hydrolysis activity of MDA5 (cor = −0.445). Furthermore, the ATP hydrolysis rate negatively correlates with the immunostimulatory potential of RNA transcripts (cor = −0.426) (Figure 7d). This result suggests that the ATPase hydrolysis activity of MDA5 is not correlated to the binding and the immunostimulatory potential of the RNA transcripts and could therefore provide a model of RNA recognition by the protein. The data are consistent with previous work on MDA5 filament formation upon dsRNA binding [26], [27]. In structural and biophysical studies, Berke et al showed that ATP hydrolysis by MDA5 causes filaments to disassemble, perhaps by inducing translocation along the RNA or triggering a conformational change in the protein. According to our data, this may explain the observed inverse correlation between the immunostimulatory activity of IVTs and their potential to induce the ATPase activity of MDA5.
Until now, in vivo RLR ligands were poorly understood and a naturally occurring MDA5 ligand could only be purified indirectly by immunoprecipitation of LGP2:RNA complexes from virus-infected cells overexpressing LGP2 [54]. By applying a combination of RNA-protein crosslinking, immunoprecipitation of endogenous proteins and RNA deep sequencing analysis, we were able to investigate RLR-associated RNA from MeV infected cells. We compared our results to the empty GFP antibody control resembling a previously published immunoprecipitation strategy [20].
Our approach reveals that MDA5 preferentially binds measles virus RNA of positive polarity, whereas RIG-I additionally binds to (−) sense RNA within the trailer region as well as in the adjacent L gene. We propose that enriched RNA of positive polarity most likely represents mRNA species, since antigenomic RNA is only generated during replication and is immediately packed into nucleocapsids [55]–[57]. For Mononegavirales, these RNA-protein complexes are considered inaccessible for cytoplasmic proteins [55], [58] and might not be ligands for RLR proteins unless they become released. We show that, unlike MDA5, RIG-I binds (+) sense RNA originating from not only the L genomic segment, but also from the 3′ end of the MeV genome, which could be either le-N read-through transcripts or abortive replication products comprising 5′-triphosphate ends [45], [46]. Furthermore, we hypothesize that RIG-I specific enriched RNA of negative polarity represents abortive replication products also having 5′-triphosphate ends [20]–[23]. Additionally, 5′-copyback DI sequences combining vRNA of positive and negative polarity were found both in RIG-I and MDA5 immunoprecipitates and may contribute to recognition [49].
Bioinformatics analysis and in vitro transcription experiments revealed a correlation between AU content and read coverage of the obtained sequences or IVTs, respectively. As shown before [59], this indicates that RNA rich in AU could serve as a putative ligand for RIG-I and MDA5, or in a secondary manner lead to a specific structure that is recognized by both proteins. The slightly weaker correlation of RIG-I associated sequences with their AU content compared to MDA5 bound RNAs could be explained by additional sequences or triphosphate RNAs recognized by RIG-I that originate from regions less rich in AU.
Interestingly, ATP hydrolysis assays performed with recombinant MDA5 and RNA transcripts indicate that the AU content of RNA negatively correlates with the ATP hydrolysis rate of the protein. This inverse correlation between the immunostimulatory potential of RNAs and their capability to stimulate ATP hydrolysis by MDA5 lets us speculate that the ATPase activity might not be necessary for, or even interfere with, the immunoactivity of RNA ligands. Although this observation disagrees with recent findings about the role of ATP hydrolysis in RIG-I oligomerization on 5′-triphosphate dsRNA [60], we assume that MDA5 and RIG-I differ markedly in their mechanical activation and the role of ATP hydrolysis. Our data is supported by results suggesting that MDA5 filament formation is abrogated in an ATP-sensitive manner. By electron microscopy (EM) analysis it was shown that MDA5 filaments disassemble in the presence of ATP, indicating that ATP hydrolysis triggers the translocation of the protein along the dsRNA molecule or reduces the binding affinity, thereby interfering with downstream signaling [26], [27]. In light of the available data in the literature we therefore hypothesize that the ATPase activity of the MDA5 helicase domain contributes to substrate specificity by detaching the protein from low affinity substrates. To further test this hypothesis we generated RIG-IE373Q and MDA5E444Q, which are mutated in the “Walker B” ATP hydrolysis motif [61], slowing down or abrogating the ATP hydrolysis activity of the proteins, while preserving formation of ATP complexes. Overexpression of these mutant proteins from transfected plasmids showed a dramatic increase in their immunostimulatory potential in the absence of any viral ligands in comparison to expressed wild-type MDA5 (Figure S9). Furthermore, pull-down studies with the RIG-I Walker B mutant revealed an increase in the amount of recovered RNA while their immunostimulatory potential decreased (data not shown). The increased immunostimulation of ATPase deficient RLRs is consistent with the model that RNAs that lead to a reduced ATP-hydrolysis rate are more proficient in immunostimulation, possibly by stabilizing RLR∶RNA complexes. The negative correlation between AU-rich sequences and the ATP hydrolysis rate suggests that MDA5 binds AU-rich RNA in preference to GC-rich RNA. This would lead to a stronger interaction between RNA and MDA5 and result in a higher immunostimulatory signal. In order to test this hypothesis, we performed binding assays with MDA5 and IVTs but we were not able to demonstrate differences in the binding affinities between the different transcripts that might support this theory (data not shown). Finally, we speculate that RNA ligands for RLR proteins could be divided into two classes. The first class would comprise RNA molecules originating from the 5′-triphosphate ends of the genome or antigenome. These molecules could be generated in the course of read-through transcription and abortive replication [45], [46] and could therefore represent preferred ligands of RIG-I, as shown previously [20]. The second class of RNA molecules could be recognized by both receptor proteins. Our data suggest that recognition of these RNAs might occur through the AU composition of sequences [34]. This second class might also prominently include defective interfering (DI) particles generated during MeV replication. For MDA5, however, our deep sequencing data show that the (−) strand portion of the DIs is either relatively short or the fraction of DIs binding to MDA5 is magnitudes lower than the binding to L derived (+) sense RNAs and therefore not easily detectable during sequencing. A more detailed analysis of the deep sequencing data is currently ongoing in order to shed more light on the complex nature of the DIs involved.
It will be interesting to see what types of RNA associate with RIG-I and MDA5 during infections with different viruses and to what extent the AU composition and DI generation contributes to RNA recognition in these types of viruses. In particular, the finding that both RIG-I and MDA5 localize to AU rich regions suggests partially overlapping roles in detection of different viruses. The specificity of RIG-I and MDA5 for certain viruses may lie not only in the detection of 5′-triphosphate by RIG-I, but also in the heterogeneity of viral evasion strategies [62]. Our findings support a model for the recognition of AU-rich sequences by RIG-I and MDA5 from MeV-infected cells. Consistently, we find a similar correlation for in vitro transcribed RNA from the Mengo virus genome.
In general, the data support previous experiments indicating that MeV is mainly recognized by RIG-I, while MDA5 seems to play a minor role [4], [5], [13], [63]. It could be possible that RIG-I initially recognizes le-N read-through transcripts or abortive replication products containing 5′-triphosphate ends, leading to the activation of the signaling cascade. In a second round of recognition, RIG-I and MDA5 then recognize viral transcripts that are rich in AU. To further test this hypothesis, time dependent experiments need to be carried out.
One feature of the applied crosslinking technique is the introduction of specific T to C transitions at the interaction sites of 4SU-labeled RNA and the protein upon UV light exposure [37]. By identifying these point mutations in the deep sequencing data, one can exactly pinpoint the RNA sequences that interact with the protein of interest. However, our bioinformatics analysis did not reveal significant enrichment of T to C mutations, which could be explained by the rather low incorporation efficiency of the photoreactive nucleoside into viral RNA, consistent with the low incorporation level of 4SU into host RNA. Nevertheless, by increasing the incorporation efficiency in future studies, the identification of point mutants could further narrow down the precise binding sites of RLRs.
In summary, our approach provides a first insight into the molecular basis of vRNA derived from MeV interaction with MDA5 in living cells and reveals a preference for binding of AU-rich regions originated from (+)-sense RNA of the L gene. In vitro, these RNA molecules appear to be a poorer stimulator of the ATPase activity of MDA5, and result in more stable MDA5 filaments and support better downstream signaling.
Infection experiments were carried out in A459 human lung carcinoma cells. HEK 293T ISRE-FF reporter cells (stable expression of firefly luciferase under the control of an interferon stimulated response element) were used for interferon stimulation luciferase reporter gene assays. All cells were maintained in Dulbecco's Modified Eagle Medium supplemented with 2 mM L-glutamine, 1% Penicillin-Streptomycin and 10% FBS (all purchased from Invitrogen). Viruses used for infections were recombinant measles virus with a sequence identical to the vaccine strain Schwarz (AF266291.1.), Sendai virus, Sendai virus defective interfering particles H4 (kindly provided by Dominique Garcin, Geneva, Switzerland), Mengo virus strain pMC0 (kindly provided by Anne Krug, TU Munich, Germany) and EMCV. Primary antibodies to human MDA5 (AT113) and RIG-I (Alme-1) were purchased from Enzo Life Science (Loerrach, Germany). Antibody to GFP (ab1218) was obtained from Abcam (Cambridge, UK). Secondary antibodies were supplied by GE Healthcare (Buckinghamshire, UK).
A549 cells were infected with virus with an MOI of 1.0 in the presence of 400 µM 4SU. Infection was allowed to proceed for 24 h and living cells were washed with PBS (10 mM phosphate, 137 mM NaCl, 2.7 mM KCl, pH 7.5) and exposed to 1 J/cm2 365 nm UV light using a photocrosslinker (Vilbert Lourmat). Cells were harvested and incubated in Nonidet P-40 lysis buffer (50 mM HEPES, 150 mM KCl, 1 mM NaF, 10 µM ZnCl2, 0.5% NP-40, 0.5 mM DTT, protease inhibitor, pH 7.5) for 10 min on ice. The lysate was cleared by centrifugation and endogenous proteins were immunoprecipitated for 4 h with the respective antibodies (1 µg/mL) bound to protein G Dynabeads (Life Technologies). The beads were washed five times with high-salt wash buffer (50 mM HEPES, 500 mM KCl, 0.05% NP-40, 0.5 mM DTT, protease inhibitor, pH 7.5) and incubated with proteinase K (Thermo Scientific) for 30 min at 55°C. The RNA was isolated by phenol/chloroform/isoamylalcohol extraction and subjected to further analysis.
A549 cells were infected with MeV with an MOI of 1. Cells were harvested 24 hpi. Total RNA was isolated according to manufacturer's protocol of the RNeasy Protect Mini Kit (Qiagen) and subjected to Illumina deep sequencing.
Immunoactivity experiments were carried out in 24-well plates. 2.5×105 HEK 293T ISRE-FF reporter cells were transfected with 250 ng of recovered RNA, 500 ng in vitro transcripts or 500 ng plasmid DNA using Lipofectamine 2000 (Invitrogen) according to manufacturer's protocol. After 24 h incubation, cells were subjected to immunoactivity experiments using the Dual-Glo luciferase assay system (Promega) according to manufacturer's instructions. The luciferase activity was determined in a 96-well plate reader. Significance of differences in luciferase activity between samples were determined via an unpaired t-test.
Isolated RNA was prepared for Illumina sequencing using the mRNA-Seq library preparation kit (Epicentre) according to manufacturer's protocol. To remove ribosomal RNA species from the sequencing libraries a Ribo-Zero rRNA removal kit (Epicentre) was used. Quality of RNA-Seq libraries was validated on a DNA1500 chip for the Bioanalyzer 2100 (Agilent). Sequencing was performed on the Illumina Genome Analyzer in the Gene Center sequencing facility (LAFUGA). Obtained sequences were processed with the FASTX toolkit (http://hannonlab.cshl.edu/fastx_toolkit/) in order to remove adapter sequences and reads with PHRED scores below 30. Remaining sequences were mapped to human and viral genomes by utilization of the Bowtie algorithm [64], allowing maximal one mismatch per unique read. The Bowtie sequence alignments were converted with SAMtools [65] to pileup format, which was subsequently used for further data analysis. Relative sequence abundances were analyzed between RLR pull-down samples and the GFP control. Specific read enrichments were calculated by determining the relative sequence abundance at each position on the genomic segment and calculating the average of the RLR/GFP ratios over a dynamic window of 200 reads. Relative sequence abundances with log2 ratios above +1 were defined as significantly enriched in the RLR library.
RNA secondary structure prediction from measles virus genome or in vitro transcripts was performed by utilization of RNAfold from the ViennaRNA package [53] using standard parameter settings. For this purpose, the genome was divided into 201 nt fragments with a shifting window size of 5 nt. The sequences were folded in silico and the linear relationship between different data sets was quantified with the Pearson correlation coefficient.
DNase treatment of the immunoprecipitated RNAs and qPCR was performed as previously described [66]. The primer pairs used for quantification were identical to those published [67]. For cDNA synthesis a random hexanucleotide mix was used (Roche). Full length MeV vac2 cDNA with a known concentration was used for standard generation. Copy number values obtained for MDA5 and RIG-I were normalized to the control GFP.
Specific primers for reverse transcription (Roche transcriptor transcriptase) and the subsequent PCR (Biozym Phusion Polymerase) were adapted from Calain et al [47]. PCR products were analyzed on agarose gels and stained with ethidium bromide.
Templates were generated for in vitro transcription in a PCR adding the T7 promoter sequence (TAATACGACTCACTATA GGG) to the 5′ end of the desired MeV or Mengo virus genomic fragment, respectively (for oligonucleotides see Tables S2 und S4 respectively). PCR products were subsequently purified on agarose gels. RNA was transcribed using the Ambion Megashortscript T7 Kit according to the manufacturer's protocols. The reaction was incubated overnight at 37°C and RNA was precipitated using LiCl at −20°C for 30 minutes. Afterwards, RNA was subjected to triphosphate digestion using FastAP (Fermentas) according to the manufacturer's instructions and purified on denaturing 8 M urea/10% polyacrylamide gels at 25 mA constant current. Gel slices containing RNA were incubated overnight with 450 µL probe elution buffer (0.5 M ammonium acetate, 1 mM EDTA, 0.2% SDS). Eluted RNA was isolated by phenol/chloroform/isoamylalcohol extraction and precipitated with ethanol.
ATPase hydrolysis activity was determined using [γ-P32] ATP. Mouse MDA5 was purified as described previously [32] and 1.6 µM of protein was preincubated with 80 nM in vitro transcribed RNA for 10 min at room temperature. The reaction was initiated by addition of ATPase hydrolysis buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 1.5 mM MgCl2, and 2 mM DTT) containing 2 mM ATP and 0.2 µCi [γ-P32] ATP. The hydrolysis rate was monitored over 1 h and analyzed by thin layer chromatography (TLC).
Sequences encoding full-length human RIG-I with N-terminal FLAG-tag and full-length human MDA5 with N-terminal FLAG-tag were cloned into pcDNA5 FRT/TO (Invitrogen). Mutants (FLAG-RIG-I E373Q and FLAG-MDA5 E444Q) were generated by site directed mutagenesis with PfuUltra (Agilent).
|
10.1371/journal.pgen.1004390 | Glycogen Synthase Kinase (GSK) 3β Phosphorylates and Protects Nuclear Myosin 1c from Proteasome-Mediated Degradation to Activate rDNA Transcription in Early G1 Cells | Nuclear myosin 1c (NM1) mediates RNA polymerase I (pol I) transcription activation and cell cycle progression by facilitating PCAF-mediated H3K9 acetylation, but the molecular mechanism by which NM1 is regulated remains unclear. Here, we report that at early G1 the glycogen synthase kinase (GSK) 3β phosphorylates and stabilizes NM1, allowing for NM1 association with the chromatin. Genomic analysis by ChIP-Seq showed that this mechanism occurs on the rDNA as active GSK3β selectively occupies the gene. ChIP assays and transmission electron microscopy in GSK3β−/− mouse embryonic fibroblasts indicated that at G1 rRNA synthesis is suppressed due to decreased H3K9 acetylation leading to a chromatin state incompatible with transcription. We found that GSK3β directly phosphorylates the endogenous NM1 on a single serine residue (Ser-1020) located within the NM1 C-terminus. In G1 this phosphorylation event stabilizes NM1 and prevents NM1 polyubiquitination by the E3 ligase UBR5 and proteasome-mediated degradation. We conclude that GSK3β-mediated phosphorylation of NM1 is required for pol I transcription activation.
| Nuclear actin and myosin are essential regulators of gene expression. At the exit of mitosis, nuclear myosin 1c (NM1) mediates RNA polymerase I (pol I) transcription activation and cell cycle progression by modulating assembly of the chromatin remodeling complex WICH with the subunits WSTF and SNF2h and, crucially, facilitating H3K9 acetylation by the histone acetyl transferase PCAF. The molecular mechanism by which NM1 is regulated remains however unknown. Here, we conducted a genome-wide screen and demonstrate that GSK3β is selectively coupled to the rDNA transcription unit. In embryonic fibroblasts lacking GSK3β there is a significant drop in rRNA synthesis levels and the rDNA is devoid of actin, NM1 and SNF2h. Concomitantly with a transcriptional block we reveal decreased levels of histone H3 acetylation by the histone acetyl transferase PCAF. At G1, transcriptional repression in the GSK3β knockout mouse embryonic fibroblasts, leads to NM1 ubiquitination by the E3 ligase UBR5 and proteasome-mediated degradation. We conclude that GSK3β suppresses NM1 degradation through the ubiquitin-proteasome system, facilitates NM1 association with the rDNA chromatin and transcription activation at G1. We therefore propose a novel and fundamental role for GSK3β as essential regulator of rRNA synthesis and cell cycle progression.
| rRNA genes are transcribed by RNA polymerase I (pol I) into a large precursor (pre)-rRNA which is cleaved into 18S, 5.8S and 28S rRNAs for incorporation into ribosomal subunits [1], [2]. Pol I, in complex with the transcription initiation factor TIF1A, is first recruited to the gene promoter via the upstream binding factor (UBF) and the selectivity factor 1 (SL1) [3]. After promoter assembly, pol I transcription requires the synergy between actin and nuclear myosin 1c (NM1) [4], [5]. The interaction between pol I-associated actin with the chromatin-bound NM1 is required for transcription activation [6]–[9]. NM1 interacts with the chromatin through its C-terminal tail and it is also part of the multiprotein assembly B-WICH that contains the WICH chromatin remodeling complex with the subunits WSTF and the ATPase SNF2h but does not comprise actin [9]–[12]. While WSTF bookmarks the position of the chromatin remodeling complex on the rDNA transcription unit, NM1 interacts with SNF2h, stabilizes the WICH complex but, crucially, facilitates recruitment of the histone acetyl transferase (HAT) PCAF [9]. An important structural role has therefore been ascribed to NM1 that connects pol I with the chromatin through direct interactions with chromatin and the pol I-associated actin, respectively. This mechanism depends on the myosin ATPase activity. Further, this mechanism activates transcription by providing the permissive chromatin that in turn facilitates polymerase function across the active gene through modulating WICH assembly and PCAF recruitment [9]. At the exit of mitosis, this mechanism is critical for cell cycle progression when pol I transcription must be re-activated [9]. However, how NM1 is regulated at the onset of pol I transcription activation is not known.
GSK3β is a proline-directed serine/threonine kinase regulated by phosphorylation. The unphosphorylated form of GSK3β is enzymatically active [13], [14]. GSK3β is inactivated through activation of several signaling pathways including Wnt signaling that either leads to serine phosphorylation [15]–[17], or disrupts multiprotein complexes that contain GSK3β and its substrates [18]. GSK3β regulates cellular metabolism, the cytoskeleton and gene expression [16]. GSK3β also mediates cell cycle progression by phosphorylating pro-proliferative factors for degradation or by phosphorylating and stabilizing anti-proliferative factors. c-Myc is an example of short-lived proteins that is ubiquitinated in a GSK3β -dependent manner by the F-box protein Fbw7 and subsequently degraded by the proteasome [19]. GSK3β also controls expression of cyclin D1, which is phosphorylated to promote nuclear export and subsequent degradation [20]. In contrast, GSK3β-mediated phosphorylation of a single serine residue (Ser-118) in the estrogen receptor α leads to the stabilization of the receptor and protects it from proteasome-mediated degradation [21]. This dual mode of activity ensures that cell cycle progression, growth and proliferation are kept under tight regulation.
Previous work has shown a link between GSK3β and pol I-specific transcription factors [22]. Induction of granulocytic differentiation in murine myeloid cells results in the degradation of UBF via GSK3β and the ubiquitin/proteasome system [23]. Furthermore, enzymatically active GSK3β interacts with the member of the SL1 complex TAFI110 and suppresses pol I transcription in a H-RAS dependent manner [24]. GSK3β has therefore been suggested to suppress pol I transcription by repressing assembly of transcription-competent polymerase at rRNA gene promoter in transformed cells.
Here, we studied whether GSK3β has a more fundamental role in pol I transcription in non-transformed cells. A genome-wide screen showed that GSK3β is selectively distributed across the entire rDNA transcription unit. Further, GSK3β is required for rDNA association of numerous factors required for pol I transcription. In GSK3β−/− mouse embryonic fibroblasts (MEFs) we found decreased levels of occupancy of actin, NM1 and SNF2h, at both promoter and transcribed sequences. These mechanisms, along with ultrastructural analysis of nucleoli in the GSK3β−/− MEFs, correlate with decreased pol I transcription through loss of permissive chromatin. Further, in G1-arrested GSK3β−/− MEFs, NM1 becomes specifically ubiquitinated by the E3 ligase UBR5 and degraded by the proteasome. These observations collectively suggest that GSK3β suppresses NM1 degradation through the ubiquitin-proteasome system, facilitates NM1 association with the rDNA chromatin and promotes pol I transcription activation at G1. We therefore propose a novel and fundamental role for GSK3β as a key regulator of rRNA synthesis.
We confirmed the localization of GSK3β on the rDNA with a novel antibody termed CGR11 against the first nine N-terminal amino acids of the protein (Figure 1A). Since the epitope contains Serine 9, which is kept unphosphorylated in the active form of GSK3β [25], the CGR11 antibody is designed to preferentially target active GSK3β. The CGR11 antibody specifically detected a single protein of 48 kDa on immunoblots of nuclear extracts from HeLa cells and wild type mouse embryonic fibroblasts (GSK3β+/+ MEFs), similarly to the commercial pan- GSK3β antibody 27C10 (Figure 1B), but not in the GSK3β−/− MEFs.
We used the GSK3β antibodies CGR11 and 27C10 to study GSK3β occupancy along the rDNA transcription unit by chromatin immunoprecipitation (ChIP). We prepared crosslinked chromatin from both GSK3β+/+ MEFs and GSK3β−/− MEFs and subjected the chromatin to immunoprecipitations with the CGR11 and 27C10 antibodies. The precipitated DNA was analyzed by quantitative real-time PCR (qPCR) using primers amplifying fragments of 45S (promoter), 18S, 5.8S, 28S rDNA and the IGSs (intergenic sequences). The qPCR analysis shows that both antibodies precipitated rDNA from the chromatin isolated from GSK3β+/+ MEFs (Figure 1C). In contrast we did not get any signal when chromatin isolated from GSK3β−/− MEFs was used in the immunoprecipitations (Figure 1C). We conclude that GSK3β specifically associates with the rDNA.
For further assessment of GSK3β protein occupancy throughout the rDNA we performed a genome-wide screen by ChIP followed by next-generation sequencing (ChIP-Seq). We subjected crosslinked chromatin isolated from GSK3β+/+ MEFs to immunoprecipitations with the CGR11 antibody and the DNA fragments were sequenced directly. Compared to background genomic binding levels the rDNA repeat showed GSK3β binding that was approximately two orders of magnitude higher. High levels of binding were found within the rRNA gene sequence, its upstream promoter elements and in the IGS extending 12 kb downstream of the rDNA (Figure 1D; Figure S1). In contrast large segments of the IGS appeared devoid of GSK3β binding (Figure 1D). The pattern of GSK3β binding across the rDNA transcription unit is similar to the binding patterns of pol I and UBF to the rDNA repeats [26]–[28]. Interestingly, very low levels of GSK3β binding to other genomic loci were detected (Table S1). Although the presence of GSK3β in the IGSs is not yet understood, the distribution of GSK3β all along the rDNA transcription unit, including externally transcribed sequences (ETSs), 18S, 5.8S, 28S and internally transcribed sequences (ITSs), suggests that GSK3β has a primary role in pol I transcription regulation.
To evaluate the possible involvement of GSK3β in rDNA transcription, we isolated total RNA from wild type and GSK3β−/− MEFs and measured relative pre-rRNA levels by quantitative reverse transcription real time PCR (qRT-PCR). Using primers amplifying 45S pre-rRNA, in the GSK3β−/− MEFs we detected a fivefold drop in the amount of nascent transcript relative to β-actin mRNA levels (Figure 2A). Twofold decrease in the levels of 45S pre-rRNA levels were also detected in HeLa cells subjected to GSK3β gene silencing by RNAi (Figure S2). To confirm these results in living cells, GSK3β+/+ MEFs and GSK3β−/− MEFs were treated with DRB to selectively block RNA polymerase II transcription and then subjected to in situ run on assays. In these experiments incorporation of the cell permeable fluorine-conjugated UTP analogue FUrD was allowed for 10 minutes and the FUrD incorporated in nascent rRNA transcripts was monitored by immunofluorescence and confocal microscopy [29]. Consistent with the qRT-PCR analysis these experiments showed that in the absence of GSK3β, FUrD incorporation in nascent nucleolar transcripts was down-regulated (Figure 2B).
We next analyzed how GSK3β affects pol I transcription. We started by applying methylated DNA immunoprecipitation (MeDIP), for unbiased detection of methylated DNA [30], [31]. Genomic DNA obtained from GSK3β+/+ MEFs and GSK3β−/− MEFs was randomly sheared by sonication and immunoprecipitated with a monoclonal antibody that recognizes 5-methylcytidine. qPCR analysis on the immunoprecipitated DNA using primers amplifying rRNA gene promoter show that the methylation levels did not change in the absence of GSK3β when compared to the reference TSH2B gene, a region of the histone H2B gene which is known to be methylated (Figure 2C). To determine whether the absence of GSK3β affected rDNA occupancy of the pol I machinery, we performed ChIP on crosslinked chromatin from GSK3β+/+ MEFs and GSK3β−/− MEFs with a human autoimmune serum against active pol I (S57299), an antibody to UBF, and antibodies to actin, WSTF, SNF2h, NM1, GSK3β (CGR11) and non-specific rabbit IgGs. In GSK3β−/− MEFs we detected modest increments in the amounts of promoter and 18S co-precipitated with the pol I and UBF antibodies (Figure 2D). In contrast, we detected drops in the amounts of promoter, 18S and IGSs co-precipitated with antibodies to actin, NM1 and SNF2h (Figure 2E). The WSTF antibody precipitated promoter, 18S and IGSs with similar efficiencies from both GSK3β+/+ MEFs and GSK3β−/− MEFs chromatin (Figure 2E). These results indicate that as consequence of GSK3β knockout the rDNA occupancies of pol I machinery, actin and certain components of the B-WICH complex are altered. These changes were accompanied by decreased PCAF occupancy and H3K9 acetylation levels (Figure 2F). However, the levels of H3K4me3 were not altered. We conclude that in the GSK3β−/− MEFs reduced levels of rRNA synthesis are primarily due to a chromatin state which is not compatible with transcription. Morphological analyses of nucleoli in GSK3β−/− MEFs were consistent with this hypothesis. GSK3β−/− and GSK3β+/+ MEFs were synchronized in G1 to avoid differences in cell cycle progression, double-stained with antibodies against nucleolin and UBF, and analyzed by fluorescence microscopy. The overall nuclear size and shape were similar in the two cell lines, but the GSK3β−/− MEFs showed a significant increase in the number of nucleoli per cell, and the nucleoli were smaller (Figures 3A–B). The GSK3β−/− nucleoli were UBF-positive, in accordance with our ChIP results (Figure 2D).
We also analyzed the ultrastructure of the nucleoli in GSK3β−/− and GSK3β+/+ MEFs by transmission electron microscopy. In normal GSK3β+/+ MEFs, the nucleoli were well developed and showed a normal morphology with three typical components: dense fibrillar centers (DFCs), fibrillar component (FC) and granular component (GC). The GSK3β−/− cells were instead characterized by reduced nucleoli that often lacked a well-defined compartmentalization and displayed reduced amounts of GC (Figure 3C). We could occasionally observe GSK3β−/− MEFs with large nucleoli, but these nucleoli exhibited a highly vacuolated ultrastructure that was never observed in GSK3β+/+ nucleoli (Figure S3). Interestingly, the nucleolar alterations that we observed in GSK3β−/− MEFs do not resemble the nucleolar disruption phenomenon that has been described in response to a variety of stress conditions [32]. The morphology of the GSK3β−/− nucleoli and the fact that they contain UBF, support instead the idea that a larger number of nucleolar organizer regions (NORs) becomes activated in GSK3β−/− MEFs than in GSK3β+/+ MEFs. However, these numerous GSK3β−/− NORs do not engage in efficient rRNA production and fail to assemble fully structured nucleoli. The morphological analysis of GSK3β−/− and GSK3β+/+ MEFs also revealed differences in the patterns of chromatin condensation between the two cell types. Staining of GSK3β−/− MEFs with DAPI revealed the existence of small patches of dense chromatin scattered throughout the nucleoplasm, often in association with the nuclear periphery, whereas GSK3β+/+ MEFs were characterized by fewer and larger areas of densely packaged chromatin (Figure 3D). This difference was confirmed at the ultrastructural level (Figure 3E).
In summary, the morphological analyses reveal severe defects in nucleolar function and chromatin organization and together with the molecular analyses presented above support the conclusion that GSK3β contributes to pol I transcription activation by indirectly inducing a permissive chromatin state.
Actin, NM1, SNF2h and PCAF levels along the rDNA transcription unit are dependent on GSK3β (Figure 2E–F). To test whether GSK3β interacts with actin, NM1, SNF2h, WSTF or PCAF, we applied immunoprecipitations to nuclear lysates from GSK3β+/+ MEFs. Briefly, we incubated nuclear lysates with the anti- GSK3β antibody CGR11. Analysis of the immunoprecipitated fractions by immunoblotting showed that endogenous GSK3β co-precipitated NM1 as well as actin whereas SNF2h, PCAF, and WSTF were not co-immunoprecipitated (Figure 4A). These results show that in the nucleus GSK3β is part of the same complex with NM1 and actin.
To evaluate whether GSK3β targets any specific regions of NM1, we used HEK293T cell lines stably expressing V5-tagged wild-type NM1 (V5-wtNM1), a V5-tagged NM1 mutant with impaired actin binding function (V5-RK605AA NM1) as well as V5-tagged deletion constructs that lack IQ motifs (V5-ΔIQ NM1) or the tail domain (V5-ΔC NM1) [8], [9] (Figure 4B). We subjected total lysates from each of the above cell lines to immunoprecipitations with an anti-V5 antibody to pull down the NM1 constructs. Analysis of the co-immunoprecipitated proteins on immunoblots with the CGR11 antibody showed specific co-precipitations of endogenous GSK3β with V5-wtNM1 but not with the V5-RK605AA NM1 (Figure 4C). Further, GSK3β co-precipitated with V5-ΔIQ NM1 and with lower efficiency, also with V5-ΔC NM1 (Figure 4C). Interestingly, a recent study supports the possibility of a direct interaction with NM1 as three putative GSK consensus sites were identified in the C-terminus of the human NM1 [33]. Our present results suggest that NM1 and GSK3β are part of the same complex and further indicate that the association is negatively affected when NM1 cannot interact with actin.
To start evaluating whether GSK3β phosphorylates NM1, we resolved extracts from growing GSK3β+/+ MEFs or GSK3β−/− MEFs by phosphate affinity SDS PAGE [34]. In these assays, extensive protein phosphorylation is revealed by comparing changes in the protein electrophoretic mobility on immunoblots of lysates treated or untreated with alkaline phosphatase. The results of immunoblots of growing GSK3β+/+ MEFs lysates for NM1 showed specific gel retardations which were lost upon phosphatase treatment (Figure 4D, lanes 1, 2). Similar results were obtained upon analysis of GSK3β−/− MEFs lysates (Figure 4D, lanes 5, 6), altogether suggesting that in growing cells NM1 is extensively phosphorylated but not by GSK3β. We next applied phosphate affinity SDS PAGE to resolve lysates from GSK3β+/+ MEFs or GSK3β−/− MEFs blocked in G1 by serum starvation (see also Figure S4). Even though we did not reveal the same extent of gel retardation as in growing cells, upon alkaline phosphatase treatment on wild type lysates we observed considerable reduction in the amount of NM1 (Figure 4D, lanes 3,4). Remarkably, analysis of immunoblots of lysates from the GSK3β−/− MEFs blocked in G1 showed decreased NM1 levels independent of the alkaline phosphatase treatment (Figure 4D, lanes 7, 8). In support of this observation, analysis on immunoblots of lysates from GSK3β−/− MEFs blocked in G1 by contact inhibition [35] independently revealed a similar drop in the NM1 protein levels (see also Figure S5A–B). Immunoblots performed on the phosphate affinity SDS PAGE for actin revealed that endogenous actin is phosphorylated, but phosphorylation appears to be independent of GSK3β since it was not affected in either growing or G1-blocked GSK3β−/− MEFs (Figure 4D). Furthermore, in contrast to NM1, the expression levels of endogenous actin were not affected in growing or G1-arrested GSK3β−/− MEFs. We therefore conclude that at G1 NM1 is specifically stabilized by GSK3β and this regulation possibly occurs through a direct interaction.
We next investigated whether GSK3β directly phosphorylates NM1. To start addressing this point, we prepared G1 lysates from GSK3β−/− MEFs, untreated or treated with the proteasome inhibitor MG132, under which conditions the NM1 levels are rescued (see also Figure 5D). Lysates were supplemented with γ-33P-ATP and with purified recombinant GSK3β. Following incubation, endogenous NM1 was immunoprecipitated, resolved by SDS PAGE and visualized by phosphorimaging (Figure 4E). The results show that a fraction of the immunoprecipitated NM1 was phosphorylated by recombinant GSK3β in the presence of MG132 (Figure 4E). To confirm that endogenous NM1 is a substrate for GSK3β, we subjected lysates from G1-arrested GSK3β −/− MEFs treated with MG132 to immunoprecipitations with anti-NM1 or anti-actin antibodies. After the immunoprecipitations, the beads were washed and the bound NM1 or actin were incubated with γ-33P-ATP and with purified recombinant GSK3β. Following incubation, the endogenous NM1 and actin were eluted from the beads and resolved by SDS PAGE. Analysis by phosphorimaging shows that a radioactive band was detected only for NM1 but not for actin (Figure 4F), indicating that NM1 is a direct substrate for GSK3β. Consistent with previous observations [36], in the same assay, a degree of GSK3β autophosphorylation was also detected. To further endorse the specificity of these results and identify potential phosphorylation sites in the NM1 primary sequence, we immunoprecipitated endogenous NM1 from nuclear extracts of MG132-treated GSK3β+/+ MEFs and GSK3β−/− MEFs arrested in G1. The immunoprecipitated protein fraction was resolved by SDS-PAGE and subjected to in gel digestion. The resulting peptides were extracted and analyzed by tandem mass spectrometry. Analysis of the immunoprecipitated NM1 from the G1 GSK3β+/+ MEFs nuclear lysate identified the peptide DGIIDFTSGSELLITK in both its phosphorylated (Figure 4G) and non-phosphorylated state (Figure S6) with mascot ion scores of 61 and 94, respectively. The MS/MS data indicates that within the above peptide the phosphorylation is present at Serine 8 which in the mouse full length NM1 amino acid sequence corresponds to the Serine 1020 located in the NM1 C-terminal tail (Accession number, Q9WTI7-3). In contrast, the same analysis performed on the endogenous NM1 immunoprecipitated from the G1 GSK3β−/− MEFs nuclear lysate identified the non-phosphorylated peptide DGIIDFTSGSELLITK with a mascot ion score of 78 (Figure 4H) but did not reveal its phosphorylated form. These results show that Ser-1020 is directly phosphorylated by GSK3β in early G1.
We conclude that the endogenous NM1 is a bona fide phosphorylation substrate for GSK3β. Phosphorylation specifically targets the NM1 C-terminus at Ser-1020 and occurs in G1. We suggest that at the exit of mitosis GSK3β phosphorylation stabilizes NM1 from proteasome-mediated degradation.
Cell cycle profiling by flow cytometry and NM1 steady state expression analysis in GSK3β+/+ MEFs and GSK3β−/− MEFs after release from a G1 block confirmed a specific down regulation of NM1 (Figure 5A; Figure S7). In the absence of GSK3β NM1 down-regulation is at the protein level since qRT-PCR analysis of the relative NM1 mRNA levels in the GSK3β−/− MEFs were not altered in comparison to those in the GSK3β+/+ MEFs (Figure 5B). Furthermore, we isolated total RNA from wild type and GSK3β−/− MEFs blocked in early G1 and measured relative 45S pre-rRNA levels by qRT-PCR. Using both serum starvation and contact inhibition to arrest cells in G1, we detected significant drops in the amount of nascent transcript relative to β-tubulin mRNA levels (Figure 5C; Figure 5SC).
We next synchronized GSK3β−/− MEFs in G1 by serum starvation and treated with the proteasome inhibitor MG132. After release from the G1 block, total lysates were collected at 0 h, 4 h and 8 h and analyzed for NM1 protein levels on immunoblots. In lysates from untreated GSK3β−/− MEFs, NM1 expression was down at 0 h and 4 h after release from the G1 block and was rescued around 8 h after the release (Figure 5D). In contrast, immunoblots of lysates from MG132-treated GSK3β−/− MEFs showed a marginal increase in the NM1 protein levels between 0 h and 8 h after the G1 block release (Figure 5D), altogether suggesting selective degradation of NM1 by the proteasome at early G1. Since treatment with MG132 rescued the levels of NM1, we further evaluated whether by inhibiting the proteasome we could also rescue the ability of NM1 to bind to the rDNA chromatin. We therefore performed ChIP on crosslinked chromatin from GSK3β+/+ and GSK3β−/− MEFs blocked in G1, treated or untreated with MG132, using antibodies to NM1 and GSK3β (CGR11). The precipitated DNA was analyzed by qPCR with primers specific for the rDNA promoter. In the GSK3β−/− MEFs the NM1 antibodies precipitated the promoter 2-fold less efficiently (Figure 5E), possibly due to NM1 degradation by the proteasome. NM1 occupancy on the promoter was not restored even in the presence of MG132 (Figure 5E). On the contrary, in ChIP experiments performed on chromatin from GSK3β+/+ MEFs at G1 the anti-NM1 antibody co-precipitated the promoter, independently of the MG132 treatment (Figure 5E). These results show that at G1, rescuing the NM1 protein levels by proteasome inhibition does not restore the ability of NM1 to efficiently bind the chromatin in the absence of GSK3β. To find out whether GSK3β-mediated phosphorylation of NM1 modulates association with the rDNA, we treated GSK3β+/+ MEFs arrested at G1 with the cell-permeable GSK3β inhibitor 6-bromoindirubin-30-oxime (BIO) [37], [38]. On immunoblots of lysates of BIO-treated GSK3β+/+ MEFs the reactivity of CGR11 to GSK3β was considerably decreased whereas the pan-GSK3β antibody 27C10 and the antibodies to actin and NM1 were not affected by the BIO treatment (Figure 5F). Using the CGR11 antibody we performed ChIP on crosslinked chromatin from BIO-treated GSK3β+/+ MEFs to monitor occupancies of NM1 and active GSK3β at the rDNA promoter. The qPCR results show that upon inhibition of the GSK3β kinase activity with BIO, GSK3β does not co-precipitate rDNA and under the same conditions, NM1 shows a 50% drop in rDNA binding (Figure 5G).
We conclude that at G1 GSK3β phosphorylates NM1 to facilitate NM1 association with the rDNA chromatin while simultaneously protecting NM1 from degradation by the proteasome.
We next determined whether NM1 is a GSK3β-dependent substrate for ubiquitination. For this purpose cells were transiently transfected with a plasmid encompassing a HA-tagged version of the ubiquitin open reading frame. Following expression of the HA-ubiquitin, cells were arrested in G1 by serum starvation and treated with MG132. We prepared total lysates and subjected them to immunoprecipitations with the anti-NM1 antibody. The fractions of co-immunoprecipitated proteins were analyzed on immunoblots for HA-tagged ubiquitin. The results show that in contrast to GSK3β+/+ MEFs, in G1 lysates from GSK3β−/− MEFs treated with MG132, NM1 becomes polyubiquitinated (Figure 6A). NM1 polyubiquitination was not detected in lysates from growing GSK3β−/− MEFs even in the presence of MG132 (Figure S8A), whereas in lysates prepared from growing GSK3β+/+ MEFs or HEK293T cells NM1 appears to be polyubiquitinated (Figure S8B–C). To confirm these results in an independent cellular system, HeLa cells were incubated with a master mix containing the HA-tagged ubiquitin plasmid and siRNA oligonucleotides for specific GSK3β gene silencing (GSK3β RNAi) or control scrambled siRNA oligonucleotides (scrRNAi) (Figure 6B; Figure S2). HeLa cells were maintained at G1 by serum starvation and were treated with MG132. Lysates were subjected to immunoprecipitations with anti-NM1 antibodies and the co-immunoprecipitated protein fractions were analyzed on immunoblots for HA-tagged ubiquitin. The results show a marked increase in the levels of endogenous NM1 polyubiquitination in lysates of GSK3β-silenced HeLa cells treated with MG132 (Figure 6B, lanes 3 and 6). We conclude that NM1 polyubiquitination is dependent on GSK3β only at G1.
To identify possible E3 ligases involved in NM1 ubiquitination, we subjected nuclear lysates from untreated or MG132-treated GSK3β+/+ MEFs to immunoprecipitations with antibodies to NM1. The co-immunoprecipitated proteins were subjected to mass spectrometry analysis by nLC-MS/MS. Within the subset of co-immunoprecipitated proteins we found two candidate E3 ligases, UBR5 and F-box/WD repeat-containing protein 8 (Fbxw8) (Table S2). Remarkably, UBR5 and Fbxw8 were not co-precipitated with NM1 from MG132-treated GSK3β+/+ MEFs nuclear lysates (Table S3), suggesting a functional association with NM1. To find out whether UBR5 and Fbxw8 target NM1 for ubiquitination in a GSK3β-dependent manner at G1, we silenced the UBR5 and Fbxw8 genes in the GSK3β−/− MEFs expressing HA-tagged ubiquitin and maintained at G1. After treatment with MG132, lysates were subjected to immunoprecipitations with anti-NM1 antibodies and the co-immunoprecipitated protein fractions were analyzed on immunoblots for HA-tagged ubiquitin. The results show that silencing Fbxw8 did not affect the level of NM1 polyubiquitination compared to control scrRNAi (Figure 6C). On the contrary NM1 polyubiquitination was not observed upon UBR5 gene silencing (Figure 6C). We conclude that in the absence of GSK3β NM1 polyubiquitination is specifically mediated by UBR5 at G1.
In summary, NM1 phosphorylation by GSK3β blocks NM1 ubiquitination by UBR5 and degradation by the proteasome, leads to NM1 association with the chromatin and promotes rDNA transcription activation at G1.
By interacting with some components of the SL1 complex, in cells transformed with oncogenic H-RAS GSK3β functions as negative regulator of rDNA transcription suppressing assembly of transcription-competent pol I at the gene promoter [24]. Here we show for the first time that GSK3β also has a positive role on the basal mechanism that leads to activation of rRNA synthesis. Genomic analysis of GSK3β by ChIP-Seq showed that GSK3β selectively distributes across the entire rDNA transcription unit and to a certain degree, GSK3β also binds to intergenic sequences. This association with the gene is shown to be functional since in the GSK3β knockout MEFs we found a fivefold reduction in rRNA synthesis levels, which correlated with increased levels of pol I occupancy at the rRNA gene promoter. Furthermore, results from the ChIP analysis showed decreased occupancy for actin, NM1 and SNF2h at the promoter and across the gene. We have recently shown that NM1 binds the rDNA chromatin, to promote the activation of pol I transcription by stabilizing the B-WICH complex, such that it can subsequently recruit the HAT PCAF. These mechanisms contribute to the permissive chromatin required for pol I transcription activation [9]. These results are compatible with the pol I transcriptional drop observed in the GSK3β knockout MEFs where the rRNA gene promoter is almost quantitatively devoid of the HAT PCAF and displays a fourfold down-regulation in the levels of H3K9 acetylation. Local impairment of H3 acetylation is accompanied by moderate impairment of the chromatin remodeling function due to the absence of SNF2h at the gene promoter. The morphological analysis of GSK3β−/− MEFs suggests that in the absence of GSK3β activity the cell activates additional NORs, probably in an attempt to compensate for the reduced rRNA production. The GSK3β−/− nucleoli are not fully functional but do recruit nucleolin and UBF, which indicates that these nucleoli are functional to some extent, in agreement with our molecular analysis. The GSK3β−/− MEFs also display differences in the overall patterns of chromatin condensation. These chromatin changes could either be a direct consequence of impaired GSK3β function or an indirect response to the reduced ribosome biosynthetic capacity of the cell. In any case, the severe structural alterations observed in the nuclei of GSK3β−/− cells support the important role of GSK3β in rRNA biogenesis. We conclude that GSK3β contributes to transcription activation and maintenance by regulating local rDNA chromatin modifications.
Our results suggest that across the rDNA transcription unit GSK3β performs its regulatory function by targeting pol I-associated factors. GSK3β is a promiscuous enzyme [39] that phosphorylates serine or threonine at position 4 of the consensus sites S/TXXXS/T[PO3] [40], [41], but also single Ser residues outside the above mentioned consensus sequence [21]. About 20% of the mammalian proteome contains multiple putative GSK3 phosphorylation sites [33]. UBF has, for instance, five putative GSK consensus sites and it is directly phosphorylated by GSK3β in vitro and in vivo [23]. Both NM1 and actin were also found to be potential GSK3β substrates [33]. Accordingly, we found that GSK3β co-precipitated NM1 and actin from nuclear lysates, but not SNF2h, WSTF or PCAF. This suggests that GSK3β is part of the same complex with actin and NM1. We found however that only NM1 is directly phosphorylated by GSK3β and that NM1 phosphorylation by GSK3β occurs on a single Serine residue (Ser-1020), located within the murine NM1 C-terminal tail domain. Mass spectrometry analysis identified a single NM1 phosphopeptide in G1-arrested GSK3β+/+ MEFs but not in the GSK3β−/− MEFs. We cannot exclude the presence of other GSK3β phosphorylation sites; Ser-1020 phosphorylation however, seems to occur outside the canonical GSK3β consensus site. Considering that the state of NM1 phosphorylation is not dependent on GSK3β in growing cells, we conclude that the Ser-1020 phosphorylation by GSK3β exclusively occurs in G1.
GSK3β-dependent phosphorylation can lead to either stabilization of the substrates or further polyubiquitination and degradation by the proteasome [42], [43]. UBF phosphorylation by GSK3β promotes UBF degradation by the ubiquitin-proteasome system, concomitantly with differentiation of myeloid cells [23]. Our results therefore indicate that at early G1, GSK3β phosphorylates NM1 to prevent NM1 polyubiquitination and degradation by the proteasome. Consistently, in the absence of GSK3β, we discovered that at G1 NM1 is polyubiquitinated by UBR5 and rapidly degraded. UBR5, also termed EDD/hHyd, is an E3 ligase that targets the N-terminus of its substrates and interacts with GSK3β [44], [45]. UBR5 resides in the nucleolus and it is known to associate with SIRT7 [46]. Furthermore, UBR5 seems to have a huge impact on cell cycle progression. UBR5 regulates S-phase and G2/M DNA damage checkpoints, it induces cell cycle arrest by increasing p53 levels and has been recently implicated in cellular proliferation [47]–[49]. Since in G1-arrested GSK3β−/− MEFs we discovered an almost quantitative drop in the levels of nascent rRNA, we now hypothesize that GSK3β-mediated phosphorylation protects NM1 from UBR5-mediated polyubiquitination, and this mechanism is important for pol I transcription activation and progression through the cell cycle (Figure 7).
A key question is why NM1 phosphorylation by GSK3β is important for rRNA synthesis at a specific temporal window of the cell cycle. We recently demonstrated that NM1 binds the rDNA chromatin through its C-terminal tail, interacts with the pol I-associated actin and this interaction is dependent on the myosin ATPase cycle. Association with the chromatin stabilizes actin-binding and allows for the establishment of permissive chromatin required for pol I transcription activation and cell cycle progression [9]. In the GSK3β knockout cells arrested in G1, the rRNA gene promoter displays reduced NM1 levels even when the NM1 protein expression is rescued by MG132 treatment. Similarly, NM1 promoter occupancy levels drop after inhibition of GSK3β activity in living cells by treatment with BIO. The NM1 C-terminal tail is necessary for chromatin association [9]. It is therefore possible that phosphorylation by GSK3β is a primary requirement for NM1 to bind the rRNA gene promoter. Since NM1 association with the rDNA is a condition for actin occupancy [9], it is tempting to speculate that by enhancing association of NM1 with the chromatin, phosphorylation at Ser-1020 indirectly stabilizes the actomyosin complex on the rDNA; this may be achieved by tethering of GSK3β to NM1 through actin.
How the NM1 phosphorylation by GSK3β is restricted to the G1 phase of the cell cycle is not known and deserves further investigation. Our working model is however that at G1 GSK3β-mediated NM1 phosphorylation unleashes a domino effect that maintains the actin-NM1 complex and B-WICH assembly on the rDNA chromatin. This mechanism therefore stabilizes the multiprotein complex that contains GSK3β and its substrate NM1 and contributes to defining the structure and organization of the pol I machinery with respect to its chromatin template for pol I transcription activation and cell cycle progression.
In summary we propose a novel gatekeeping function for GSK3β across the rDNA, where GSK3β targets NM1 and consequently controls local chromatin modifications compatible with rRNA synthesis by instructing G1 cells to slow down NM1 degradation. In the GSK3β knockouts, cell cycle progression and in particular, the transition to S-phase occurs more rapidly than in wild type cells. The dual mode of GSK3β activity ensures that cell cycle progression is kept under tight regulation in proliferating cells [20]. It is therefore tempting to speculate that NM1 is a novel proliferative factor that is directly stabilized by GSK3β at G1 and it is likely to be altered concomitantly with inactivation of GSK3β, possibly in response to intracellular signaling.
The CGR11 antibody against GSK3β was designed as peptide specific polyclonal antibody against the N-terminal amino acid sequence GRPRTTSFAE and affinity purified against the same epitope (Agrisera AB, Sweden). The anti-GSK3β antibody 27C10 was purchased from Cell Signaling (9315). Antibodies against WSTF (ab50850), SNF2h (ab3749), H3K9Ac (ab10812), Ki67 (ab15580) and the non-specific rabbit IgGs (ab46540) were from Abcam. The mouse anti-PCAF (sc13124) and anti-UBF antibodies, the rabbit anti-UBF (sc9131), anti-Nucleolin (sc-13057) and anti-Cyclin D1 (sc-718) antibodies were purchased from Santa Cruz Biotech, whereas the anti-β-actin antibody (clone AC74) was from Sigma Aldrich. The V5 epitope antibody (A190-120A) was purchased from Bethyl Laboratories. The human autoimmune sera S57299 specific for the RPA194 pol I subunit was a kind gift of U. Scheer (Wurzburg University, Germany) [10] and the antibody against NM1 has previously been characterized [6]. The monoclonal antibody to bromouridine triphosphate (BrdU) to monitor FUrd incorporation was from Sigma Aldrich. Species-specific secondary antibodies conjugated to Cy2, Alexa 488, Alexa 568, Alexa 594 or Texas-Red were purchased from Invitrogen and Jackson ImmunoResearch. The secondary antibodies conjugated to FITC and Texas-Red were from DakoCytomation. DNA was revealed by DAPI staining (300 nM for 3 min at room temperature, RT).
The GSK3β +/+ MEFs and GSK3β−/− MEFs were a kind gift from J.R. Woodgett (University of Toronto, Canada). MEFs, HeLa and HEK293T cells were grown in DMEM medium (Gibco), supplemented with 10% foetal bovine serum (Gibco) and a 1% penicillin/streptomycin cocktail (Gibco). For synchronization in G1, subconfluent cells were grown in serum-free media for 24 hours or grown until contact inhibition [35]. The cells were released from the G1 block by adding back serum. Where indicated, GSK3β+/+ MEFs, GSK3β−/− MEFs and HeLa cells were incubated with MG132 (Cayman Chemical, ref no 10012628) to a final concentration of 40 µM for 3 h at 37°C. GSK3β+/+ MEFs were also incubated with BIO (Sigma Aldrich, B1686) to a final concentration of 1 µM for 24 hrs at 37°C. HEK293T cells and HEK293T cells constitutively expressing V5-tagged wtNM1 and RK605AA NM1 point mutant, as well as ΔIQ NM1 and ΔC NM1 deletion mutants were previously characterized [9] and are gifts of I. Grummt (University of Heidelberg, Germany). The plasmid expressing the HA-tagged Ubiquitin was a gift of O. Sangfelt (Karolinska Institutet, Sweden).
Flow cytometry (FACS) was performed as described [9]. Briefly, cells were collected by trypsinization and fixed in 70% ethanol on ice for 15 min. The DNA was stained with propidium iodine (PI) solution containing 50 µg/ml PI, 0.1 mg/ml RNasaA and 0.05% Trition X-100 in PBS (phosphate buffer saline) at 37°C for 40 min. Cells were then FACS on FACSCalibur (Becton Dickinson) and 10000 cells were counted. The experiment was repeated three times.
ChIP on growing or synchronized GSK3β+/+ MEFs, GSK3β−/− MEFs was performed as previously described [9]. Briefly, formaldehyde cross-linked chromatin was obtained from growing cells and from early G1 cells, treated or untreated with BIO (1 µM) and MG132 (40 µM) as indicated. Cross-linked chromatin was immunoprecipitated with antibodies to pol I (S57299), UBF, WSTF, SNF2h, NM1, Actin, H3K9Ac, PCAF, GSK3β (CGR11 and 27C10) and non-specific rabbit IgGs. DNA-protein complexes were analyzed by qPCR with specific primers amplifying multiple regions of the rRNA gene, including promoter, 18S, 5.8S, 28S and IGS (see Table S4 for sequences). qPCR was performed using SYBR-green from Applied Biosystems according to the manufacturer's instructions. The primer concentration was 2.5 mM and the samples analyzed by Rotor-Gene 6000 series software 1.7. The PCR conditions were: hold 95°C for 3 minutes, followed by cycles of 95°C for 3 seconds, 60°C for 20 seconds, 72°C for 3 seconds. The results were analyzed using an average of Ct of IgG as background. The 2ΔCt of each sample in triplicates was related to the 2ΔCt of the input sample.
For ChIP-Seq analysis, crosslinked chromatin from GSK3β+/+ MEFs was subjected to immunoprecipitations with the GSK3β antibody CGR11. 5 ng of precipitated DNA was used to prepare sequencing libraries at the Bejing Genome Institute (Hong Kong) using the Illumina HiSeq 2000 platform. For sequencing data alignment and analysis, the current assembly of mouse reference genome is missing the ribosomal rDNA repeats, of which there are approximately 400, located on chromosomes 12, 14 and 15. To compensate for this we utilized the procedure of Zentner et al (2011) [28] and constructed a custom reference sequence that contained a non-masked rDNA repeat, taken from BK000964, which was added to the end of chromosome 12 of the MM9 assembly. This allows for the identification and mapping of sequences present within the rDNA repeat region that are associated with the ChIP pulldown. Without the addition of the rDNA to the assembly these sequences would be removed from the analysis pipeline. The method has previously been shown to be a robust technique to identify regions within the genomic rDNA region that are associated with transcription factors, silencing or enhancing factors or histone modifications. The ChIP-Seq data sets are available in the Gene Expression Omnibus (GEO) database with accession number GSE57153. The analysis procedure involved the use of the SOAP2 program to map the reads to the constructed reference genome. Sequences with more than two mismatches were discarded from further analysis. The resulting individual sequences were remapped back to the annotated UCSC MM9 reference sequence, which allows for the identification of peaks corresponding to the levels of association of the ChIP target with those loci.
MeDIP assays were carried out essentially as previously described [30], [31]. In brief, genomic DNA was extracted using the Qiagen QIAmp DNA kit from GSK3β+/+ and GSK3β−/− MEFs. DNA was sonicated to make 200–500 bp fragments and subsequently denatured. Immunoprecipitation with anti-5-methyl-cytodin (Abcam) antibody was done overnight. The complexes were captured with protein A Sepharose. The methylated DNA was finally eluted with minElute PCR purification kit (Qiagen) and qPCR for the rDNA promoter region was run. Bars represent percent of input.
Immunoprecipitation assays were performed as previously described [9]. Endogenous GSK3β from nuclear extracts of growing GSK3β+/+ MEFs nuclear lysates were incubated with the CGR11 antibody or control non-specific rabbit IgGs. Constitutively expressed V5-tagged wtNM1, RK605AA NM1, ΔIQ NM1 and ΔC NM1 mutants from HEK293T cells lysates were incubated with the anti-V5 epitope antibody and control non-specific rabbit IgGs. The antibodies were subsequently precipitated with Protein G Sepharose (Invitrogen). The beads were washed with 1XPBS supplemented with 1 mM PMSF, 0.2 % NP-40 and then resuspended in SDS-loading buffer and heat denatured. Bound proteins were resolved by SDS-PAGE and analyzed on immunoblots for GSK3β, NM1, WSTF, SNF2h, PCAF, actin or V5. Endogenous NM1 from nuclear extracts of growing GSK3β+/+ MEFs treated or untreated with MG132 (40 µM for 3 hrs at 37°C) were incubated with the CGR11 antibody or control non-specific rabbit IgGs. The co-immunoprecipitated protein fractions were resolved by SDS-containing gel electrophoresis and in gel digested with trypsin (minus the heavy and light chain gel sections). The tryptic peptides were analyzed on a RSLC nanoLC system coupled to a Velos I system (LTQ Orbitrap Velos Pro).
To monitor phosphorylation of the endogenous NM1 protein, GSK3β−/− MEFs were arrested in G1 by serum starvation and treated with 40 µM MG132 for 3 hrs at 37°C. Lysates were prepared in 20 mM Hepes pH 7.4, 0.05 mM ATP, 10 mM MgCl2, 1 mM dithiothreitol, 2 mM sodium orthovanadate, and then incubated with 5 µCi of γ-33P-ATP and recombinant GSK3β (Abcam) for 30 min at 30°C. The lysates were next subjected to immunoprecipitations with the anti-NM1 antibody as previously described [9]. The immunoprecipitated NM1 was resolved by SDS PAGE and the proportion of phosphorylated NM1 within the immunoprecipitated protein fraction was detected by phosphorimaging. Alternatively, the kinase assays were performed on immunoprecipitated NM1 or actin from lysates of GSK3β−/− MEFs treated with MG132 still coupled to the Sepharose beads. Briefly, the beads were washed with 1X PBS containing 0.5% NP-40 and supplemented with γ-33P-ATP and recombinant GSK3β for 30 min at 30°C. The immunoprecipitated protein fraction was detected by phosphorimaging and immunoblotting for NM1 and actin.
Phosphate affinity gel electrophoresis for detection of endogenous NM1 and actin phosphorylation was performed as previously described [9]. Lysates prepared from growing or G1 synchronized GSK3β+/+ MEFs and GSK3β−/− MEFs were separated by 8% SDS-PAGE containing 25 µM Phos-tagTM AAL-107 according to the manufacturer's instructions (MANAC Incorporated) and 50 µM MnCl2, and transferred to a PVDF membrane using a transfer buffer containing 25 mM Tris, 86 mM glycine and 10% methanol. Immunoblots were analyzed for NM1 and actin. Where indicated extracts were subjected to alkaline phosphatase treatment as described in the instruction manual provided by the manufacturer (New England Biolabs). For identification of phosphorylated residues, NM1 was immunoprecipitated from nuclear lysates prepared from GSK3β+/+ MEFs and GSK3β−/− MEFs, resolved by SDS PAGE and in gel digested with trypsin. The tryptic peptides were analyzed by tandem mass spectrometry.
GSK3β+/+ MEFs, GSK3β−/− MEFs, HeLa and HEK293T cells on 10 cm dishes were transfected with 7–10 ng of a plasmid expressing HA-tagged ubiquitin using Lipofectamine 2000 as described in the instruction manual (Invitrogen). Following 24 h expression cells were treated with 40 µM MG132 for 3 hrs at 37°C and lysed in SDS containing lysis buffer (1 % SDS, 25 µl saturated NEM in PBS). Lysates were denatured and the SDS diluted to 0.1% with 1X PBS. Lysates were subjected to immunoprecipitations with the anti-NM1 antibody overnight at 4°C and precipitated with Protein G Sepharose. The beads were washed in 0.5% NP-40 buffer, resuspended in SDS-containing buffer and heat denatured. Samples were resolved by SDS-PAGE and transferred on PVDF membrane for immunodetection of ubiquitin with anti-HA epitope antibody. Where indicated, ubiquitination assays were performed on GSK3β−/− MEFs and HeLa cells subjected to GSK3β, UBR5 or Fbxw8 gene silencing by RNAi (see below).
For the GSK3β gene silencing, HeLa cells were subjected to GSK3β RNAi oligonucleotides (target sequence 5′ GGACCCAAAUGUCAAACUA) or control scrambled RNAi (scrRNAi) oligonucleotides (5′ UCGUUGCAGGAUAUGUAGUUUUU). GSK3β gene silencing duplexes and control scrambled versions were purchased from Dharmacon and applied by transfection with Lipofectamine RNAiMax (Invitrogen) at a final concentration of 400 pmol for 24 hrs. For the UBR5 and Fbxw8 genes silencing, GSK3β−/− MEFs were subjected to RNAi oligonucleotides (Dharmacon) to UBR5 (target sequence 5′-GGGUGUACAUUCUUUAAUA) and Fbxw8 (target sequence 5′-CGCCAAGGAGCACACAUUA) applied by transfection with Lipofectamine RNAiMax at a final concentration of 400 pmol for 24 hrs.
To reveal active pol I transcription foci, living GSK3β+/+ MEFs and GSK3β−/− MEFs grown on cover slips were pre-incubated with DMEM supplemented with 75 µM DRB (Sigma Aldrich) for 1 h. The FURD (Sigma-Aldrich) was then added to a final concentration of 2 mM and cellular uptake was allowed for up to 10 min [9], [29]. Cells were fixed with a 3.7% formaldehyde solution in PBS at room temperature and permeabilized with a 0.5% Triton X-100 solution in PBS. For detection of incorporated FURD, fixed cells were incubated with a mouse monoclonal antibody to BrdU followed by a Cy3-conjugated goat anti-mouse secondary antibody. Fluorescence images were obtained from a confocal microscope (Zeiss LSM meta) with 63X oil objective NA 1.3. Images were collected and analyzed using the LSM software.
For analysis of nascent pre-rRNA, total RNA was extracted from growing GSK3β+/+ MEFs, GSK3β−/− MEFs and GSK3β-silenced HeLa cells with the TRI reagent as specified by the manufacturer (Sigma). 1 ng of RNA was reversed transcribed and analysis by qRT-PCR with specific primers amplifying mouse and human 45S pre-rRNA relative to β-actin mRNA. qRT-PCR was performed using SYBR-green from Applied Biosystems according to the manufacturer's instructions (see also above for further details). Where indicated the same analysis was performed on nascent 45S pre-rRNA in GSK3β+/+ MEFs, GSK3β−/− MEFs synchronized in G1, both by serum starvation and contact inhibition. Relative 45S pre-rRNA levels were measured by qRT-PCR against the levels of β-tubulin mRNA (primers sequences are shown in Table S4). The qRT-PCR values are shown as bars diagrams. Error bars represent the standard deviation of three independent experiments. Significances were obtained by Student's T-test, two-sample, equal variance.
For immunolocalization of nucleolin and UBF, GSK3β+/+ MEFs and GSK3β−/− MEFs cells were grown on coverslips to subconfluence and arrested in G1 by growing in serum-free medium for 24 hours. The cells were fixed with 4% formaldehyde in PBS for 15 min, permeabilized with 0.1% Triton X-100 in PBS for 13 min at room temperature and stained with primary antibodies against Nucleolin and UBF following standard procedures. Secondary antibodies conjugated to FITC and Texas-Red were used to visualize Nucleolin and UBF, respectively. The slides were mounted in Vectashield containing DAPI (Vector Laboratories), and examined and photographed with an Axioplan fluorescence microscope (Carl Zeiss). Cells in random areas of the preparations were classified into three groups according to the number of positively stained nucleoli per cell (one to three, four to eight, more than eight).
Subconfluent GSK3β+/+ MEFs and GSK3β−/− MEFs cells were arrested in G1 by growing in serum-free medium for 24 hours. The cells were pelleted and fixed with 2% glutaraldehyde (Merck) in Sorensen's phosphate buffer, washed with Sorensen's buffer, and embedded in 2% low melting point agarose. The agarose blocks were cut into small pieces, dehydrated in a graded series of ethanol at room temperature, and embedded in Agar 100 resin (Agar Scientific Ltd). The embedded cell pellets were cut into 50 nm thin sections, mounted on 100 mesh copper grids, and stained with 2% uranyl acetate in 50% ethanol for 5 min at room temperature. The specimens were examined and photographed in a transmission electron microscope Tecnai G2 Spirit BioTwin (FEI Company) at 80 kV. Photoshop software (Adobe) was used for the preparation of composite images and for adjustment of intensity and contrast.
|
10.1371/journal.pgen.1007793 | Synthetic STARR-seq reveals how DNA shape and sequence modulate transcriptional output and noise | The binding of transcription factors to short recognition sequences plays a pivotal role in controlling the expression of genes. The sequence and shape characteristics of binding sites influence DNA binding specificity and have also been implicated in modulating the activity of transcription factors downstream of binding. To quantitatively assess the transcriptional activity of tens of thousands of designed synthetic sites in parallel, we developed a synthetic version of STARR-seq (synSTARR-seq). We used the approach to systematically analyze how variations in the recognition sequence of the glucocorticoid receptor (GR) affect transcriptional regulation. Our approach resulted in the identification of a novel highly active functional GR binding sequence and revealed that sequence variation both within and flanking GR’s core binding site can modulate GR activity without apparent changes in DNA binding affinity. Notably, we found that the sequence composition of variants with similar activity profiles was highly diverse. In contrast, groups of variants with similar activity profiles showed specific DNA shape characteristics indicating that DNA shape may be a better predictor of activity than DNA sequence. Finally, using single cell experiments with individual enhancer variants, we obtained clues indicating that the architecture of the response element can independently tune expression mean and cell-to cell variability in gene expression (noise). Together, our studies establish synSTARR as a powerful method to systematically study how DNA sequence and shape modulate transcriptional output and noise.
| The expression level of genes is controlled by transcription factors, which are proteins that bind to genomic response elements that contain their recognition DNA sequence. Importantly, genes are not simply turned on but need to be expressed at the right level. This is, at least in part, assured by the sequence composition of genomic response elements. Here, we studied how the recognition DNA sequence influences gene regulation by a transcription factor called the glucocorticoid receptor. Specifically, we developed a method to test the activity of variants in a highly parallelized setting where everything is kept identical except for the sequence of the binding site. The systematic analysis of tens of thousands of sequence variants facilitated the identification of a previously unknown sequence variant with high activity. Moreover, we report how sequence variation of the response element influences cell-to-cell variability in expression levels. Finally, we observe similar activity profiles for distinct sequence variants that share similar three-dimensional DNA shape characteristics arguing that the three-dimensional perception of DNA by the glucocorticoid receptor, modulates its activity towards individual target genes.
| The interplay between transcription factors (TFs) and genomically encoded cis-regulatory elements plays a key role in specifying where and when genes are expressed. In addition, the architecture of cis-regulatory elements influences the expression level of individual genes. For example, transcriptional output can be tuned by varying the number of TF binding sites, either for a given TF or for distinct TFs, present at an enhancer [1, 2]. Moreover, differences in its DNA-binding sites can modulate the magnitude of transcriptional activation, as exemplified by the glucocorticoid receptor (GR), a hormone-activated TF [3–5]. The sequence differences can reside within the 15 base pair (bp) core GR binding sequence (GBS) consisting of two imperfect 6 bp palindromic half-sites separated by a 3 bp spacer. Although the effects on activity are more modest than those observed for changes within the core, sequences directly flanking the core also modulate GR activity [3]. However, these sequence-induced changes in activity cannot be explained by affinity [3]. Instead, the flanking nucleotides induce structural changes in both DNA and the DNA binding domain of GR, arguing for their role in tuning GR activity [3].
Notably, the expression level of a gene is typically measured for populations of cells and thus masks that expression levels can vary considerably between individual cells of an isogenic population [6–9]. This variability in the expression level of a gene, called expression noise, results in phenotypic diversity, which can play a role in organismal responses to environmental changes (so called bet-hedging) and in cell fate decisions during development. Expression noise can be explained by the stochastic nature of the individual steps that decode the information encoded in the genome. For example, transcription occurs in bursts [7, 10–12], which can induce variability in gene expression due to differences in burst frequency and in the number of transcripts generated per burst (burst size) [13]. Noise levels are gene-specific, which can be explained in part by differences in the sequence composition of cis-regulatory elements [11, 14–16]. For instance, the sequence composition of promoters influences expression variability with high burst size and noise for promoters containing a TATA box [15, 17]. In addition, chromatin and the presence or absence of nucleosome-disfavoring sequences have been linked to transcriptional noise [16–19]. Finally, noise levels can also be tuned by the number and by the affinity of TF binding sites [11, 16].
Many fundamental insights regarding the role of sequence in tuning transcriptional output and noise have come from reporter studies [20, 21]. A key advantage of reporters is that they can provide quantitative information in a controlled setting where everything is kept identical except for the sequence of the region of interest. Until recently, a limitation of reporter studies was that sequence variants had to be tested one at a time. However, the recent development of several parallelized reporter assays allows the simultaneous assessment of many sequence variants [21]. One of these parallelized methods is STARR-seq (Self-Transcribing Active Regulatory Region sequencing) [22]. In this assay, candidate sequences are placed downstream of a minimal promoter, such that active enhancers drive their own expression and high-throughput sequencing reveals both the sequence identity and quantitative information regarding the activity of each sequence variant. The STARR-seq method has been used to assay enhancer activity genome-wide [22, 23], to study regions of interest isolated either by Chromatin Immunoprecipitation (ChIP) or a capture-based approach [24, 25], and to study the effect of hormones on enhancer activity [25, 26].
Here, we adapted the STARR-seq method to systematically study how sequence variation both within the 15 bp GBS and in the region directly flanking it modulate GR activity. Specifically, we generated STARR-seq libraries using designed synthetic oligos (synSTARR-seq) with randomized nucleotides flanking the core GBS to show that the flanks modulate transcriptional output by almost an order of magnitude. When grouping sequences based on their ability to either enhance or blunt GBS activity, we found that each group contained a broad spectrum of highly diverse sequences, but striking similarities in their DNA shape characteristics. Using the same approach, we also assayed the effect of sequence variation within the core GBS. Finally, using single cell experiments with individual enhancer variants, we study how the sequence composition of the response element influences expression mean and noise. Together, our studies establish synSTARR-seq as a powerful method to study how DNA sequence and shape modulate transcriptional output and noise.
To test if we could use the STARR-seq reporter [22] to study how sequence variation of the GR binding site influences GR activity, we first tested if a single GBS is sufficient to facilitate GR-dependent transcriptional activation of the reporter. Therefore, we constructed STARR reporters containing either a single GBS as candidate enhancer (Fig 1A), a randomized sequence or as positive control a larger GBS-containing sequence derived from a GR-bound region close to the GR target gene FKBP5. The resulting reporters were transfected into U2OS cells stably expressing GR (U2OS-GR) [27] and their response to treatment with dexamethasone (dex), a synthetic glucocorticoid hormone, was measured. As expected, no marked hormone-dependent induction was observed for the reporter with the randomized sequence. This was true both at the level of RNA (Fig 1B) and at the level of the GFP reporter protein (S1 Fig). In contrast, we observed a robust hormone-dependent activation both at the level of RNA and GFP protein for reporters with either a single GBS or with the larger genomic FKBP5 fragment (Fig 1B and S1A Fig), showing that a single GBS is sufficient for GR-dependent activation of the STARR-seq reporter.
Our previous work has shown that the sequence directly flanking GBSs can modulate DNA shape and GR activity [3]. For a parallelized and thorough analysis of sequence variants flanking a GBS, we generated STARR-seq libraries for two GBS variants, we previously named Cgt and Sgk, that showed a strong influence of flanking nucleotides on activity [3]. Specifically, we generated libraries using designed synthetic sequences (synSTARR-seq) containing a GBS with five consecutive randomized nucleotides directly flanking the imperfect half site (Fig 1A and S2A Fig). Next, we transfected the GBS flank libraries into U2OS-GR cells to determine the activity of each of the 1024 flank variants present in the library. We performed three biological replicates for each condition and found that the results were highly reproducible (r ≥ 0.91 for vehicle treated cells, r ≥ 0.98 for dex treated cells; Fig 1C and S1B–S1E Fig). Notably, we retain duplicate reads in our analysis, which is essential to get quantitative information for individual sequence variants of the library. To calculate the activity for each flank variant, we used DESeq2 [28] to compare the RNA-seq read number between dex- and vehicle (ethanol) treated cells (Fig 1A). This resulted in the identification of 189 flank variants with significantly higher activity (enhancing flanks), 125 flank variants with significantly lower activity (blunting flanks) and 710 flank variants that did not induce significant changes in activity (neutral flanks). To test the accuracy of the synSTARR-seq data, we cloned 5 flank variants from each activity group (enhancing, blunting and neutral) and assayed the activity of each variant individually by qPCR. Consistent with what we observed for the synSTARR library, the activity of blunting flanks was significantly lower than for the neutral flanks whereas the activity of the enhancing flanks was significantly higher (Fig 1D). Notably, all flank variants tested were activated upon dex treatment ranging from 2.1 to 15.3 fold (627% higher) depending on the sequence of the flank. Together, our results show that the synSTARR-seq assay produces reproducible and quantitative information and can be used for a high-throughput analysis of the effect of the flanking sequence on GBS activity.
To assess how the sequence composition of the flanking region influences GBS activity, we ranked the flank variants by their activity and used a color chart representation to plot the sequence at each position for the Cgt (Fig 2A) and Sgk GBS (S2A Fig), respectively. In addition, we generated consensus sequence motifs for the significantly enhancing and blunting variants (Fig 2B and S2B Fig). Notably, these consensus sequence motifs treat each sequence equally and do not take the quantitative information regarding the activity of each sequence into account. To take advantage of the quantitative information provided by the synSTARR-seq assay, we used kpLogo [29], which uses the fold change as weight for each sequence variant, and statistically evaluates the enrichment/depletion of specific nucleotides at each position. The resulting probability logo can be interpreted as an activity logo that visualizes for each position which nucleotides are associated with either higher (letters above the coordinates) or lower (below the coordinates) GBS activity (Fig 2C and S2C Fig). The activity logo, consensus motifs and color chart highlight several sequence features for enhancing and blunting flank variants. For example, high activity is associated with a T at position 8 for both the Cgt and Sgk GBS, which matches what we found previously when we studied the activity of endogenous GR-bound regions [3]. In addition, the most active flank variants preferentially have an A at position 9 followed by a C at position 10 (Fig 2A and S2A Fig). To validate that this “TAC” signature results in high activity, we shuffled the sequence to either TCA or CAT and found that this indeed resulted in markedly lower activity (Fig 2D). For blunting flank variants, we observed a preference for an A at position 8 and a bias against having a C at position 10 (Fig 2A and 2C and S2A and S2C Fig). However, altogether we find that the consensus motifs for enhancing and blunting flanks only have low information content and that a broad spectrum of distinct sequences can enhance or blunt the activity of the adjacent GBS (Fig 2B and S2B Fig).
Our previous work [3] indicates that DNA shape can influence GR activity downstream of binding. Consistent with this notion, we measured similar Kd values for flanks variants from the different activity classes (Fig 2E). These findings are also in agreement with published work showing that the nucleotides directly flanking GBSs have little effect on GR affinity [30]. To examine if the flank effects might be explained by differences in DNA shape, we calculated the predicted minor groove width, roll, propeller twist and helix twist [31] for enhancing and blunting flank variants (Fig 3A and S2D Fig and S3 Fig). Consistent with a role for DNA shape in modulating GR activity, we found shape characteristics that differ between enhancing and blunting flanks. For example, we observed a wider minor groove at position 6, and to a lesser degree at position 7 for blunting flanks of the Cgt GBS, when compared to enhancing flanks (Fig 3A and S4A Fig). In addition, blunting flanks for the Cgt GBS have a narrower minor groove than enhancing flanks for positions 8–12 (Fig 3A and S4A Fig), a region with several non-specific minor groove contacts with the C-terminal end of the DNA binding domain of GR [5]. For the Sgk GBS library, we find similar shape characteristics associated with blunting flanks with a wider minor groove at position 6 and a narrower minor groove for positions 8–12 (S2D Fig and S4B Fig). DNA-shape-based hierarchical clustering recapitulates these characteristics in cluster 4, containing many more blunting flanks than any of the other clusters, for both the Cgt and Sgk GBS flank libraries (Fig 3B and 3C and S2E and S2G Fig). Of note, the consensus motifs for cluster 4 and for the other shape clusters have only low information content (Fig 3D and S2F Fig) indicating that distinct sequences can give rise to similar shape characteristics with shared effects on the activity of the adjacent GBS.
Together, these synSTARR-seq experiments uncover how sequence variation in the flanking region of the GBS influences activity and point at a role for DNA shape in modulating GBS activity.
We next generated an additional synSTARR-seq library to study the effect of variation within the 15bp core sequence. This library contains a fixed GBS half site followed by eight consecutive randomized nucleotides (Fig 4A). The library, containing over 65,000 variants, was transfected into U2OS-GR cells and the read count for each variant was determined both in the presence and absence of hormone treatment. Compared to the flank library, we observed a lower correlation between experiments, especially for variants with a low read count (S5 Fig). Specifically, when we compared the read count between biological replicates, we found that sequences with a read count below 100 were typically detected in only one of the replicates. Therefore, we decided to remove sequences with a mean read count below 100 across all experiments. Next, we analyzed data from three biological replicates to determine the activity of variants in the library (Fig 4B). To validate the measured activities, we cloned 4 sequences that repress, 4 that show a weak activation (log2 fold change <2) and 8 strongly activating GBS variants. Consistent with the results from our screen, the three groups showed distinct levels of activity (Fig 4B and 4C). However, for the group of repressed GBS variants we did not recapitulate the observed repression in our screen (Fig 4C), indicating that these variants might behave differently in isolation. Alternatively, what looks like repression might be a consequence of issues with data normalization, which assumes that the distribution of the log fold changes is centered on 0, which is not given when GBS variants can activate but not repress gene expression. Notably, a lack of GR-dependent transcriptional repression was also reported in another study using the STARR-seq approach to study the regulatory activity of GR-bound genomic regions [25] indicating that GR might not be able to repress transcription in the STARR-seq context.
Given that the observed repression was not reproducible, we concentrated our analysis on 1696 sequences that facilitated significant GR-dependent transcriptional activation. Consistent with activation, we found that the consensus motif for activating sequence variants recapitulates the known GR consensus sequence with the second half site 3-bp downstream of the fixed first half site of our library (Fig 4D). Accordingly, the GBS motif weight, which serves as a proxy for DNA binding affinity, is higher for activating sequences when compared to sequences that did not respond to hormone treatment (Fig 4G). However, the score for the top 10% most active sequences was not higher than for all active variants (Fig 4G), arguing that higher affinity does not drive the high levels of activation. As expected and consistent with the GR consensus motif, the color chart (Fig 4D) and activity logo (Fig 4E) highlight a strong preference for a G at position 3 and accordingly GBS activity is significantly lower for variants with a nucleotide other than G at this position (S6A Fig). The activity logo also highlights that a G at position 2 is associated with lower activity (Fig 4E and 4F).
Previous studies have shown that the sequence of the spacer can modulate GBS activity [4, 5]. Therefore, we compared the activity of all 16 spacer variants in our library that match the GBS consensus for the second half site at the key positions 3, 4 and 6 (S7A Fig). In line with a role for the spacer in modulating transcriptional output, we find significant differences between the spacer variants (S7B Fig). For example, the activity for variants with an AC spacer is significantly higher than for several other spacer variants (S7B Fig) whereas the activity for GT variants is significantly lower (p.adj < 0.01) than either AA, AC or TC variants (S7B Fig).
Unexpectedly, the activity logo and top of the color chart indicated a high activity for variants with a C at position 2 (Fig 4D and 4E), instead of a consensus T observed in the GR consensus motif and from in vitro experiments studying the effect of DNA sequence on GR DNA binding affinity [30]. A careful examination of the sequence composition of the most active variants also revealed a preference for TC at the preceding positions within the spacer (Figs 4E and 5A). To test if the high activity for sequences with a C at position 2 depends on the nucleotide composition of the preceding nucleotides, we changed them to GG and found that this resulted in a marked reduction in GR-dependent activation (Fig 5B and S8A Fig). In addition, we compared the activity between variants with a T or a C at position 2. The activity was higher for the C variant when preceded by TC. However, when we changed the preceding nucleotides to GG the activation was stronger for the T than the C variant (Fig 5B and S8A Fig). These experiments indicated that the high activity for the C variant depends on the preceding nucleotides.
Interestingly, the most active variants resemble the sequence composition of the “combi” motif we identified previously [32]. The combi motif contains only a single GR half site followed by TTCC and we found evidence that GR binds this sequence as a monomer in conjunction with a partnering protein [32]. Similar to the combi motif, several of the most active variants (Fig 5A) contain a GR half site followed by TTCC. However, whereas the combi motif lacks a second GR half site, the motif for the 25 most active variants from our screen (named “combi2”) also contains a recognizable second GR half site (Fig 5A). To gain insight into the mode of GR binding at the combi2 motif, we examined published ChIP-exo data [32]. ChIP-exo is an assay that combines ChIP with a subsequent exonuclease step [33] which results in a base-pair resolution picture of GR binding. The ChIP-exo signal takes the form of sequence-specific peak patterns (footprint profiles), detectable on both strands with the program ExoProfiler [32]. We applied ExoProfiler to scan GR-bound regions with the combi2 motif (Fig 5D and 5E, solid lines). As control, we analyzed the footprint profile for the canonical GR consensus motif (Fig 5D; JASPAR MA0113.2) and recovered peak pairs on the forward and reverse flanks that demarcate the protection provided by each of the monomers of the GR dimer (Fig 5E, shaded area). The signal for the first half site is essentially the same and a similar pattern is also observed for the second half site, indicating that GR binds as a dimer on regions bearing the combi2 motif, however with additional signal (highlighted with black arrows in Fig 5E). In addition, we compared the footprint profile between the original combi (Fig 5D; [32]) and the combi2 motif (Fig 5F). Again, the position and shape of the peaks are compatible for the first half site but the ChIPexo signal for the second half site looks markedly different. The aforementioned additional signal for the combi2 motif aligns with the position of the second peak pair of the combi motif (Fig 5F), indicating that the footprint profile for the combi2 motif appears to be a composite of the signal for homodimeric GR binding at canonical GBSs and the signal for monomeric GR binding together with another protein. Our previous work suggests that this partnering protein on combi motif might be Tead or ETS2. The ChIP-exo profile thus points to three alternative binding configurations on combi2: homodimeric GR, monomeric GR binding with Tead/ETS2 or the simultaneous binding of homodimeric GR complex together with Tead/ETS2. Structural modeling suggests that this third mode is possible given the absence of obvious sterical clashes that would prevent this mode of binding (Fig 5G). However, additional functional studies are needed to determine if GR indeed partners with Tead/ ETS2, or possibly with other proteins, at the combi2 motif.
To assess if DNA shape could play a role in modulating GBS activity, we calculated the predicted minor groove width for all 1696 significantly activated sequences ranked by activity (S6B Fig). Comparison of the top 20% most active and bottom 20% least active sequence variants highlighted two regions with significant differences. First, consistent with our findings for the flank library, we find that a wider minor groove at positions 6 and 7 correlates with weaker activity (S6B and S6C Fig). Second, we find that a narrower minor groove in the spacer (position -1 and 0) correlates with weaker activity (S6B and S6C Fig). As we observed for the flank variants, the different activity classes do not show a distinct sequence signature (S6B Fig) again arguing that DNA shape might modulate GBS activity.
Together, the findings for our half site library suggest a role for both DNA shape and sequence in tuning the activity of GBS variants. Moreover, our screen uncovered a novel high-activity functional GR binding sequence variant.
Thus far, we analyzed the effect of sequence composition on transcriptional output by analyzing mean expression levels for populations of cells. To test if sequence variation in the enhancer influences cell-to-cell variability in gene expression (noise), we measured GFP levels for individual STARR constructs in single cells (Fig 6A and 6B). Cells were transfected with individual constructs along with an mCherry expression construct to remove extrinsic noise, for example caused by differences in transfection efficiency. We first analyzed sequence variants containing a single GBS (single GBS group) including known GBSs; two variants matching the combi2 sequence motif and the Cgt GBS with an enhancing flank variant. Consistent with previous findings [5], we found that GBS variants from the single GBS group induced different mean levels of GFP expression. For example, the mean GFP level upon dex treatment was lower for the Pal GBS than for the Cgt variant (Fig 6C, orange and red squares). In line with findings by others [16], we observed that transcriptional noise scales with mean expression with lower noise for variants with higher mean expression (Fig 6C). Next, we assayed two additional groups of sequences with distinct binding sites architectures that both result in more robust GR-dependent activation when compared to single GBS variants (Fig 6A). The first group contained three instead of one GBS copy (triple GBS group) whereas the second group (composite group) contains a GBS flanked by a sequence motif for either AP1, ETS1 or SP1, three sequence motifs that can act synergistically with GR [34, 35]. As expected, the mean GFP expression was higher for each member of both the triple GBS and the composite group when compared to the single GBS group (Fig 6A and 6C). Interestingly, the increase in mean expression we found for the groups of triple GBS and composite enhancers was not accompanied by a decrease in expression noise (Fig 6C). The high noise to mean expression ratio was especially striking for several triple GBS variants (3xPal, 3xCgt, 3xSgk and 3x Fkbp5-2) but observed in general for each member of the groups of triple and composite enhancers when compared to the single GBS group. Furthermore, enhancer variants with similar mean expression levels (e.g. 3xSgk and Ets1+FKBP5-2) can have vastly different noise levels indicating that binding sites architecture can independently tune both mean expression and cell-to-cell variability in gene expression with noisier expression for enhancers with multiple GBSs.
In this study, we developed a modified version of the STARR-seq method where we used designed synthetic oligonucleotides to assay how sequence variation within and around the GBS influence GBS activity. This facilitated the thorough and parallelized assessment of 1024 flank variants on GBS activity in a highly reproducible and quantitative fashion (Fig 1 and S1 Fig). Similarly, we assessed over 65,000 variants to study how variations in one of the half sites and the spacer influence GBS activity. Taken together, we find that variation in both the half site and in the region flanking the GBS influences GR activity. Quantitatively however, changes in the half site have more profound effects on activity than those in the flanking region (Fig 2D, Fig 4B). A key advantage of using designed sequences over the analysis of genomic regions is that variants can be compared in a context where everything is identical except for the sequence of the GR binding site. Notably, the sequence of the binding site is just one of several signals that are integrated at genomic response elements to modulate GR-dependent transcriptional responses. Therefore, our synSTARR-seq data for a single GBS is unlikely to yield accurate predictions for the activity of GR response elements in the genomic context. Other inputs that could improve predictions include chromatin environment and information regarding how the presence or absence of binding sites for other TFs influences GR-dependent activation. The synSTARR-seq approach can readily be adapted to study how combinations of signals are integrated. For example, principles of combinatorial regulation can be studied using designed sequences for which the GBS is flanked by binding sites for other TFs. Similarly, the assay can be used to investigate the cross-talk between GBS sequence, ligand chemistry, type of core promoter and GR splice isoforms.
Importantly, our findings for the synthetic STARR-seq assay are consistent with GR-dependent regulation of endogenous target genes. Specifically, the nucleotide directly flanking the GBS is preferentially a T for both enhancing flanks in our synSTARR-seq experiments and for the motif we previously found for genomic GR binding sites associated with genes that show the most robust response to GR activation [3]. Moreover, we uncovered a novel functional GR binding sequence variant with high activity, which we called combi2. Consistent with the high activity of the combi2 motif observed in the synSTARR assay, genes with nearby GR-bound peaks matching the combi2 motif were, on average, slightly more activated by GR than genes with peaks matching the consensus motif (based on RNA-seq data we generated for U2OS-GR cells treated for 4h with either 1μM dexamethasone or 0.1% ethanol as vehicle control; Fig 5C). Other sequence preferences we uncovered for flanks that enhance GBS activity include an A followed by a C at positions 9 and 10 respectively (Fig 2A and 2C and S2A and S2C Fig). One possible explanation for the increased activity is that this sequence generates an additional GR half site or a binding site for another TF. However, the ChIP-exo profile for GBSs flanked by nAC looked essentially the same as the profile for the canonical GBS (S9 Fig), arguing against the binding of an additional factor. Alternatively, the flanking nAC could influence GR’s DNA binding affinity. However, a comprehensive analysis of the effect of sequence variation within and in the regions flanking GR binding sites showed that the flanks essentially do not influence the binding affinity of GR [30]. Accordingly, we found similar Kd values for the AC flank when compared to variants with lower activity (Fig 2E) indicating that the change in activity is not driven by affinity. Together, the synSTARR-seq approach uncovered how sequence variation modulates GR activity, which confirmed previous findings based on a small number of sequences but also provided new insights into mechanisms that modulate GR-dependent regulation of endogenous target genes.
We were surprised to find that the consensus motifs for enhancing and blunting flanks displayed low information content indicating that a broad spectrum of distinct sequences can enhance or blunt the activity of the adjacent GBS (Fig 2 and S2 Fig). However, when looking at DNA shape we found specific shape characteristics for each group (Fig 3A, S2 Fig and S3 Fig). This indicates that distinct sequences can induce similar DNA shape characteristics with analogous effects on GBS activity. This finding was corroborated by our analysis of the spacer, which is not directly contacted by GR, yet influences GR activity. Also here we found distinct spacer shape characteristics for the most and least active GBS variants, without a clear sequence signature for each group (S7B Fig). Furthermore, we trained a model to distinguish between high and low activity GBSs based on either DNA sequence or on predicted minor groove width information. Assessment of the accuracy of the models using ROC curves showed that a single shape parameter, minor groove width, can be used to distinguish quite accurately between blunting and enhancing flanks (S10A Fig) and also between the top and bottom 20% active variants from our half site library (S10B Fig). Together, our findings which are based on a systematic analysis of many sequence variants are consistent with previous studies based on a small number of binding sites, showing that GR activity can be modulated by DNA shape [3, 4]. Notably, although the role of DNA shape in modulating the affinity of TFs for DNA has been well documented [36–38], we find that DNA shape modulates GR activity without apparent changes in DNA binding affinity (Fig 2E, [30]). This is consistent with a model where DNA shape acts as an allosteric ligand which induces structural changes in associated TFs which in turn changes the composition and regulatory activity of the complexes formed at the response element [5, 39–41]. Another, not mutually exclusive, explanation for flank-dependent modulation of transcriptional output is that flank variants serve as binding sites for other TFs that act additively or synergistically with GR. Further support for the importance of DNA shape comes from the analysis of the conservation of non-coding regions of the genome. This analysis uncovered greater conservation at the level of DNA shape than on the basis of nucleotide sequence indicating that DNA structure may be a better predictor of function than DNA sequence [42]. Accordingly, incorporation of DNA shape characteristics improves in vivo prediction of TF binding sites [43] and, based on our findings, could also improve the prediction of TF binding site activity.
We also explored if GFP protein expression levels of individual cells can be used to study how enhancer architecture influences cell-to-cell variability in gene expression. A similar approach was used to study how sequence variation of the promoter influences transcriptional noise in yeast [16]. Notably, the only difference between the reporters we assayed is their enhancer sequence, which is downstream of the ORF for the GFP protein. For sequences with related enhancer architectures (e.g. singe GBS variants), we observed that transcriptional noise scales with mean expression, such that higher expression levels are associated with lower noise (Fig 6C). This is consistent with a two-state promoter model where increases in mean expression are driven by an upsurge in transcription burst frequency [44]. Similarly, the estrogen receptor, a hormone receptor closely related to GR, modulates transcription by changing the frequency of transcriptional bursting [12]. When we compare distinct enhancer architectures, we find that expression mean and noise can be uncoupled. Specifically, the noise to mean expression ratio is higher for response elements harboring multiple TF binding sites when compared to the group of single GBS variants, indicating that the increase in expression might be accompanied by an increase in the number of transcripts produced during each burst. This finding is consistent with studies in yeast showing that increasing the number of binding sites for GCN4 results in increased expression with relatively high noise levels [16]. Notably, both multiple binding sites for GR and a combination of a GR binding site and a binding site for another TF result in an increased noise to mean expression ratio (Fig 6). Genomic GR response elements typically contain multiple GR binding sites and motifs for other TFs [25] arguing that endogenous GR-driven gene expression might be quite noisy with high levels of cell-to-cell variability in gene expression levels. However, if our findings for synthetic response elements reflect what happens at genomic response elements is unclear given that they differ in several ways. For example, in contrast to most endogenous response elements, the GBSs in our reporters are separated by 4 bp. Furthermore, each GBS sequence is identical in our reporters whereas most genomic GBS sequences have a unique sequence composition.
Our results are consistent with a model in which the architecture of the enhancer influences transcriptional burst size and frequency. However, more sophisticated single-cell studies of nascent transcripts are needed for a detailed understanding of the role of enhancer architecture given that our studies are based on the measurement of steady state fluctuations in protein levels. For example, in our experimental approach we cannot rule out that other mechanisms, including differences in RNA stability and translation rates, could contribute to the cell-to-cell variability in expression observed. Nonetheless, our findings argue that differences in enhancer architecture might contribute to gene-specific tuning of expression mean to noise ratios of GR target genes.
Taken together, we present synSTARR, an approach to measure how designed binding site variants influence transcriptional output and noise. The systematic analysis of sequence variants presented here resulted in the identification of a novel functional GR binding sequence and provides evidence for an important role of DNA shape in tuning GR activity without apparent changes in DNA binding affinity. Our simple approach using designed sequences can be applied to other TFs and can be used to systematically unravel how the interplay between sequence and other signaling inputs at response elements modulate transcriptional output.
Plasmids. STARR reporter constructs were generated by digesting the human STARR-seq vector [22] with SalI-HF and AgeI-HF and subsequent insertion of fragments of interest by in-Fusion HD cloning (TaKaRa). All inserts had the following sequence composition: 5’- TAGAGCATGCACCGGACACTCTTTCCCTACACGACGCTCT—-INSERT—-AGATCGGAAGAGCACACGTCTGAACTCCAGTCACTCGACGAATTCGGCC-3’. Sequence homologous to the STARR reporter construct in bold; Sequence for p5 and p7 adaptors underlined. The exact sequence of the insert for each construct used in this study is listed in S1 Table.
Cell lines, transient transfections and luciferase assays. U2OS cells stably transfected with rat GRα (U2OS-GR) [27] were grown in DMEM supplemented with 5% FBS. Transient transfections were done essentially as described [5] using either lipofectamine and plus reagents (Invitrogen) or using kit V for nucleofections (Lonza).
Synthetic STARR-seq. Library design and generation: To generate GBS variant libraries, oligos containing degenerate nucleotides (N) at defined positions were ordered from IDT as “DNA Ultramer oligonucleotide” (sequence listed below). The oligonucleotides were made double stranded using Phusion polymerase (NEB; 98°C for 35 sec, 72°C for 5 min) using the revPrimer (GGCCGAATTCGTCGAGTGAC). The resulting double stranded inserts (25ng) were recombined with 100ng linearized (SalI-HF and AgeI-HF) STARR-seq vector [22] by in-Fusion cloning in 5 parallel reactions. After pooling the reactions, the DNA was cleaned up using AMPure XP beads (Beckman Coulter), transformed into MegaX DH10B cells (Invitrogen) and plasmid DNA was isolated using a Plasmid Plus Maxi kit (Qiagen). STARR-seq: For STARR-seq experiments, 5 million U2OS-GR cells were transfected with 5 μg library-DNA by nucleofection using kit V (Lonza). The next day, cells were treated for 4 h with 1 μM dexamethasone or with 0.1% ethanol as vehicle control. Reverse transcription and amplification of cDNA for subsequence Illumina 50bp paired-end sequencing were done as described [22].
Cgt flank library DNA Ultramer oligonucleotide:
TAGAGCATGCACCGGACACTCTTTCCCTACACGACGCTCTTCCGATCTCAGCGCAAGAACAtttTGTACGNNNNNCTAGATCGGAAGAGCACACGTCTGAACTCCAGTCACTCGACGAATTCGGCC
Sgk flank library DNA Ultramer oligonucleotide:
TAGAGCATGCACCGGACACTCTTTCCCTACACGACGCTCTTCCGATCTCAGCGCAAGAACAtttTGTCCGNNNNNCTAGATCGGAAGAGCACACGTCTGAACTCCAGTCACTCGACGAATTCGGCC
GBS half site library DNA Ultramer oligonucleotide:
TAGAGCATGCACCGGACACTCTTTCCCTACACGACGCTCTTCCGATCTCAGCGAAAGAACAtNNNNNNNNCGTCGCTAGATCGGAAGAGCACACGTCTGAACTCCAGTCACTCGACGAATTCGGCC
RNA-seq U2OS-GR cells (Fig 5C). U2OS-GR cells were treated for 4h with either 1μM dexamethasone or 0.1% ethanol as vehicle control. RNA was isolated from 1.2 million cells using the RNeasy kit from Qiagen. Sequencing libraries were prepared using the TruSeq RNA library Prep Kit (Illumina). Prior to reverse transcription, poly adenylated RNA was isolated using oligo d(T) beads. Paired end 50bp reads from Illumina sequencing were mapped against the human hg19 reference genome using STAR [45] (options:—alignIntronMin 20—alignIntronMax 500000—chimSegmentMin 10—outFilterMismatchNoverLmax 0.05—outFilterMatchNmin 10—outFilterScoreMinOverLread 0—outFilterMatchNminOverLread 0—outFilterMismatchNmax 10—outFilterMultimapNmax 5). Differential gene expression between dex and etoh conditions from three biological replicates was calculated with DESeq2 [28], default parameters except betaPrior = FALSE.
Electrophoretic mobility shift assays. EMSAs were performed as described previously [3] using Cy-5 labeled oligos as listed in S2 Table.
RNA isolation, reverse transcription and qPCR analysis. RNA was isolated from cells treated for either 4 h or overnight with 1 μM dexamethasone or with 0.1% ethanol vehicle. Total RNA was reverse transcribed using gene-specific primers for GFP (CAAACTCATCAATGTATCTTATCATG) and RPL19 (GAGGCCAGTATGTACAGACAAAGTGG) which was used for data normalization. qPCR and data analysis were done as described [5]. Primer pairs for qPCR: hRPL19-fw: ATGTATCACAGCCTGTACCTG, hRPL19rev: TTCTTGGTCTCTTCCTCCTTG, GFP-fw: GGCCAGCTGTTGGGGTGTC, GFP-rev: TTGGGACAACTCCAGTGAAGA.
Noise-Measurements. For noise measurements, U2OS-GR cells were transfected using lipofectamine and plus (Invitrogen) essentially as described [5]. In short: The day before transfection, 40,000 U2OS-GR cells were seeded per well of a 24 well plate. The following day, cells were transfected with individual STARR reporter constructs (20ng/well) along with a SV-40 mCherry expression construct (20ng/well) and empty p6R plasmid (100 ng/ well). Transfected cells were treated overnight with either 1μM dexamethasone or with 0.1% ethanol vehicle control. Fluorescence intensity was measured using an Accuri C6 flow cytometer (BD Biosciences) and the yellow laser (552nM) and filter 610/20 for mCherry and the deepblue laser (473nM) and filter 510/20 to measure GFP. Gates were set for mCherry and GFP and only cells showing both mCherry and GFP fluorescence were included in the analysis. Relative expression of GFP (GFP/Cherry), from 800–1600 individual dexamethasone-treated cells, was used to calculate mean expression and the standard deviation of cell populations. Mean and standard deviation for noise (CV2) and for relative GFP expression were derived from three biological replicates.
Analysis of synSTARR-seq data. RNA-seq reads were filtered and only sequences exactly matching the insert sequence in length and nucleotide composition were included in the analysis. The number of occurrences for each sequence variants was counted for each experimental condition and differentially expressed sequences were identified using DESeq2 [28] using a p adjusted value <0.01 as cut-off. To fit the dispersion curve to the mean distribution, we used the local smoothed dispersion (DESeqwithfitType = "local"). Notably, each of the constructs of the flank libraries contains a functional GBS. Therefore, flanks that blunt activity will appear repressed after hormone treated because their fraction in the total pool of sequences decreases relative to flank variants with higher activities. For the flank libraries, we obtained information for each sequence variant (1024) in the library. For the half site library, we identified 61,582 out of the 65,536 possible variants present in this library. We found that including sequences with low read coverage resulted in many false positive differentially expressed GBS variants. To avoid this, we only included sequences with a mean read count above 100 across all experiments, leaving us with information for 33,689 sequence variants. The pearson correlation coefficient for replicates was calculated using the ggscatter function of the ggpubr library in R.
Boxplots comparing groups of sequence variants as specified in the figure legends show center lines for the median; box limits indicate the 25th and 75th percentiles; whiskers extend 1.5 times the interquartile range from the 25th and 75th percentiles.
Sequence logos to depict the consensus motif for groups of sequences were generated using WebLogo [46]. The probability logo (activity motif) was generated with kpLogo [29] using as input the sequence and fold change (dex/etoh) for each variant and the default settings for weighted sequences.
Motif weight. The motif weight for each variant was calculated using the RSAT matrix-scan program [47, 48]. Specifically, the motif weight was calculated using Transfac motif M00205 truncated to the core 15bp, and a custom background model created with RSAT create background program, trained on human open chromatin available at UCSC genome browser (http://genome.ucsc.edu/cgi-bin/hgTrackUi?db=hg19&g=wgEncodeRegDnaseClustered). Boxplots comparing groups of sequence variants show center lines for the median; box limits indicate the 25th and 75th percentiles; whiskers extend 1.5 times the interquartile range from the 25th and 75th percentiles.
Comparison of ChIP-seq peak height between combi2 and canonical GBS motif. GR ChIP-seq data sets for U2OS-GR cells were downloaded as processed peaks from EBI ArrayExpress (E-MTAB-2731). ChIP-seq peaks in a 40 kb window centered on the transcription start site of differentially expressed genes (RNA-seq data: E-MTAB-6738) were scanned using RSAT matrix-scan [47, 48] for the occurrence of either a GBS-match (Transfac matrix M00205, p value cut-off: 10−4) or the combi2 matrix we generated (Fig 5D, p-value cut-off 10−4). Next, peaks were grouped by motif match and median peak height was calculated for each group and the p-value comparing both groups was calculated using a Wilcoxon rank-sum test to produce Supplementary S8B Fig.
Comparison of gene regulation. To compare the level of activation between genes with nearby peaks with either a GBS match (Transfac matrix M00205, p value cut-off: 10−4) or a combi2 match (motif Fig 5D, p-value cut-off 10−4), we first scanned ChIP-seq peaks (U2OS-GR cells: E-MTAB-2731) in a 40 kb window centered on the transcription start site (using all annotated TSSs from Ensembl GRCH37) for motif matches using RSAT matrix-scan [47, 48]. Only peaks with an exclusive motif match were retained to generate a boxplot comparing the log2 fold change for genes of each group (RNA-seq data: E-MTAB-2731). Center lines show the median, box limits indicating the 25th and 75th percentiles and whiskers extending 1.5 times the interquartile range from the 25th and 75th percentiles. p-value comparing the log2 fold change for both groups was calculated using a Wilcoxon rank-sum test to produce Fig 5C.
DNA shape prediction. We used DNAshapeR [31] to predict the minor groove width, roll, propeller twist or helix twist for sequence variants of interest. Boxplots for individual nucleotide position show center lines for the median; box limits indicate the 25th and 75th percentiles; whiskers extend 1.5 times the interquartile range from the 25th and 75th percentiles. The Wilcoxon rank-sum test was used to calculate the p-values comparing nucleotide position variants between groups. Individual sites were clustered using K-means clustering with k = 4 clusters nstart = 20 and 100 restarts with the function 'kmeans' from the R 'stats' package.
Classification of GBS activity. To assess classifier performance we generate ROC curves using 10-fold cross-validation. Four different models were tested to classify GBS activity into blunting or enhancing. A mononucleotide model consisting of sequence motifs estimated from relative nucleotide frequencies within the two classes. Class affiliation is predicted with a likelihood ratio test. We also tested a similar model based on dinucleotides. In addition, we tested two random forest (RF) classifiers with 100 trees, based on sequence and shape information. We used the R package "randomForest" for constructing the classifiers [49]. Since RF classifiers are not designed for categorical data, we coded nucleotide sequences using 00 for 'A', 01 for 'C', 10 for 'G', and 11 for 'T'.
ChIP-exo footprint profiles. ChIP-exo footprint profiles were generated using the ExoProfiler package [32] and published ChIP-exo (EBI ArrayExpress E-MTAB-2955) and ChIP-seq (E-MTAB-2956) data for IMR90 cells as input. Peaks were scanned using either the JASPAR MA0113.2 motif [50], the PWM for the combi1 motif [32], the combi2 motif (Fig 5D) or for the AC flank variant, the motif depicted in S9A Fig. Hits were included if the p-value was <10−4. Overlay plots for distinct motifs were generated by aligning the profiles on the GBS and normalizing the signal for each motif variant to 1.
Structural alignment of GR:ETS1 complex. Structural alignment of the GR:ETS1 complex on a combi2 sequence was done as described previously [32] except that both GR dimer halves are retained in the resulting model. In short: A structural model of the DNA hybrid sequence (AGAACATTCCGGCACT) was generated using 3D-Dart [51] using the ETS1 structure (PDB entry 1K79) and the GR structure (PDB entry 3G6U). GR and the ETS2 binding motifs were aligned using the CE-align algorithm [52] to the 3D-DART DNA model of the hybrid sequence.
Data were deposited in ArrayExpress under the accession numbers: E-MTAB-6738 (RNA-seq U2OS-GR) and E-MTAB-6737 (synSTARR-seq U2OS-GR). In addition, we used the previously deposited datasets: E-MTAB-2731 (ChIP-seq U2OS cells), E-MTAB-2955 and E-MTAB-2956 (ChIP-seq and ChIP-exo data IMR90).
|
10.1371/journal.ppat.1004362 | EhCoactosin Stabilizes Actin Filaments in the Protist Parasite Entamoeba histolytica | Entamoeba histolytica is a protist parasite that is the causative agent of amoebiasis, and is a highly motile organism. The motility is essential for its survival and pathogenesis, and a dynamic actin cytoskeleton is required for this process. EhCoactosin, an actin-binding protein of the ADF/cofilin family, participates in actin dynamics, and here we report our studies of this protein using both structural and functional approaches. The X-ray crystal structure of EhCoactosin resembles that of human coactosin-like protein, with major differences in the distribution of surface charges and the orientation of terminal regions. According to in vitro binding assays, full-length EhCoactosin binds both F- and G-actin. Instead of acting to depolymerize or severe F-actin, EhCoactosin directly stabilizes the polymer. When EhCoactosin was visualized in E. histolytica cells using either confocal imaging or total internal reflectance microscopy, it was found to colocalize with F-actin at phagocytic cups. Over-expression of this protein stabilized F-actin and inhibited the phagocytic process. EhCoactosin appears to be an unusual type of coactosin involved in E. histolytica actin dynamics.
| E. histolytica is an important pathogen and a major cause of morbidity and mortality in developing nations. High level of motility and phagocytosis is responsible for the parasite invading different tissues of the host. Phagocytosis and motility depend on highly dynamic actin cytoskeleton of this organism. The mechanisms of actin dynamics is not well understood in E. histolytica. Here we report that coactosin like molecule from E. histolytica, EhCoactosin is involved in F-actin stabilization. The crystal structure obtained for the protein provides explanation for some functional differences observed with respect to the human homologue, such as ability to bind G-actin. Moreover, computational modelling along with crystal structure helps to explain the F-actin binding and stabilization by wild type protein. The mutational analysis further suggests that F-actin binding property does not depend on conserved Lys75 residue as observed in Human coactosin like protein (HCLP) but other regions present in protein are involved in binding. Overexpression of this protein in trophozoites leads to stabilization of actin filaments which are not accessible to actin remodelling machinery thereby reducing the growth of parasite due to decreased rate of actin dependent endocytosis. Overall, EhCoactosin behaves as F-actin stabilizing protein in vitro and it also participates in processes like phagocytosis and pseudopod formation.
| Human amoebiasis is caused by the protist parasite E. histolytica. The parasite is highly motile and displays high level of phagocytic activity in the trophozoite stage. Motility and phagocytosis are essential processes for the survival and invasion of host tissues by the parasite, and largely depends on a highly dynamic actin cytoskeleton. Moreover, there are other processes, such as phagocytosis that also require dynamic actin filament reorganization. Molecular mechanisms that regulate actin dynamics in E. histolytica have not been studied in detail. Preliminary investigations suggest an overall similarity with those described in other eukaryotic cells, but with crucial differences. For example, a number of calcium-sensing calcium-binding proteins appear to directly regulate actin recruitment and dynamics [1], [2], [3]. Several actin-binding proteins are encoded by the E. histolytica genome and many of these proteins are homologs of those that have been studied in other systems. Not many of these amebic actin-binding proteins have been characterized. Understanding structural-functional relationship of these proteins would help to decipher mechanisms of actin dynamics in E. histolytica.
In E. histolytica as well as many other cells, actin dynamics involves both assembly and disassembly of filaments regulated by several actin-binding proteins. The actin-binding protein coactosin was first identified in Dictyostelium discoidedeum and has been classified as a member of actin depolymerising factor (ADF)/cofilin family [4]. The ADF/cofilin family members are expressed in all eukaryotes studied to date. The human coactosin-like protein (HCLP) binds F-actin and interferes with capping of filaments. However it does not affect actin polymerisation [5]. HCLP is also known to bind 5-lipooxygenase [6]. The binding of members of the ADF/cofilin family to the F-actin results in severing and depolymerisation of F-actin [7]. However the precise function of this family may vary from actin nucleation to severing depending on the cellular concentration gradient of cofilin [7].
The E. histolytica genome contains only one copy of the coactosin gene, whose product we refer to as EhCoactosin. Since the role of EhCoactosin in the actin dynamics of E. histolytica has not been previously investigated, we have carried out structural and functional analyses of this protein and present the results here. They show that a single conserved ADF homology domain of EhCoactosin is involved in binding F-actin, and that F-actin is stabilized when EhCoactosin is bound. Moreover, mutation of conserved lysine 75 to alanine does not result in loss of F-actin binding, in contrast to that observed in the case of HCLP, and the binding of this mutant EhCoactosin yields a similar level of F-actin stabilization as does the binding of native EhCoactosin. But deletion of complete F-loop completely abolishes G-actin binding with loss of F-actin stabilization activity, albeit still binds to F-actin. We also propose a mechanism for the binding of EhCoactosin to actin based on a structural model obtained by X-ray crystallography. Overall our results suggest that EhCoactosin displays some features not seen in coactosin from other organisms.
Motility and phagocytosis are important processes for biology of E. histolytica as these are involved in providing nutrition and pathogenesis. It is well known that actin dynamics is key in regulation of above mentioned processes. In E. histolytica not many proteins that regulate actin dynamics have been described. Our group is analysing systematically the E. histolytica homologs of known actin-binding proteins both functionally as well as structurally. In this article we have described E. histolytica homolog of coactosin like protein.
A multiple sequence alignment of EhCoactosin [Acc No XP_650926.1 from the NCBI database] with homologous proteins from different organisms allowed us to identify numerous residues that are conserved in this family of proteins, as well as those unique to EhCoactosin (Figure 1). The amebic Coactosin sequence displays 40% similarity with both human and D. discoidedeum CLPs. Among the conserved residues is a critical lysine at position 75, known to be involved in F-actin binding [8].
The binding of EhCoactosin to F-actin was assessed by a sedimentation assay as described previously (1). The full-length wild-type (WT) protein binds F-actin, as it was found in the pellet fraction after ultracentrifugation (Figure 2A). A similar level of F-actin binding was also observed for truncated versions of EhCoactosin where either the N-terminal seven amino acid residues (EhCoΔN, Figure 2B) or C-terminal 14 residues (EhCoΔC, Figure. 2C) were deleted. In an attempt to narrow down the specific region involved in actin binding, we deleted the F-loop of EhCoactosin (from 71–76 amino acids) and also mutated the critical Lys75 residue. The F-loop deleted version of EhCoactosin (ΔF) retained actin binding property (Figure 2D). The K75A mutant of EhCoactosin was able to bind F-actin, (Figure 2E), which is in contrast to the complete loss of F-actin binding caused by the same mutation in HCLP [8]. These observations suggest that F-actin binding by EhCoactosin does not solely depend on F-loop and Lys75.
G-actin binding was determined by a G-actin sequestering and solid phase assay as described previously [1]. G-actin sequestering assay uses fluorescently labelled G-actin and when a protein binds the labelled actin its florescence decreases mostly in dose dependent manner. The WT EhCoactosin shows G-actin sequestering in dose dependent manner (Figure 2F) while the EhCoΔF has no G-actin binding activity as seen in Figure 2G. We also confirmed G-actin binding for other truncated versions of proteins by this assay. EhCoΔC and EhCoΔN showed dose-dependent G-actin binding but affinity of EhCoΔN was more than EhCoΔC as 10 µM of EhCoΔN was able to sequester same amount of G-actin as 25 µM of EhCoΔC (Figure S1A and B).
EhCoactosin displays specific G-actin binding was also confirmed by binding to a plate coated with G-actin. The level of binding was 2-fold higher than that of the known G-actin-binding protein EhCaBP1 [1] (Figure S1(C)).While EhCoΔC showed a 33% decrease in binding when compared to the WT protein, EhCoΔN exhibited a 2-fold increase in G-actin binding in comparison to WT. The homolog pfADF1, which binds G-actin strongly [9] is positively charged at the N-terminal region compared to EhCoactosin.
The deletion of N-terminal residues in EhCoactosin exposes more positive charges in this region (Figure S2) which, by analogy with pfADF1, may explain the increased affinity of this mutant for G-actin. The F-loop deleted (EhCoΔF) version exhibited complete loss of G-actin binding which was also observed with G-actin sequestering assay.
The role of EhCoactosin in F-actin stabilization was determined by a pyrene-actin assay where fluorescence of pyrene-labelled F-actin decreases upon depolymerisation. The assay showed relative stabilization of F-actin by EhCoactosin compared to that by Xenopus cofilin1 (Xac1) (Figure 3A), and the stabilization effect was confirmed by the ability of EhCoactosin to antagonize the F-actin severing activity of Xac1 (Figure 3B) [10]. That is, while addition of Xac1 led to a sharp decrease in fluorescence, indicating its severing effect on F-actin, in the presence of EhCoactosin no decrease in fluorescence was observed and values were similar to that seen with only actin. The results suggest that EhCoactosin may be protecting F-actin from severing (Figure 3B). We also checked possibility of interaction between Xac1 and EhCoactosin by pull down assay which may lead to similar results. We found that Xac1 and EhCoactosin and its mutants do not interact directly with each other (Figure S3).
EhCoΔC and EhCoΔN showed actin stabilization similar to that of the wild-type protein (Figure 3C and 3E), and a similar stabilization effect was also observed in the case of the K75A mutant (Figure 3G). Moreover, both truncated versions and K75A mutant of EhCoactosin antagonised Xac1-dependent F-actin severing (Figure 3D, 3F and 3H). However, EhCoΔN and Xac1 at 2∶1 ratio did show mild F-actin severing (Figure 3D); the apparent weaker protection conferred by this mutant may be result of its high affinity for G-actin (Figure S1). However, the EhCoΔF had lesser F-actin stabilizing property than WT protein (Figure 3I) as in presence of the protein F-actin depolymerised to an extent. Also EhCoΔF was not able to protect F-actin from Xac1 activity (Figure 3J). These results indicate that F-loop is very essential for stable F and G-actin binding. The deletion of F-loop results in lower affinity towards F-actin making it accessible for Xac1 activity. Hence the whole F-loop plays an essential role in stable binding rather than conserved lysine residue at 75th position.
EhCoactosin consists of a central core of β-sheets surrounded by α-helices. The central core is made up of five strands: β1(26–32), β2(37–44), β3(60–69), and β4(76–85) forming antiparallel strands while β5-strand (113–117) forms parallel strand with β3 and β4. The central β-sheets are flanked on both sides by a total of five helices; α1(9–17) and α3(92–107) are located on the N-terminal side, and α2(48–54), α4(120–122) and α5(125–137) are located on the C-terminal side (Figure 7A and B). This arrangement of secondary structural elements is a common structural feature of proteins belonging to the ADF/cofilin family.
EhCoactosin has a long N-terminal end protruding outside with Ser repeats and this signature Ser repeats is expected to bind G-actin as seen in PfADF1 [9], however wild type EhCoactosin binds to F-actin and EhCoΔN shows higher affinity for G-actin, indicating the “Ser” repeats on the N-terminal are not involved in G-actin binding. The loop connecting strands β3 and β4, which has a conserved lysine at position 75, is called the “F-loop” and it is expected to participate in stabilizing and binding to F-actin [11]. As described in more detail below, the surface of EhCoactosin is highly negatively charged, and this F-loop is part of the negatively charged surface. The N-terminal end and the F-loop are at two opposite sides of the globular structure (Figure 7) suggesting that EhCoactosin binds G-actin and F-actin in very different ways.
Although there is a general similarity of the overall conformation of EhCoactosin with that of related proteins in other organisms, the surface charge distributions of EhCoactosin is markedly distinctive. The surfaces of both sides of EhCoactosin are quite negatively charged, although one surface has overall higher level of negative charge as compared to the other surface. Just a small positively charged surface is found in the α3 and α4 region, as well as is between the β4 and α3 regions, and a hydrophobic pocket is formed between β3 and α5 (Figure 8A and 8A′). In contrast, human coactosin-like protein (HCLP) is positively charged on one side, while negatively charged on the other, which is a characteristic feature of the ADF/cofilin family. The F-loop surface, which is negatively charged on both sides in EhCoactosin, is positively charged on one side and hydrophobic on the other in HCLP (Figure 8B and 8B′).
The surface charge distributions of pfADF1 and pfADF2 also differ from that of EhCoactosin. For pfADF1, one side is highly positively charged and the other has a relatively hydrophobic surface. The N-terminal region of pfADF1 is positively charged relative to that of EhCoactosin [11]. Also, α1 of pfADF1 has three positively charged residues and is relatively long whereas in EhCoactosin it is relatively small and negatively charged [Figure 8C and 8C′]. The surface of pfADF2, while more negatively charged than that of pfADF1, is less negatively charged than that of EhCoactosin (Figure 8D and 8D′). Note that the N-terminal regions of EhCoactosin and PfADF2 were also found to be different; while, as indicated above, the former has Ser repeats, the latter does not [12], [13].
The overall structure of EhCoactosin is quite similar to that of human coactosin-like protein (HCLP), with an RMSD of 1.56 Å and few major differences. The N-terminal regions of the two proteins do deviate by up to 14.7 Å, with that of HCLP bent towards the inside of the structure while in EhCoactosin this N-terminal region is extended. Also, α1 of HCLP is longer by 3 residues compared to that of EhCoactosin (Figure 9A).
The overall structure of EhCoactosin is also fairly similar to the structures of the two types of ADF proteins of Plasmodium falciparum, pfADF1 and pfADF2. Although pfADF1 is functionally different than other ADF/Cofilin proteins, since it binds G-actin [12] and only transiently interacts with F-Actin [12], its overall structure differs from that of EhCoactosin by an RMSD of just 2.0 Å. Certain structural differences are quite notable: The F-loop is absent in pfADF1; β3 and β4 of EhCoactosin, which are extended towards its F-loop, are shorter in pfADF1; and a long C-terminal α-helix present in EhCoactosin is absent in pfADF1. All these observations suggest that the F-loop, β3, β4 and the C-terminal helix of EhCoactosin could be involved in binding to F-actin (Figure 9B). Note also that in pfADF1, the N-terminal end is relatively short, and connected to a short β-sheet, which is a characteristic feature of ADF/cofilin, while in EhCoactosin the N-terminal region is long with characteristic serine repeats, which is thought to participate in G-actin binding. However, both these proteins bind G-actin and it is difficult to suggest a possible mechanism with this data.
The RMSD between pfADF2 and EhCoactosin is 2.13 Å. pfADF2 binds F-actin as well as G-actin [12], and in pfADF2, the F-loop, β3, β4, β5 and β6 are similar to those in EhCoactosin. Moreover, the C-terminal helix, which is missing in pfADF1, is present in pfADF2. This helix is nevertheless longer in EhCoactosin. These regions are likely to be involved in F-actin binding (Figure 9C).
EhCoactosin directly binds F-actin but the mechanism of preventing depolymerisation is not understood. The structural differences of EhCoactosin with Coactosins from other organisms may be responsible for the distinct functional properties. Properties of mutants helped us to model F-actin binding. Here we have sought to analyse the nature of interactions between actin and EhCoactosin by computational modelling. We propose different mode of binding of EhCoactosin to G-actin and F-actin to explain the actin binding properties. Based on the crystal structure of the mouse twinfilin C-terminal ADF homology domain in complex with actin [14] and the recent 9 Å EM model of human Cofilin-2 in complex with actin filaments [15] (Figure S6A and S6B), we built two different models, one for G-actin binding and one for F-actin binding to explain and understand actin binding mechanism of EhCoactosin. EhCoactosin superimposes well with the cofilin of the cofilin-actin complex filaments [15]. In the energy-minimized model, EhCoactosin fits well between the subdomain 1 of the actin monomer and the subdomain 2 of the next actin monomer (Figure 10A). As seen in the model, the N-terminal region of EhCoactosin interacts with subdomain-1 of the actin monomer1 and the C-terminal region of EhCoactosin is placed at the binding interface between two actin molecules (Figure 10B). The α-3 helix forms extensive contacts with subdomain-1 of the actin monomer-1 whereas the F-loop (S69-K75) interacts with the subdomain-2 of the adjacent actin monomer (Figure 10C and D). The C-terminal α-5 helix is docked inside the cavity formed by the two actin molecules. The N-terminal sequence and F-loop region behave like clamps anchoring well within the F-actin structure along the length of the filaments, hence resulting in its stabilization. This explains the effect of EhCoΔF as the mutation of the F-loop results in loss of F-actin stabilization suggesting F-loop is one side of the clamp interacting with F-actin. Thus EhCoΔF can bind F-actin but can't stabilize it. The homology model of the EhCoactosin-F-Actin complex suggests that various regions of the protein, such as the N-terminal sequence, helices α-3 and α-5 and the F-loop play important roles in binding F-actin – and also suggests, in agreement with our mutational studies described above, that no single region or feature of EhCoactosin is indispensible for binding F-actin. Such is the case for EhCoactosin Lys75, for example, despite it being conserved and completely responsible for F-actin binding in other systems; EhCoactosin is unique in this regard. EhCoactosin deletion mutants EhCoΔC as well as EhCoΔN also displayed F-actin binding and stabilization abilities similar to that of the wild type protein.
The model for globular monomeric actin (G-actin) binding to EhCoactosin was obtained using the mouse twinflin ADF homology domain in complex with actin (Figure S4B). Based on the energy minimized model, α3 of EhCoactosin binds the cleft between subdomain 1 & 3 of actin as shown in Figure 11. The modelling data suggest that deletion of the N-terminal region and development of positive charge may loosen interaction with a hydrophobic patch on domain 1 of actin (Figure S2). Due to this, α3 can enter in the groove between domain 1 and 3 of G-actin (see G-actin binding model, Figure 11), helping to explain our result described above that EhCoΔN binds G-actin more strongly than does wild-type EhCoactosin. Interestingly the EhCoΔF abolishes G-actin binding suggesting F-loop deletion might have altered the orientation of α3 and thus loss in G-actin binding.
The protist parasite E. histolytica undergoes extensive pseudopod extension, and displays high level of motility, phagocytosis and macro-pinocytic activities. These processes are crucial for amebic biology as these are associated with food intake and pathogenesis. Since actin dynamics drives all of these processes, we have been investigating many molecules that are known to participate in actin dynamics. Actin-binding proteins, such as those of the ADF/cofilin family, play a major role in actin dynamics. In the current study, we have investigated structural and functional features of the ADF/cofilin protein EhCoactosin. Our results indicate EhCoactosin to be both a G- and F-actin-binding protein, and that it stabilizes F-actin by direct binding. This set of unusual functional feature is due to presence of unique structural motifs not observed in other coactosins or other homologs.
EhCoactosin displays an overall conformational similarity with other ADF/cofilin family members such as HCLP, pfADF1 and pfADF2, yet also displays distinct differences (Figure S7A). Some of the features, such as presence of helices α1 and α3 at the N-terminal region as well as the F-loop, which contains conserved Lys75, are also present in coactosins from other organisms including D. discoideum which are structurally conserved in this family (Figure S7B). Distinctive features of EhCoactosin include a longer N-terminal sequence and a more negatively charged surface. As a result of the latter feature, both sides of the F-loop in EhCoactosin is negatively charged while, for example, one side of the F-loop of HCLP is positively charged while the other side is hydrophobic. The observation that certain features found in EhCoactosin are absent in other coactosins suggests that this molecule in E. histolytica may impart novel functional properties.
Our data clearly show that EhCoactosin is both an F- and G-actin-binding protein in vitro. It is associated with the actin cortex and co-localises with F-actin during pseudopod formation and erythrophagocytosis. The presence of EhCoactosin in phagocytic cups is parallel to the F-actin during the phagocytic cup formation. In vitro functional assays suggest that EhCoactosin is a F-actin stabilizing protein which implies its role in maintaining integrity at the leading edge. Nearly all coactosins studied previously, including human CLP, have not shown a direct effect on actin polymerisation or depolymerisation, although they can interfere with capping of filaments. Chick coactosin is an exception which has been shown to be involved in actin polymerisation downstream of Rac signalling and to promote polymerisation [16]. EhCoactosin is a novel member of the coactosin family with direct effect on F-actin stabilization with F-loop playing important role in binding.
The functional difference between EhCoactosin and other coactosins can be attributed mainly to increased length of the N-terminal part and altered charge distribution. These distinct properties of EhCoactosin are likely to contribute to its binding of G-actin and stabilization of F-actin. Deletion of the N-terminal part EhCoactosin, for example, increases the binding affinity for G-actin on the solid phase. The C-terminal part may also have a role in regulating G-actin binding. When it is deleted affinity for G-actin decreases but not drastically and this is similar to HCLP where C-terminal does not play significant role in F-actin binding [17]. Our in silico analysis suggests that one molecule of EhCoactosin binds to two adjacent actin molecules in the filament. The binding model also suggests that interactions between EhCoactosin and F-actin involve several regions rather than just the F-loop as in other systems. The N-terminal and F-loop of EhCoactosin function as clamps in F-actin binding and decorate the filament along its length. The long serine rich N-terminal region plays a role in F-actin binding whereas deletion of which results in F-actin severing activity. The model and solid phase data suggest that this may be due to high affinity for G-actin displayed by the mutant as a result of uninhibited binding of α3 between subdomain 1 and 3. Although the Lys75 residue is needed by HCLP for binding F-actin, it is not required in case of EhCoactosin since the mutant K75A protein has similar experimentally determined F-actin-binding and other properties as does the wild-type protein. Computational modelling also supports these results as K75A mutant does not show any significant change in binding of EhCoactosin to actin, as K75 is not directly interacting with F-actin. This implies that binding of EhCoactosin and actin involves interactions other than F-loop and Lys75 residue unlike other homologs. But complete deletion of F-loop results in loss of F-actin stabilization suggesting F-loop is one side of the clamp interacting with F-actin. Thus EhCoΔF can bind F-actin but can't stabilize it.
Our experiments have shown that EhCoactosin stabilises F-actin, but we also need to understand the underlying contributions to actin dynamics in E. histolytica since both depolymerization as well as stability of F-actin are required for critical cellular processes. Many drugs that stabilize F-actin have deleterious effect on processes that require actin dynamics [18], [19], and over-expression of EhCoactosin in E. histolytica yields cells that display impaired growth and phagocytosis, presumably due to the protein's stabilization of F-actin. This consequence of overexpression is not seen with other coactosins and appears to be a unique property of the E. histolytica protein. E. histolytica is an early branching eukaryote displaying unique biology, and although it shares many of the participants of the cytoskeleton remodelling machinery with metazoan organisms, it also uses a few novel proteins in regulating the actin cytoskeleton [1], [2], [3]. The calcium-binding proteins EhCaBP1 and EhCaBP3 are such examples, and they have been shown to be involved in actin dynamics and phagocytic cup formation [2], [3], [20]. All these studies including present study show that E. histolytica proteins can also undergo functional diversification in order to fulfil its needs, high rate of actin dynamics. The detailed study of this binding protein will lead to better understanding of the cytoskeletal remodelling in this parasite and also as well evolution of this process in other eukaryotes.
The erythrophagocytosis results indicate in vitro concentration of EhCoactosin above critical level may affect actin remodelling. Phagocytosis involves both actin polymerisation and depolymerisation which is mediated by several actin-binding proteins. The high levels of EhCoactosin in cell may promote excess stability of F-actin in vitro by preventing access of actin remodelling protein to F-actin required during the phagocytosis. Taken together this leads to increased rigidity in actin cytoskeleton which impairs its dynamic remodelling required for processes like motility and phagocytosis.
In conclusion, EhCoactosin is directly involved in F-actin stabilization, which has not been reported earlier. In vivo EhCoactosin may actively contribute to the maintenance of F-actin during erythrophagocytosis and pseudopod formation. The interactions between EhCoactosin and F-actin depend on several regions in the protein rather than specific residues such as Lys75. The evolutionary basis of development of specific interaction in higher organisms can be understood by studying primitive eukaryotes like E. histolytica. This study will also lead to better understanding of actin dynamics in this organism and as well as evolution of actin dynamics as a process in organisms.
The coding sequence of coactosin gene (GenBank accession no. XP_650926) was amplified by PCR from genomic DNA of Entamoeba histolytica strain HM1:IMSS using the forward primer 5′-CCGCCATGGCAATGTCTGGATTTGATCTTAG-3′ and the reverse primer 5′-CCGCTCGAGCTTAATTTTAGCAGCGATTTC-3′. The EhCoactosin gene was cloned in pET28b (Novagen) between Nco1 and Xho1 sites with a C-terminal 6× His tag. Four constructs were prepared for biochemical experiments: wild-type EhCoactosin (EhCoWT); an EhCoactosin in which 14 amino acid residues were deleted from the C-terminus because it was predicted to form a loop (EhCoΔC); another for which 7 residues were deleted from the N terminus (EhCoΔN), F-loop spanning from 71–76 amino acid was also deleted (EhCoΔF) and a single site substitution mutant (K75A). The cloning was confirmed by restriction digestion by Nco1 and Xho1 followed by DNA sequencing. The CAT gene of the shuttle vector pEhHYG-tetR-O-CAT (TOC) was excised using KpnI and BamHI and the EhCoactosin gene was inserted in its place in either the sense or the antisense orientation. The expression in this vector was tetracycline inducible and expressed sense (S) and antisense (AS) RNA of the gene in E. histolytica trophozoites. For the study of co-localization in E. histolytica cells we carried out HA tagging at the N-terminus of EhCoactosin. The Forward Primer 5′-CGGGGTACCATGTATCC ATATGATGTTC CAGATTATGCTATGTCTGGATTTG-3′ and the reverse primer 5′- GCGGGATCCTTAAGCATAATCTGGAACATCATATGGATAATT TGAGGTGG-3′ were used for HA tagging.
The recombinant plasmid containing the EhCoactosin gene was transformed into E. coli BL21 (DE3) cells (Novagen). Primary culture was grown overnight in 50 ml LB media from the single colony of transformed BL21 cells supplemented with 50 µg/ml Kanamycin at 37°C. Secondary culture was grown by inoculating 1% of primary culture in the same media at 37°C until the OD600 reached 1.0. The culture was induced with 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) (Sigma) and allowed to grow for another 4 hrs at the same temperature. Cells were harvested by centrifugation at 6000 rpm for 10 minutes at 4°C. These cells were stored at −80°C until further processing.
The harvested cells were resuspended and homogenized in resuspension buffer containing 50 mM Tris HCl (pH 8.0), 0.1 mM EDTA and 0.1 mM DTT. Resuspended cells were lysed with 3 cycles of flash-freezing in liquid nitrogen and subsequent thawing in water-bath at 37°C. The lysate was subjected to 5–6 cycles of sonication on ice at 25% amplitude with each pulse of 30 sec and 1 min interval. The sonicated cell lysate was centrifuged at 13,000 rpm for 30 minutes at 4°C. Supernatant was filtered with Whatman filter paper no. 1 and clear lysate was passed through a Nickel-NTA column (GE healthcare) pre-equilibrated with resuspension buffer. Thereafter, the column was washed with 2 bed volumes of buffer containing 50 mM Tris HCl (pH 8.0), 0.1 mM EDTA, 0.1 mM DTT and 10 mM imidazole. The bound protein was eluted with buffer comprising 50 mM Tris-HCl (pH 8.0), 0.1 mM EDTA, 0.1 mM DTT and 100 mM imidazole. The purified fractions of protein were concentrated using Centricon filters (Millipore) and subjected to gel filtration chromatography on HiLoad Superdex 75G 16/60 column (GE Healthcare) pre-equilibrated with buffer containing 50 mM Tris-HCl (pH 8.0), 0.5 mM EDTA, 0.5 mM DTT and 1 mM sodium azide. Homogeneity of protein was assessed on 12% SDS-PAGE (Figure S8). Peak fractions were concentrated using Centricon filters (Millipore) and concentration was estimated with A280.
Selenomethionine-labelled EhCoactosin was purified under reducing conditions using specific media (by Molecular Dimensions, United Kingdom). The concentration of selenomethionine was maintained at about 25 mg/litre. Initially, the primary culture was grown in LB medium overnight. Cells were then harvested by centrifuging at 4000 rpm for 6 min. Harvested cells were resuspended in the complete selenomethionine media, and washed once with same media to completely remove any leftover LB medium. Secondary culture was grown by inoculating 1% of primary culture in the same media at 37°C until the OD600 reached 1.0. Culture was allowed to grow at 37°C for about 4 hrs after inoculation until OD600 reached 1.0. Cells were induced with 1 mM IPTG and allowed to grow for another 4 hrs at same temperature. Cells were harvested at 6,500 rpm for 6 min and stored at −80°C for further processing. Subsequent processing and purification were done by the same method used for native EhCoactosin.
G-actin was purified from rabbit skeletal muscle acetone powder [21]. Further Actin was labelled with N-(1-pyrene) iodoacetamide (P-29, Molecular Probes) by the protocol described previously [22] for performing the pyrene-actin assay.
Native EhCoactosin protein was crystallized using the hanging drop vapor diffusion method in 24-well Linbro plates against a reservoir solution containing 25–35% PEG 1500, 100 mM sodium acetate, 0.2 mM CaCl2, 10 mM MgCl2 and 100 mM HEPES, pH 7.3–7.7. Two µl of [∼75 mg/ml] protein and 2 µl of reservoir solution were mixed and allowed to equilibrate at 16°C. The crystals that formed in these drops were flash frozen in a cryoprotectant solution containing additional 5% PEG 400 mixed with mother liquor. Selenomethionine-labelled protein was prepared and crystallized using similar conditions. The crystal appeared in condition containing 28–33% PEG 3350, 100 mM sodium acetate, 0.2 mM CaCl2, 10 mM MgCl2, 5% isopropanol and 100 mM HEPES pH 7.4–7.7 (Figure S9). The crystals were flash frozen in the same cryo-protectant. The X-ray data for selenomethionine-substituted crystals were collected at the BM14 synchrotron beamline, ESRF, Grenoble, France at a selenium peak wavelength of 0.97860 Å. Data sets were indexed and scaled using HKL2000 [23].
Anomalous data collected for Se-Met labelled EhCoactosin crystals were used to calculate FA values using the program SHELXC [24]. Each of the two heavy atoms expected were found using the program SHELXD [24]. Initial phases were calculated after density modification using SHELXE [25]. The reflection file was further used in the Autobuild program [25] of the Phenix suite [26] for automated model building. Then missing residues were traced into the electron density and refined by iterative model building using the COOT graphics package combined with REFMAC5 [27]. HEPES, Na, and water molecules were added by COOT guided by Fo-Fc electron density >3σ. The final model was validated by the Procheck [28] program of the CCP4 suite. Structure factors and co-ordinates have been validated and deposited in the Protein Data Bank with accession id 4LIZ. Data statistics are listed in Table 1.
A model of the F-actin–EhCoactosin complex was built using the 9 Å electron microscopy derived model of F-actin ADF/cofilin protein complex [16]. The crystal structure of EhCoactosin was then superimposed onto the human ADF/cofilin molecule from the EM model using the RAPIDO server [29]. The ADF/cofilin molecule used showed an extended N-terminal region which did not superimpose well. The overall structure (119 atoms), however, did superimpose well with an RMSD of 1.15 Å (Figure S4A). The final model of the complex with five actin and six coactosin molecules was then subjected to energy minimization with 2500 cycles of steepest descent and followed by 2500 cycles of steepest descent algorithm using AMBER molecular dynamics package [30]. Similarly the model of the G-actin-EhCoactosin complex was obtained using the crystal structure of mouse twinfilin C-terminal ADF homology domain in complex with actin [14]. The root mean square deviation obtained was 2.17 Å (Figure S4B). The electrostatic surface charge distribution was calculated using the ABPS plugin in PyMOL. The negative electrostatic surface is shown in red, and the positive surface in shown in blue; all surfaces are drawn at 3 e/kBT. The images were prepared using Pymol software [31].
E. histolytica strain HM-1: IMSS and all transformed parasites were maintained and grown in TYI-S-33 medium [1] containing 125 ml of 250 U ml−1 benzyl penicillin and 0.25 mg ml−1 streptomycin per 100 ml of medium. The transformants containing tetracycline inducible system were grown in presence of 10 µg ml−1 of Hygromycin B. The cells were first grown for 48 h (60–70% confluent) and then 20 µg ml−1 tetracycline was added to the medium for 36 h for induction. Cells carrying constructs with constitutive expression system (such as GFP) were maintained at 10 µg ml−1 of G418. But the experiments were carried out in presence of 30 µg ml−1 of G418.
Transfection was performed by electroporation. Briefly, trophozoites in log phase were harvested and washed with phosphate buffer saline (PBS), followed by incomplete cytomix buffer (10 mM K2HPO4/KH2PO4 (pH 7.6), 120 mM KCl, 0.15 mM CaCl2, 25 mM HEPES (pH 7.4), 2 mM EGTA, 5 mM MgCl2]. The washed cells were then re-suspended in 0.8 ml of complete cytomix buffer (incomplete cytomix containing 4 mM adenosine triphosphate, 10 mM glutathione) containing 200 mg of plasmid DNA and subjected to two consecutive pulses of 3000 V/cm (1.2 kV) at 25 mF (Bio-Rad, electroporator). The transfectants were initially allowed to grow without any selection. Drug selection was initiated after 2 days of transfection in the presence of 10 µg ml−1 G-418 for constructs with GFP or 10 µg ml−1 of hygromycin B was used for tetracycline inducible constructs.
Immunofluorescence staining was carried out as described previously [1]. Briefly E. histolytica cells resuspended in TYI-33 medium were transferred onto acetone-cleaned coverslips placed in a petri dish and allowed to adhere for 10 min at 35.5°C. The culture medium was removed and cells were fixed with 3.7% pre-warmed paraformaldehyde (PFA) for 30 min. After fixation, the cells were permeabilized with 0.1% Triton X-100/PBS for 1 min. This step was omitted for non-permeabilized cells. The fixed cells were then washed with PBS and quenched for 30 min in PBS containing 50 mM NH4Cl. The coverslips were blocked with 1% BSA/PBS for 30 min, followed by incubation with primary antibody at 37°C for 1 h. The cover slips were washed with PBS followed by 1% BSA/PBS before incubation with secondary antibody of 30 min at 37°C. Antibody dilutions used were: Anti- EhCoactosin at 1∶200, anti-HA at 1∶50, TRITC-Phalloidin at 1∶250 and anti-rabbit Alexa 488 (Molecular Probes) at 1∶300. The preparations were further washed with PBS and mounted on a glass slide using DABCO (1,4-diazbicyclo [2,2,2] octane (Sigma) 10 mg/ml in 80% glycerol). The edges of the coverslip were sealed with nail-paint to avoid drying. Confocal images were visualized using an Olympus FluoView FV1000 laser scanning microscope.
17.5 µM G-actin with 10% pyrene labeled was polymerized for one and half hour at 25°C in F-buffer (10 mM Tris-Cl pH8.0, 0.2 mM DTT, 0.7 mM ATP, 50 mM KCl, 2 mM MgCl2). Depolymerization kinetics was started with the addition of 2 µL of preassembled actin with 58 µL of F-buffer, and the volume was made up to 70 µl with HEKG5 or protein solution. N-pyrene fluorescence was monitored with excitation at 365 nm and emission was measured at 407 nM for 600 seconds (QM 40 PTI NJ). The de-polymerizing protein Xenopus cofilin1 (Xac1) was used as a positive control [10].
The solid-phase assay experiments were performed to monitor the binding of wt- and mutant EhCoactosin proteins to G-actin. The wells of the ELISA plate were coated with 5 µM G-actin in PBS buffer and incubated for 12 h at 4°C. The wells were washed with PBS-T buffer. 5 µM protein was added to the wells in duplicates. Bound protein was detected with anti-EhCoactosin antibody followed by HRPO-lined anti-rabbit IgG using the colorimetric substrate TMB (Sigma). The reaction was stopped with 2N H2SO4 and absorbance was monitored at 405 nm with ELISA plate reader (Bio-Rad, USA).
5 µM of rabbit muscle actin was polymerized for one and half hour at 25°C in F-buffer (10 mM Tris-Cl pH8.0, 0.2 mM DTT, 0.7 mM ATP, 50 mM KCl, 2 mM MgCl2). After polymerization, actin was mixed with appropriate target protein (5 µM) in a total volume of 150 µl and incubated for 30 min at RT. The samples were centrifuged at 100,000 g for 45 min at 4°C. The supernatant and pellet fractions (total) were analyzed by 12% SDS-PAGE followed by Coomassie blue staining. All target proteins were ultracentrifuged at 1,00,000× g for 1 h and the supernatant was used for the assays in order to avoid aggregates.
The human erythrocytes used in the experiments were collected from Somlata. The blood was taken by piercing the ring finger by sterile needle and transferred into a sterile tube containing PBS. 107 red blood cells (RBC) were washed with PBS and incomplete TYI-33 and were incubated with 105 amoeba for varying time periods at 37°C in 0.5 ml culture medium. The amoebae and erythrocytes were pelleted down, non-engulfed RBCs were bursted with cold distilled water and recentrifuged at 1000 g for 2 min. This step was repeated twice, followed by resuspension in 1 ml formic acid to burst amoebae containing engulfed RBCs. The absorbance was measured at 400 nm.
The human erythrocytes used in the experiments were collected from Somlata. The blood was taken by piercing the ring finger by sterile needle and transferred into a sterile tube containing PBS. The consent letter was obtained from the individual for taking blood sample before carrying out the experimental studies.
|
10.1371/journal.pgen.1000424 | An African Ancestry-Specific Allele of CTLA4 Confers Protection against Rheumatoid Arthritis in African Americans | Cytotoxic T-lymphocyte associated protein 4 (CTLA4) is a negative regulator of T-cell proliferation. Polymorphisms in CTLA4 have been inconsistently associated with susceptibility to rheumatoid arthritis (RA) in populations of European ancestry but have not been examined in African Americans. The prevalence of RA in most populations of European and Asian ancestry is ∼1.0%; RA is purportedly less common in black Africans, with little known about its prevalence in African Americans. We sought to determine if CTLA4 polymorphisms are associated with RA in African Americans. We performed a 2-stage analysis of 12 haplotype tagging single nucleotide polymorphisms (SNPs) across CTLA4 in a total of 505 African American RA patients and 712 African American controls using Illumina and TaqMan platforms. The minor allele (G) of the rs231778 SNP was 0.054 in RA patients, compared to 0.209 in controls (4.462×10−26, Fisher's exact). The presence of the G allele was associated with a substantially reduced odds ratio (OR) of having RA (AG+GG genotypes vs. AA genotype, OR 0.19, 95% CI: 0.13–0.26, p = 2.4×10−28, Fisher's exact), suggesting a protective effect. This SNP is polymorphic in the African population (minor allele frequency [MAF] 0.09 in the Yoruba population), but is very rare in other groups (MAF = 0.002 in 530 Caucasians genotyped for this study). Markers associated with RA in populations of European ancestry (rs3087243 [+60C/T] and rs231775 [+49A/G]) were not replicated in African Americans. We found no confounding of association for rs231778 after stratifying for the HLA-DRB1 shared epitope, presence of anti-cyclic citrullinated peptide antibody, or degree of admixture from the European population. An African ancestry-specific genetic variant of CTLA4 appears to be associated with protection from RA in African Americans. This finding may explain, in part, the relatively low prevalence of RA in black African populations.
| Rheumatoid arthritis (RA) is a systemic autoimmune condition affecting the synovial membranes of diarthrodial joints. The etiology of RA is unclear but is thought to result from an environmental trigger in the context of genetic predisposition. We report that a single nucleotide polymorphism (SNP) (rs231778) in CTLA4, which encodes a negative regulator of T cell activation, is associated (p = 2.4×10−28) with protection from developing RA among African Americans. rs231778 is only polymorphic in populations of African ancestry. Protective alleles such as this one may contribute to the purported lower prevalence of RA in African Americans. Our finding appears to be independent from confounding by linkage with the HLA-DRB1 shared epitope or by genetic admixture. Furthermore, we did not replicate associations of CTLA4 SNPs with RA or other autoimmune diseases previously reported in Asians and Caucasians, such as rs3087243 (+60C/T) and rs231775 (+49A/G). The associations of different SNPs with RA susceptibility specific to different populations highlight the importance of CTLA4 in the pathogenesis of RA and demonstrate the ethnic-specific genetic background that contributes to its susceptibility.
| Cytotoxic T-lymphocyte associated protein 4 (CTLA4, CD152) is a negative regulator of T-cell activation. As the T-cell activation signal propagates due to costimulatory B7 molecule (CD80, CD86) binding of CD28, cell surface expression of CTLA4 increases to compete with CD28 [1]. CTLA4 also prevents further clonal expansion of effector T-cells, including regulatory T cells (Treg) [2],[3], and can inhibit osteoclast formation [4].
Genetic variation in CTLA4 (Chromosome 2q33) could contribute to unchecked T cell or osteoclast activation with resultant onset of autoimmune disease such as rheumatoid arthritis (RA). CTLA4 was modestly associated with RA in a recent genome wide association study (GWAS) of RA in Caucasians [5]. CTLA4 single nucleotide polymorphisms (SNP), such as rs231775 (+49A/G), have been associated with multiple autoimmune conditions including RA, Addison's disease, autoimmune pancreatitis [6], autoimmune thyroid disease, celiac disease, chronic inflammatory arthritis [7]. multiple sclerosis [8], type I diabetes mellitus, Sjögren's syndrome [9], and systemic lupus erythematosus (SLE) [10]. An association with another SNP, rs3087243 (+60C/T), and RA was found in a Chinese Han population [11]; however, these results were not replicated in Irish [7], United States Caucasian [12], or, when corrected for multiple testing, British Caucasian [13] populations. Analysis of a much larger group of Caucasians from North America and Sweden associated this marker with RA [particularly with the anti-cyclic citrullinated peptide (anti-CCP) antibody positive RA subset] [14]. Given the association of CTLA4 with multiple diseases in various populations, we sought to characterize the genetic contribution of CTLA4 to RA in African Americans – a population not yet explored.
RA is purported to be less prevalent in African Americans than in Caucasians based on clinical observation and data in black continental Africans [15]–[19]. African-specific protective alleles might explain the lower disease prevalence among persons of African ancestry and should be evaluated in genetic studies with this population.
In this study, we genotyped CTLA4 haplotype tagging SNPs (htSNPs) in two groups totaling 505 African American patients with RA and 712 African American healthy controls. We found and replicated a novel protective association at an ethnic-specific intronic SNP, rs231778, in both independent groups. While this SNP is polymorphic only in the HapMap Yoruba population, we confirmed a lack of variation by genotyping 530 Caucasians. Importantly, we did not detect significant confounding for association of rs231778 when our patients were stratified by level of European admixture or by RA subclassification such as presence of the HLA-DRB1 shared epitope (SE) or anti-cyclic citrullinated peptide (anti-CCP) antibodies [20]. We also did not find association with two SNPs (rs3087243 and rs231775) previously reported to have disease associations with RA in European ancestry populations or with other autoimmune diseases.
Our data reveal a protective African ancestry-specific allele that may contribute to the purportedly lower prevalence of RA in persons of African ancestry and provide suggestions for future research into the relationship between T cell regulation and RA pathogenesis.
The Consortium for the Longitudinal Evaluation of African Americans with Early Rheumatoid Arthritis (CLEAR) Registry enrolled self-identified African Americans with RA who met the American College of Rheumatology (ACR) 1987 diagnostic criteria [21]. Participants for CLEAR were recruited from the University of Alabama at Birmingham (UAB) [coordinating center]; Emory University/Grady Hospital (Atlanta, GA); University of North Carolina at Chapel Hill; Medical University of South Carolina (Charleston, SC); and Washington University (St. Louis, MO). Recruitment occurred in two phases: enrollment of patients with early RA (<2 year disease duration) followed longitudinally until 5 years disease duration, from 2000 to 2007 (CLEAR I); and enrollment of patients with RA of any duration from the same sites as part of a cross-sectional analysis from 2007 to present (CLEAR II). Comprehensive demographic, clinical, and radiographic data are being collected on all CLEAR participants, and serum and DNA samples are being stored [22]. These data allow for stratification of RA patients [20] by presence of the HLA-DRB1 SE and anti-CCP antibody positivity. We have also measured estimated global admixture using a panel of ancestry informative markers (AIMs), as previously reported [23]. A group of healthy African American controls, for the longitudinal arm of this study, with similar sex, age, and geographic location has been recruited, as previously described [23]. All participants were recruited with informed consent under the approval of each respective Institutional Review Board. Genomic DNA was isolated using standard methods and stored at −70°C.
This study included 282 African American RA patients and 149 African American controls from the CLEAR longitudinal study (CLEAR I) and 223 African American RA patients from the CLEAR cross-sectional study (CLEAR II). We also obtained DNA samples from an additional 563 healthy African Americans from Alabama recruited for a case-control study of SLE [24] to use as controls for the CLEAR II RA patients. Demographics for CLEAR I and CLEAR II RA patients are presented in Table 1. Controls were younger than the RA patients (mean age: CLEAR I = 45±14 years, CLEAR II = 35±11 years). Similar to the patient groups, both of the control sets were predominantly female (percent female: CLEAR I = 82%, CLEAR II = 74%). In total, we analyzed 505 African American RA patients and 712 African American controls. We used RA patients and controls from CLEAR I as an initial test set and RA patients from the CLEAR II and additional Alabama controls as a replication group.
All SNPs within the CTLA4 region (±2 kb) that have a minor allele frequency (MAF) ≥0.05 in the Yoruba HapMap population (Phase II/Release 21) were genotyped: rs231775, rs231776, rs231777, rs231778, rs231779, and rs3087243. Data from the resequencing of CTLA4 in both African and European populations contracted to SeattleSNP (Dr. Debbie Nickerson, University of Washington) were kindly provided from the Population Genetics Study coordinated at UAB (Drs. Richard Kaslow and Robert Kimberly). CTLA4 SNPs detected by SeattleSNP that capture information on polymorphisms not present in HapMap for Africans with a MAF ≥0.05 were additionally genotyped: rs11571319, rs231772, rs231780, rs34031880, rs733618, and *5251. *5251 is not yet listed in dbSNP: its physical location is 54945227 in NCBI contig file NT_005403, and its surrounding sequence is ATGGTAGCCTTGCTTATTGT [G/T] GGTGGCAACCTTAATAGCAT.
Genotyping was performed by the Illumina FastTrack GoldenGate BeadXpress genotyping service (San Diego, CA) for CLEAR I for SNPs from the International HapMap Consortium. All other genotyping was performed using Applied Biosystems TaqMan Allelic Discrimination Assays (Foster City, CA) on an ABI 7900HT Genetic Analyzer. Overall, between both platforms for all SNPs, our genotyping success rate was 99.4%. We successfully genotyped rs231778 among 74 samples using both platforms with 100% reproducibility. To confirm the monomorphic nature of rs231778, we genotyped this SNP in 530 Caucasian samples from the UAB Treatment of Early Aggressive Rheumatoid Arthritis (TEAR) study.
Fisher's exact tests were performed on SAS 9.0 (Cary, NC) and exact logistic regression tests performed on LogXact 8.0 (Cambridge, MA). We controlled for potential confounding by HLA-DRB1 status, anti-CCP antibody positivity, and genetic admixture following the approach of Redden et al. [25]. Linkage Disequilibrium and haplotype analyses were performed with HaploView v3.31 [26]. All SNPs were in Hardy-Weinberg Equilibrium (tested with Chi squared tests), except rs231776 (HWE p = 0.0085), which was excluded from further analysis.
Data available from the International Haplotype Mapping Consortium (HapMap), as accessed in February 2008, appear incomplete with regard to coverage of CTLA4. Only SNPs present from the 5′ region through intron 1 (rs231775, rs231776, rs231777, rs231778, rs231779) are represented with detailed genotyping data. HapMap does not provide data for SNPs among the remaining exons and introns of CTLA4 but does present information for polymorphisms in the 3′ end of the gene, such as rs3807243. To select htSNPs that cover the remaining interior portions of this gene, we accessed resequencing data available from SeattleSNP that provided detailed genotypes on YRI and CEU populations. SeattleSNP routinely resequences only 500 basepairs into each end of a given intron. A portion of intron 1 (the longest intron) is the only region of CTLA4 not completely resequenced by SeattleSNP; however, intron 1 was completely covered by HapMap, allowing the combination of these two resources to provide the most detailed haplotype tagging strategy for this gene. See Figure 1.
Due to the limited public information on CTLA4 in African Americans, we used HaploView to calculate linkage disequilibrium (LD) across all genotyped SNPs. A plot representing the LD (r2 values) of SNPs is included as Figure 2.
We detected a protective effect for RA in African Americans with the G allele of rs231778 in both CLEAR study groups (longitudinal and cross-sectional) independently and together (CLEAR I and CLEAR II combined Fisher's exact p = 4.46×10−26). See Table 2. Because homozygotes for the G allele were rare, we compared the frequency of persons with genotypes GG and AG to those with genotype AA. From the odds ratios of the two groups combined, it can be seen that the presence of the G allele confers a protective effect (OR = 0.19, 95% CI: 0.13–0.26, p = 2.4×10−28, Fisher's exact). See Table 3. rs231778 is not in LD with any other SNP, which suggests any genetic effect it confers is likely independent. See Figure 2.
The G allele of rs231778 is relatively specific for African populations as only the A allele is detected among Asians and Caucasians genotyped in the International HapMap Project and in Caucasians genotyped by SeattleSNP. Since variation at rs231778 was not found in the HapMap (n = 24) or Perlegen (n = 60) based European samples, we genotyped an additional 530 self-identified Caucasians to assess ethnic specific variation at this site. Among these 530 subjects, only 3 were heterozygous at rs231778, and none were homozygous for the G allele, which yields a MAF of 0.0028. In the 697 healthy African American individuals we genotyped, the MAF is 0.209 illustrating the ethnic specificity of this marker.
Since the presence of the SE has been associated with susceptibility to RA in our population [23] and known to confound association with RA at other immunologically relevant loci such as PTPN22 [27], we evaluated our findings in CTLA4 for possible confounding by the HLA-DRB1 SE, the strongest known genetic risk factor for RA. We found that the MAF of rs231778 was not different within cases or controls when stratified for number of SE alleles present. Because only 4 control samples have two SE alleles, we cannot rule out any possible influence of the SE on the genetic contribution of CTLA4 in RA susceptibility, but it appears to be unlikely. See Table 4.
Since our study focuses on African Americans, a group with known recent population admixture [28], we assessed percentage of European admixture as a confounding factor in the association of RA with rs231778. Data from a genome-wide admixture panel performed at the Broad Institute from our previously reported work [23] allowed calculations of global admixture estimates (percent European ancestry) for 282 cases and 94 controls (total N = 366). Of these 366 with admixture data, there was successful genotyping for the CTLA4-containing region of Chromosome 2 in 266 cases and 81 control samples (total N = 347). We show the mean percentage of European ancestry segregated by genotype for cases and controls in Table 5. The degree of admixture was not associated with rs231778 genotype (Fisher's exact p = 0.2367). We confirmed that admixture difference between cases and controls was not significant using the robust Welch test, which produced a value of 2.308 (degrees of freedom = 215.578, p = 0.130).
We did not find a significant association with RA of the G allele among the 347 samples with complete admixture data and CTLA4 genotypes (asymptotic p = 0.0674); we suspect that this is due to the reduced statistical power of analysis of a smaller number of subjects and controls. When we based calculations upon the 366 samples used in our previous admixture-based manuscript [23], this small increase in sample size regained statistical significance of association with RA (asymptotic p = 0.0183). To illustrate further the lack of significance among the 347 samples is due to lack of power, frequency counts of genotypes among the 366 samples are incorporated in Table 5 to demonstrate a similar pattern of genotype distributions with and without these additional samples.
Nonsynonymous SNPs previously associated in other populations and autoimmune phenotypes (rs3087243 and rs231775) were not associated with RA in our study. See Table 6. We also found no association when we analyzed data based upon deduced haplotypes or at any individual SNP when stratified by RA subclassification (SE status, anti-CCP antibody status, or percent European ancestry) as has been observed with RA associations at other sites in the genome [23] and with CTLA4 SNPs in Caucasian populations [14]. See Table 6.
We found no significant association with the G allele of rs3087243, even when stratified for presence of anti-CCP antibody, as previously reported in Europeans with RA [14]; the distribution of genotypes and allele frequencies of this SNP were similar in anti-CCP antibody-positive and anti-CCP antibody-negative RA patients. Similarly, we found no significant differences in allele frequency between anti-CCP positive RA patients and anti-CCP negative RA patients at the SNP associated with RA in our study (rs231778).
In the initial analysis of the CLEAR longitudinal arm, we found a protective effect (lower allele frequency in patients than controls) of the minor allele (G) of rs231780 allele (Fisher's Exact p = 0.0123). However, upon replication in the CLEAR cross-sectional arm, this difference in MAF between cases and controls was not significant in the cross-sectional arm or in both arms combined [Fisher's exact p = 0.0667; MAF 0.097 in patients, 0.124 in controls]. See Table 6. Of note, the rs231780 SNP appears to be African-ancestry specific as well, with a MAF of 0.17 in Africans and ∼0.00 in Europeans among HapMap subjects. It is possible that our lack of association at this marker is due to a true negative state or due to lack of power for detecting a positive association, as our the p value is bordering on significance (p = 0.07). Although rs231780 is also an ethnic-specific SNP, there is not significant LD between it and the strongly associated SNP rs231778 (r2 = 0.107, D′ = 0.445). The lack of association of the African-specific SNP, rs231780, with RA might be sufficient to rule out genetic admixture as the cause of the association at rs231778.
We also found an association with rs231776 (Fisher's exact p = 0.0418) when both study groups were combined; however, this SNP was not in Hardy-Weinberg equilibrium (HWE p = 0.0085), complicating interpretation of these results.
We detected a significant novel genetic association with RA in African Americans at the CTLA4 SNP rs231778. In this case-control study, African Americans with at least one minor (G) allele were 0.19 times as likely to have RA as those without a minor allele (95% CI 0.13 0.26, Fisher's Exact p = 2.437×10−28). This P value does not appear to be subject to the inaccuracy introduced by cancellation error by complementation [29]. Our study is limited in sample size due to its exclusive focus on a minority population, which may introduce influence by bias in sample collection, genotyping errors, and lack of power. However, due to our efforts in matching patients and controls and validating our genotyping results (100% reproducibility in 74 samples on different genotyping platforms), we believe such biases have been minimized. We believe that our study is sufficiently powered to detect associations as we found a statistically significant result in two separate arms of the study.
The associated SNP, rs231778, is located in intron 1 and is not in LD with any genotyped SNP in CTLA4 such as the disease-associated rs3087243 and rs231775 markers. See Figures 1 and 2. It is possible, however, that LD could span farther than assessed in this study allowing the possibility that rs231778 is a surrogate marker for another associated polymorphism well outside of the gene boundaries of CTLA4. LD has been shown to span several megabases in African Americans, which supports this possibility [30]. Additional genotyping of 5–10 AIMs in this chromosomal region in a large number of African Americans may allow a better understanding of the long-range haplotype structure. Our study did include five African-specific SNPs (rs231772, rs231776, rs231780, rs34031880, 5251*) and one AIM, defined as a difference in MAF >0.20 between populations, (rs3087243) that did not associate with RA. The association of the African-specific allele of rs231778 and RA and the lack of association at these ethnic-specific markers supports the idea that the association of rs231778 is independent from bias by genetic admixture.
Interestingly, rs231778 is monomorphic in both Asian and Caucasian populations, according to genotype data from HapMap and SeattleSNP, and virtually absent in our genotyping of 530 Caucasians. Given the ethnic-specific status of this SNP, it is possible that our finding helps to explain the purported, but as yet unproven, observation of a lower prevalence of RA in African Americans compared to Caucasians. We would anticipate the association of such African-specific protective alleles with resistance to RA. Racial or ethnic differences have now been suggested in the association of RA with several genes, including PTPN22 [31], PADI4 [32], SLC22A4 and RUNX1 [33], and in CTLA4, particularly between Asian and Caucasian populations [10],[34]. These data highlight the need for additional research into the genetic background of RA in various populations such as African Americans to uncover additional ethnic-specific associations.
Our study included 697 healthy African American controls that possessed a MAF of 0.209 at rs231778. This finding is surprising since public resources such as the International HapMap Consortium has a MAF of 0.09 in their panel of 60 Yorubans and 0.00 in 60 Caucasians. African Americans are considered to be an admixed population with an African background and contribution of approximately 20% European genetic ancestry. In fact, we calculated that European ancestry contributes 15±5% of the genetic composition of African Americans in the CLEAR study. Therefore, we would expect a MAF for rs231778 to be between 0.00 and 0.09. Given that our participants were collected at multiple centers across the Southeastern United States (with each center having similar MAFs), that we genotyped 74 samples with 100% reproducibility on dual platforms (TaqMan and Illumina), and that our study included a larger number of samples (n = 697) than public resources (n = 60), we believe our results are accurate. Such a difference from the expected MAF may be due to reduced power in HapMap compared to this work or due to population stratification (i.e. the MAF of 0.09 for Yorubans in Nigeria could be markedly lower than elsewhere on the continent from where ancestors of our participants may have lived). More work into the genetic population structure across Africa and in admixed populations such as African Americans is needed to appreciate such differences.
Population-based differences in susceptibility to RA are observed through previous reports that show an association between RA and rs3087243 (+60C/T), a polymorphism known to affect the expression levels of soluble CTLA4 protein [35], in Swedish and North-American populations [14] or a lack of association at this locus in studies based in Massachusetts or Northern Ireland [7],[13]. We failed to find an association of rs3087243 in RA among African Americans. Even when stratifying for a clinical subclassification more strongly associated with CTLA4 [14] (anti-CCP positivity), we could not reproduce these results in African Americans. This non-replication finding may be due to genuine population-specific differences in allele frequency or different patterns of LD among African and European ancestry individuals, but our relatively small sample size precludes definitive conclusions. For example, to detect a small genetic effect [OR = 1.08 (95% CI: 1.01–1.17)] in a meta-analysis of genotypes, Plenge et al. analyzed ∼4,000 Caucasian RA samples [14], a much higher number of subjects than is available for our analysis. We also failed to find an association with the nonsynonymous SNP, rs231775 (+49A/G), which has been implicated in multiple autoimmune diseases, again possibly due to small sample size.
CTLA4 is an important molecule in preventing an inappropriate immune response and in dampening osteoclast formation [4], both of which may have implications for the pathogenesis of RA. CTLA4 stimulation functions in regulatory T cell development including proliferation and frequency [2],[3],[36], providing another possible mechanism for this protein to influence RA pathogenesis. While we do not address possible functional consequences of this polymorphism, future work may reveal a relationship between rs231778 and T cell/osteoclast development or linkage disequilibrium with a SNP outside of the CTLA4 gene boundaries that influences expression or function.
In conclusion, our results suggest a need for greater understanding of CTLA4 function and of the ethnic-specific genetic contributions to RA including relationship to disease pathogenesis.
|
10.1371/journal.pntd.0006658 | Campylobacter, a zoonotic pathogen of global importance: Prevalence and risk factors in the fast-evolving chicken meat system of Nairobi, Kenya | Campylobacteriosis is a leading foodborne zoonosis worldwide, and is frequently associated with handling and consumption of poultry meat. Various studies indicate that Campylobacter causes a substantial human disease burden in low to middle-income countries, but data regarding the organism’s epidemiology in countries like Kenya are scarce. In sub-Saharan Africa, 3.8 million deaths of children under-5 years of age are reported annually. Of those, 25% are caused by diarrheal diseases, and Campylobacter is one of the most frequently isolated bacteria from diarrheic children. With the growth of urban conglomerates, such as Kenya’s capital, Nairobi, changes in diets, food production systems, and retailing dynamics, it is likely that exposure and susceptibility to this pathogen will change. Therefore, the importance of Campylobacter disease burden in Kenya may increase further. The objectives of this study were: 1) to determine the prevalence of Campylobacter spp. in Nairobi’s small-scale chicken farms and meat retailers, and 2) to identify potential risk factors associated with its presence in those sites. The prevalence data provides the first detailed baseline for this pathogen in the urban Kenyan context. The risk factors provide context-specific insights for disease managers. A cross-sectional study of broiler, indigenous chicken farms, and chicken meat retailers, was conducted in a peri-urban, low to middle-income area (Dagoretti), and a very-low income informal settlement (Kibera) of Nairobi. Chicken faeces were collected using one pair of boot socks per farm, and 3 raw chicken meat samples were purchased per retailer. Samples were cultured for viable Campylobacter spp. using mCCDA, followed by blood agar plates in aerobic/microaerobic conditions for prevalence calculations. A questionnaire-based survey on sanitary, sourcing and selling practices was conducted at each site for risk factor identification using logistic regression analyses. A total of 171 farm premises and 53 retailers were sampled and interviewed. The prevalence results for Campylobacter spp. were between 33 to 44% for broiler and indigenous chicken farms, 60% and 64% for retailers, in Dagoretti and Kibera, respectively. Univariable logistic regression showed an association between Campylobacter spp. presence and the easiness of cleaning the display material used by the retailer. Restricting access to the flock was also associated with the pathogen’s presence. Multivariable logistic regression identified the selling of defrosted meat as a retailer risk factor (OR: 4.69; 95% CI: 1.31–19.97), calling for more investigation of the reported repetitive freezing-thawing processes and cold chain improvement options. At the farm-level, having a pen floor of material not easy to clean was found to increase the risk (OR: 2.31; 95%CI: 1.06–5.37). The relatively high prevalence of Campylobacter spp. across different areas and value chain nodes indicates a clear human exposure risk. The open nature of both small-scale broiler and indigenous chicken production practices with low biosecurity, hygiene and informal transactions, likely plays a role in this. While gradual improvement of farm biosecurity is recommended, risk factors identified suggest that consumer education and enforcement of basic food safety principles at the retailer end of the food continuum represent key targets for risk reduction in informal settings.
| Gastrointestinal disease following food-poisoning can cause severe clinical signs in humans and represent high costs for society. Examples of bacteria causing foodborne diseases include Salmonella and Campylobacter. In low to middle income countries, where resources are limited and a significant part of the population cannot always afford treatment, foodborne diseases such as Campylobacteriosis can play an important role in child mortality. Chickens and undercooked chicken meat have been found to commonly harbour this bacterium. In countries like Kenya, where fast urbanisation is occurring and chicken farming systems are intensifying, diets and food retailing infrastructure are also changing. Scientific research has not yet well documented how widely distributed Campylobacter is in such changing contexts, and which risk factors can favour its presence. In this study, the researchers have investigated small chicken farms and chicken meat sellers in Nairobi, Kenya’s capital, to better understand the risk that Campylobacter could represent for human health.
| Campylobacteriosis is one of the leading bacterial foodborne zoonosis globally [1], with handling and consumption of chicken meat identified as a major risk factor in high-income countries [2]. The estimated public health impact of Campylobacter-induced enteric disease is around 0.35 million disability-adjusted life years per year for EU-27, with annual costs estimated at about 2.4 billion euros [3]. Despite intensive research on the pathogen and testing of a range of control measures, campylobacteriosis has been the most frequently reported gastronintestinal disease in Europe since 2005 [4,5]. An overall rate of 59.8 cases of campylobacteriosis per 100,000 population was reported in 2014 for the European Union and two European Economic Area countries, ranging from 1.3 to 197.4 by country [4]. While differences in reporting, climatic conditions, and chicken production systems may explain differences in incidence, the epidemiology of the bacteria remains poorly understood and other factors may be involved. In low and middle-income countries (LMIC), surveillance for Campylobacter seldom exists in people and chickens, and data regarding the organism’s presence, risk factors and impacts are scarce. Yet, the disease burden of Campylobacter in the global South should not be underestimated. In Sub-Saharan Africa alone, 3.8 million deaths in children under 5 years are reported annually, 25% of which are caused by diarrheal diseases [6]. This bacterium is among the most common pathogens found in diarrheic children in LMIC [7]. In a multisite birth cohort study from 2009 to 2012 (MAL-ED study) in Asian, Latin American and African countries, Campylobacter spp. were the most frequently detected pathogens, occurring in 84.9% of 1892 children, and contributed the highest burden of diarrhoea in the first year of life. Campylobacter infection in children was associated with growth deficits across sites [8,9]. A Campylobacter isolation rate of 8% was reported in all-age diarrheic patients in Ethiopia [10], compared to 12% (higher than for Salmonella and Shigella) in Kenya [11]. A Campylobacter prevalence of 19% was reported in children under 5 from Morogoro, Tanzania [12], whereas a study in Western Kenya health centers isolated Campylobacter spp. from 42% of diarrheic children under 5 [13]. Hence, a better understanding of the sources of Campylobacter is needed to reduce diarrhoea-related child mortality. With the aim to address these data gaps, this study focused on Nairobi, Kenya, to investigate the epidemiology of Campylobacter spp. in a likely source, namely the chicken meat production system, in this setting.
Kenya’s capital, Nairobi, illustrates the global trend of fast urbanisation in LMIC countries. The human population has grown from 350,000 in 1962 to 3,375,000 in 2009, whilst the spread of informal settlements has led to over 60% of city’s population residing in conditions of significant poverty [14]. In parallel, the middle-class has been growing rapidly, with increasing demands in terms of food quality, and a surge in supermarkets and fast food outlets [15]. To meet the increasing demand in poultry meat, poultry production systems have been intensifying in Kenya [16]. An increase in commercial chicken farming, generally using imported fast growing broiler breeds such as the Cobb 500, is observed in and around Kenyan urban centres such as Nairobi, Mombasa, Nakuru, Kisumu and Nyeri, where the demand for poultry meat and market access for chicken producers are greater in comparison to rural areas. Outside of urban areas, indigenous chickens (i.e. local breeds which grow slowly and are used for egg and meat production) are the main chicken species kept [17]. These changes in retailing dynamics, diets, and poultry production systems, are altering the epidemiological setting for campylobacteriosis. While indications of protective immunity against the bacteria in adults [18] may have led to the disease been seen as low priority in LMIC, the evolving Nairobi setting may lead to significant changes in exposure and susceptibility to the disease in the population, and calls for a better understanding of the pathogen’s epidemiology.
While poultry is recognised as a major source of Campylobacter spp. [2], western studies have identified consumption of poultry meat, undercooked red meat, raw milk, untreated water, contaminated raw foods like salads, contact with pets and farm animals, and international travels as risk factors for disease in humans [2,16,17]. Studies investigating the disease in the global South are sparse. In the MAL-ED study covering 8 low-resource sites in Asia, Latin America and Africa, factors associated with a reduced risk of Campylobacter detection in children regular surveillance stools included treatment of drinking water, exclusive breastfeeding, access to an improved latrine, and recent macrolide antibiotic use [8]. C. jejuni and C. coli have been isolated from chickens, goats and sheep in Nigeria and similarities between strains isolated in chickens and humans suggest that poultry is an important source of human campylobacteriosis [7]. Risk factors identified for Campylobacter infection in people include home slaughtering and eating undercooked meat in Cambodia [19], the presence of animals or uncovered garbage in the cooking area, and lack of piped water in Egypt [7], contact with animals and HIV infection in Burkina Faso [20], young age, consumption of chicken meat and prepared salad in Tanzania [21], and poor hygienic conditions in LMIC in general [7].
The most important source of Campylobacter infection in chickens is thought to be the external environment [2]. Risk factors for Campylobacter presence reported for intensive commercial production systems in high-income countries include the use of contaminated water [22]; flock thinning (partial depopulation), carry-over from a previous flock following inadequate cleaning and disinfection, increasing bird age at slaughter and number of birds reared per year on farm [23,19]; organic rearing [24], broiler houses older than 15 years old, and long downtime between flocks [25]. A 2004 study in Senegalese broiler chickens found a 63% Campylobacter prevalence. On-farm presence of laying hens, cattle and sheep, lack of exclusive clothing for poultry workers, and use of chick transport cartons as feeders were found to increase the risk of infection in chickens, whereas thorough cleaning and disinfection of the poultry house were protective [26]. Two studies in South African broiler flocks found Campylobacter prevalence in chickens to be higher in rural areas (68%), compared to commercial indoor broiler flocks (47%) or layer flocks and (94%)[27]. To the authors’ knowledge, only one risk factor study has been published so far for chicken meat production systems in Nairobi, Kenya, which identified cleaning of the poultry house before restocking as a risk factor [28].
To mitigate carcass contamination by the intestinal tract of positive birds during the slaughter process [29], industrial slaughter and processing facilities in high-income countries use a variety of strategies such as chemical treatment, irradiation or freezing of carcasses to reduce the bacterial count [2,25]. At retailer level, general hygiene measures to prevent cross-contamination between the meat, retailer’s hands, contact surfaces and utensils, are recommended to minimise Campylobacter spread [30]. National prevalence in chickens and chicken meat have been found to vary greatly worldwide, from 4.9 to 100% in EU broiler carcasses [31], with a mean prevalence in broiler meat across Europe in 2015 of 46.7% [5], and from 8 to 100% in poultry meat at retail level across 32 different countries globally [32]. The mean prevalence reported for poultry meat in Senegal and South Africa in the latter study was 73.1%. In the sub- Saharan Africa context, Campylobacter prevalence in chicken meat was found to be 81.9% in poultry processing plants [33] and 100% in retail outlets in Nigeria [34], and varied between 11.1% and 100% in South African supermarkets [35]. One study found a prevalence for thermophilic Campylobacters of 77% (C. jejuni 59%, C. coli 39% and C. laridis 2%) in raw chicken sourced from butcheries, markets and supermarkets in Nairobi, Kenya [11], while studies in Ghana and Ethiopia found a prevalence close to 22% [36,37].
Risk factors identified for industrial commercial chicken production in high-income countries are highly context-specific and cannot be applied directly to informal meat production systems, such as small-scale Nairobi chicken farms, where biosecurity is limited, even in commercial broiler operations. Except for a few large integrated broiler companies and high-end supermarkets chains, informal production and retailing still dominate [31,14]. The lack of literature on Campylobacter risk factors in food animals and food animal products in LMIC, where rearing systems and level of hygiene may differ greatly from Western settings, represents a major gap [38].
Considering the public health importance of Campylobacter, especially for vulnerable groups in LMIC, poultry’s predominant role in the global North as a risk factor, and the scarcity of epidemiological data in the context of rapid African urbanisation, the objectives of this study were: 1) to determine the prevalence of Campylobacter spp. in Nairobi’s small-scale chicken farms and chicken meat retailers, and 2) to identify potential risk factors associated with the presence of Campylobacter spp. in those same sites. The data provide a system-wide picture of the risks of exposure to Campylobacter at farm and retailer levels, and the first detailed baseline for this pathogen in the urban Kenyan context, whilst the identified risk factors help understand its epidemiology and provide insights for Kenyan disease managers.
The selection of Nairobi was based on the following criteria: representativeness of growing urban centers in East Africa, transitioning urban landscape and evolving chicken production systems. Nairobi, one of the major fast-growing urban centers in East Africa, with both a growing middle class and expanding informal settlements, is a prime candidate to investigate the epidemiology of Campylobacter in the context of transitioning urban landscape and chicken production systems. The study design was informed by previous work on the chicken meat value chains in Nairobi [39]. In the latter study, small-scale broiler and indigenous chicken farms, and small-scale broiler meat retailers were identified as key nodes, and were therefore targeted for the understanding of the risk of exposure to Campylobacter. Small-scale chicken farms were defined as a flock of 2 to 100 birds for indigenous chicken flocks, and 2 to 800 birds for broiler flocks, and small-scale broiler meat retailers were defined as any premise selling raw broiler meat (butchery) or a mix of raw and cooked meat (combination of butchery and small restaurant), not belonging to a franchise; they were found to be the most numerous chicken meat value chain actors in Nairobi. Poultry abattoirs and indigenous chicken meat retailers were found to be rare in Nairobi (indigenous chickens are commonly sold on-farm directly to consumers), and were therefore excluded. Large integrated broiler companies could not be sampled due to the sensitivity of the business information. A cross-sectional survey of small-scale broiler and indigenous chicken farms as well as broiler meat retailers was conducted (layer chickens were excluded). In order to provide a representative picture of Nairobi’s food system and major types of urban landscape found in the city, two areas of different wealth levels and production systems were purposely selected. Dagoretti, a low to middle-income, peri-urban area, characterised by a rural-like landscape with pockets of residential areas and moderate population density, was selected as a major livestock raising area within Nairobi, and due to its easy accessibility for the research team. Kibera, characterised by high population density, fully urban landscape and lower livestock activity, was selected as it represented the largest very low-income informal settlement (slum) in Nairobi. As major differences in the value chain structure and risky practices had been identified by Carron et al., 2017 [39], these two areas were targeted to test the hypothesis whether socio-economic status could affect the presence and survival of Campylobacter at farm and retailer level.
Sample sizes were calculated for independent populations (see S1 Appendix for more information on sample size calculation), namely two types of chicken production systems per area, and one retailer group per area, using an expected Campylobacter spp. prevalence of 50%, a 10% confidence limit and 90% confidence interval. No regular records of farms were available to guide the selection of farms to be sampled. The team worked with community elders that had been recruited to participate in the project to create a census of all broiler farms in each area. Since the number of broiler farms was limited (close to or below the calculated sample size), all were targeted for sampling. Because it is a common practice in the study sites to own indigenous chickens, it was not realistic to undertake a census of indigenous chicken farms. This resulted in an overall sample size for Dagoretti and Kibera, respectively, of 42 and 8 small-scale broiler farms, 67 and 63 small-scale indigenous chicken farms, and 21 to 40 small-scale broiler meat retailers per area. Using the target sample size for indigenous farms, a corresponding number of random GPS coordinates within each area was computer-generated using ArcGIS. The first farm found North of each GPS point by the sampling team was targeted for sampling. A census approach was used for broiler meat retailers, as these were reported by the elders to be few, and located along a few main streets in each area. In Dagoretti, the census of all butcheries selling chicken meat was performed by walking or driving along the main streets and asking employees whether they sold chicken. GPS coordinates for each retailer selling chicken were recorded. Due to the small number of retailers (close to or below the calculated sample size), it was decided to sample all retailers willing to participate. In Kibera, due to security issues, a key informant was asked to perform a similar retailer census with support from the local elders.
On chicken farms, one sample each of chicken faeces and/or housing litter was collected using boot socks dampened with sterile saline [40]. Three meat samples (100g or more) from different chicken carcasses were bought from each retailer. Each boot sock pair or meat sample was put in a sterile ziplock bag and stored in a cool box with ice packs, until testing at the laboratory within 5 hours of collection. All samples were cultured for viable Campylobacter spp, using a protocol based on the ENIGMA consortium 2017 study [40]. Boot sock samples were enriched using 50 ml of Exeter broth and incubated at 42°C under aerobic conditions with a minimal air space for 24 hours before sub-culturing. A 50g piece of each meat sample was cut aseptically, added to 200ml of saline and subject to stomaching for 1 minute; 5ml of the stomacher content was added to 5ml of double strength Exeter broth, and 10ml of the enriched sample incubated similarly to boot sock samples. All samples were then cultured for viable Campylobacter spp. at the Kenya Medical Research Institute (KEMRI) of Nairobi. Samples were first plated onto mCCDA and incubated for 48 hours at 42°C under microaerobic conditions (using CampyGen microaerobic gas pack in jars). Plates were visually examined for suspect Campylobacter colonies using colony size, shape and surface colour, and other key characteristic: C. jejuni 2.0–3.0 mm, flat/entire/glossy, grey/white, can be efflorescent (spreading moist), and C. coli 1.0–2.5 mm, convex/entire/glossy, creamy grey moist. For each mCCDA plate that showed growth for Campylobacter spp., four suspect colonies were subcultured on two different Columbia blood agar plates. One plate was incubated under microaerobic conditions for 48 hours at 42°C, and the other under aerobic conditions at 37°C for 48 hours. Growth in microaerobic conditions only was considered as positive for Campylobacter spp. A subset of isolates (428/560) was confirmed by LPX-PCR [41].
In order to evaluate risk factors for Campylobacter spp. exposure, a questionnaire was used to collect data from each site visited. The farmers’ and retailers’ questionnaire (S2 and S3 Appendices) covered the following categories of variables/themes (Table 1): 1) Farm or retailer’s environment and characteristics, 2) Management practices, 3) Biosecurity, health or sanitary practices, and 4) Sourcing and selling of chickens/chicken products. Questionnaires were written in English and conducted using Open Data Kit (ODK, https://opendatakit.org/about/tools/) software on electronic tablets. Sites and samples were identified by scanning unique barcodes. Enumerators were Kenyan citizens familiar with the city, bilingual in English and Kiswahili. Pre-sampling training of the enumerators on the questionnaires took place.
Prior to data collection, ethical approvals were sought from the ILRI-IREC (International Livestock Research Institute—Institutional Ethical Research Committee, project reference ILRI-IREC2016-01). ILRI-IREC is accredited by the National Commission for Science, Technology and Innovation (NACOSTI) in Kenya. Approval from the Royal Veterinary College (RVC) ethical committee was also received (project reference: URN 2015 1453). Permission to interview people was obtained from the Ministry of Agriculture and the local Veterinary Authorities. The study’s objectives and participants’ rights were explained in Kiswahili to farmers and retailers upon arrival at the site. Verbal and written consent to participate in the study were obtained before initiating data collection.
Variables in the survey data with too many missing observations (>25%) and variables with no substantial variability (>95% responses identical) were not kept for analysis. This first variable screening lead to a total of 45 farm exposure variables and 43 retailer variables for inclusion in the risk factor analysis. Using Excel and R version 3.3.2 (2016-10-31), each site (farm or retailer) barcode and meta data was linked to the corresponding sample barcodes and laboratory results. Inconsistencies and data gaps were reviewed with the field coordinator and discussed with laboratory technician to clean the database.
Using a chicken farm or retailer as a sampling unit, a site with one or more positive samples on culture classified as positive for Campylobacter spp. A sample was considered positive if at least one isolate was obtained and comfirmed by culture. Culture prevalences were calculated using QuickCalcs (GraphPad, http://www.graphpad.com). A Fisher’s exact test was used to compare prevalence between groups. The LPX-PCR results obtained for a subset of the samples were used to confirm the culture prevalences.
A two-step statistical analysis using univariable logistic regression followed by multivariable logistic regression, was performed to identify risk factors for the presence of Campylobacter spp. at farm, or retailer level, respectively. No derivative analysis was made for farms or retailers from a specific area (Dagoretti or Kibera), or for a specific type of farm (broiler vs. indigenous chicken) due to the limited size of each subgroup. Rather, area and type of farms were included as confounders in the models.
A univariable analysis was performed in order to identify possible associations between the 88 selected exposure variables and the the presence of Campylobacter spp. using univariable logistic regression for each of the predictors. Odds ratio (OR) and 95% confidence intervals (CI) were calculated. All variables with a p-value (calculated with a likelihood ratio test) lower than 0.2 were retained for assessment in the multivariable analyses, except if the variable belonged to a nested question not applicable to the whole “farm” or “retailer” population.
Two multivariable logistic regression analyses were conducted independently, one for retailers, one for farmers. In each analysis, variables selected during the respective univariable analysis were included in the initial model. A stepwise backward selection procedure was used to refine models until all variables remaining in each model met the criterion of a p-value ≤0.05. Two-way interactions between predictors were assessed using a likelihood ratio test and considered significant if p ≤0.05.
In order to evaluate potential collinearity effect between predictors, the levels of association between risk factors identified during the univariable analysis were assessed using a Fisher test; risk factors with more than two-fold changes in the logistic regression coefficients were also checked during the selection process.
As data collection took place following a sampling frame designed for investigating Campylobacter spp. prevalence in two Nairobi areas and two types of chicken farms, multiple logistic regression models were built to account for the potential confounding effect of the study design. One farm model included “farm area” and “farm type” variables, despite their non-significance in the univariable analysis, whilst a second model did not include them. Similarly, one retailer model included the “retailer area”, and another did not include this variable.
The predicted probability was calculated for each observation based on the final model and the fit was assessed using the distribution of the model’s residuals, residuals close to zero suggesting a good fit [42]. Finally, an R-squared value was calculated [43]. All statistical analyses were performed using the statistical software R version 3.3.2 (2016-10-31).
Two broiler farms declined participating in the survey due to fear of pathogen introduction. Another 4 indigenous chicken farms declined sampling with no specified reasons. An estimated 25% of broiler farms on the census could either not be reached, or did not raise broilers over the course of the sampling months. In total, 171 farms were sampled; 18 and 7 small-scale broiler farms, and 78 and 68 small-scale indigenous chicken farms, in Dagoretti and Kibera, respectively. One questionnaire was administered per site, but as some sites had multiple sheds, 181 boot sock pairs were collected. An estimated 10% of small-scale broiler meat retailers declined participation in the survey, mainly due to the absence of the owner on the premises, or lack of time. A total of 53 retailers were successfully surveyed, 25 in Dagoretti, and 28 in Kibera; and 183 meat samples were collected.
The culture prevalence of Campylobacter spp. in small-scale farms varied between 33 and 44% across types of production systems and areas, whereas the prevalence in retailers was 60% in Dagoretti and 64% in Kibera (Table 2). While Campylobacter spp. prevalence at retailer level was higher than at farm level, no statistically significant difference was found between the types of site. Out of the 429 isolates tested by LPX-PCR, only 1 was not confirmed as Campylobacter, suggesting reliable culture resuts. A total of 63% of the 428 Campylobacter isolates were C. Jejuni.
This study is the first to document Campylobacter spp. prevalence in both small-scale chicken farm and chicken meat retailer levels in Nairobi and to investigate factors determining the heterogeneity of Campylobacter presence in these settings. The results provide valuable insights into the potential risks of human exposure in an otherwise undocumented context. The great variability found in Campylobacter prevalence across broiler batches or carcasses in the EU, the limited number of similar studies in the East African context, and the differences in epidemiological units used in the literature (e.g. retailer versus carcass-level prevalence), make it difficult to compare these results directly with other studies. However, the relatively high Campylobacter prevalence results found in Nairobi retailers echoes some of the prevalence reported (73.1% or higher) in retail poultry meat in sub-Saharan Africa [28,36,37]. In Nairobi, Kenya, isolation rates of 59% for C. jejuni, 39% for C. coli, and 2% for C. laridis were found in raw chicken sourced from butcheries, markets and supermarkets [11], with chicken meat tested less than 24 hours after slaughter showing a higher prevalence (85.3%). Time since slaughter might aso have influenced results in our study, since meat samples were not collected at the slaughter plant. Few studies identified much lower Campylobacter prevalence values, such as 21.7% in retail raw chicken meat tested in Ethiopia [36], and 21.9% of commercial chicken carcasses swabbed in Ghana [37]. Broiler flock prevalence in our study are moderately lower than in other sub-Saharan African studies (47% to 68% Campylobacter prevalence overall) [10,44], which might be due to the small number of broiler farms sampled, to a difference in size of commercial flocks, or a difference in sampling unit and testing methods. Few studies found a prevalence lower than 30%. A Ghanaian study found Campylobacter in 22.5% of ceacal samples [37], a Tanzanian study in 42.5% of chickens (various breeds) using cloalcal swabs [12] and an Ethiopian study, in 28.9% of chickens (various breeds) [45]. A study from 1988 found a prevalence of 51.5% in Kenyan broilers [46], whereas a 2018 study found an overall prevalence of 69.5% in Nairobi chickens [28]. The prevalence results in our study are indicative of a relatively uniform distribution of the pathogen across the chicken meat system studied. This can most likely be explained by the informal nature and overall lack of biosecurity in these systems, which is unlikely to limit the introduction of Campylobacter into either indigenous or broiler flocks. Unlike in Europe and North America, practices used in broiler versus “backyard” indigenous chicken farms in Nairobi share more similarity. Due to a lack of resources, small-scale Nairobi broiler rearing infrastructure is heterogeneous, using suboptimal materials, and often in proximity to other livestock. Flock management is often lead by irregular market access, with limited sanitary considerations. This is exacerbated in informal settlements where space is lacking, and resources are further limited. In such areas, a broiler flock can be found under a vegetable shop stall or staircase. The limited number of broiler farms observations, as fewer broiler farms than expected were identified, also limits the power of this study to identify differences between management systems. Indeed, studies in Ethiopia and Tanzania have identified marked differences in prevalence between broiler and indigenous chicken flocks, with conflicting results. Two studies in Tanzania found a higher Campylobacter prevalence in indigenous chickens (76.49% and 75%) compared to broilers (26.4% and 50%) [47,12]. Another Tanzanian study found no significant difference between broilers and indigenous chickens, but rather a higher prevalence in local chickens from rural areas compared to those in urban areas [48], while an Ethiopian study found significantly higher Campylobacter isolation rates in animals (chicken, sheep, cattle and pigs) in urban areas (56.7%) compared to rural areas (26.7%)[49]. Finally, a 2018 study found a prevalence of Campylobacter of 91.07% in broilers, 70.96% in layers, and 61.04% in indigenous chickens in peri-urban areas of Nairobi, Kenya [28]. The higher prevalence found in meat sellers compared to farms in our study may be explained by the risk of cross-contamination between chicken meat products of mixed sources during meat handling, cutting, storage and display. A Ugandan study found Campylobacter survived much better on wooden cutting boards than plastic or metal ones [50], wooden boards being widely used in Nairobi retailers. Nairobi-specific factors that may affect Campylobacter’s survival include the average temperature, which is constantly above 16°C, or the precipitation which is high (80 to 191 mm) during the two rainy seasons. Indeed, unlike the reported summer and autumn peaks of campylobacteriosis in Europe and North America, seasonality of Campylobacter has not been reported in LMIC, potentially due to a lack of study in tis setting [51]. The common practice of freezing and defrosting chicken meat in Nairobi, further discussed below, could also influence the bacteria’s presence. In addition to investigating the prevalence of Campylobacter in the meat system, determining the level of contamination of the chicken meat sampled in Nairobi retailers would have brought an additional key indication of the risk of human exposure, but was not feasible due to resource limitations. A higher load of Campylobacter on meat increases the risk of contamination of the direct meat environment and spread within a household, or site. The European Food Safety Authority (EFSA) has estimated a public health risk reduction of 50%–90% could be achieved, if all broiler batches complied with the critical limit of <1000 and <500 CFU/g of neck and breast skin, respectively [52]. However, the infectious dose for Campylobacter being low at a few hundred cells (500 or less) [53], prevalence of the bacteria at retailer-level was considered an appropriate indicator of the risk of exposure in this study.
Few explanatory variables were found to have a significant association with the presence of Campylobacter spp. in the univariable analysis or were identified as risk factors in the multivariable analysis. Retailers using a display material “not easy to clean”(e.g. made of wood or porous material) were shown to have higher odds of Campylobacter spp. presence, compared to those using a display material easy to clean. This is in line with literature describing lower levels of hygiene at retail-level as a risk factor [53]. A risk assessment of Campylobacteriosis linked to chicken meals prepared by households in Dakar, Senegal, determined that washing of cooking utensils during food preparation was not sufficient to significantly reduce the risk of Campylobacteriosis, whereas changing knife, board and dishes between pre and post-cooking was [54]. “Selling defrosted meat” increased the odds of Campylobacter spp. presence in both steps of the analysis. This finding is surprising given that freezing can be used as a strategy to reduce numbers of Campylobacters present on the meat [55,56,57,58]. However, freezing-thawing of chicken meat was found to be a common retailer practice in Nairobi and could favour re-contamination of the meat. Multiple retailers interviewed described how they turned off their freezer during the day to soften the meat for cutting, and turned it back on at night to preserve unsold meat until the next day. Freezing fresh chicken meat for 24 hours has been shown to reduce the log number of viable Campylobacters by up to 2.5 [55,56], and a 2–3 day freezing period to diminish the risk by 50–90% [52]. However, freezing temperatures in Nairobi are not verified, and incomplete freezing may be common. The repetitive freezing-thawing-refreezing practice observed in chicken retailers is driven by resource scarcity, and the demand from consumers for small quantities of chicken meat. The latter has led to a selling culture of cutting small pieces of meat from a whole carcass in the presence of the customer. Hardly any retailers were found to freeze small meat pieces in individual packaging. This may be due to customers wanting to see the carcass of origin. Where cold chain infrastructure is more affordable and food hygiene is strictly regulated and enforced, multiple freezing-thawing cycles are not allowed. Studies have found that the refrigeration prior to freezing, as well as the type of meat surface (e.g. skin versus meat muscle, or ground chicken) will affect the number of Campylobacter cells surviving freezing [32,36]. A 2013 study by the UK Food Standards Agency determined that the freezing temperature and length of time taken to freeze chicken livers influenced the bacteria’s survival [55]. Another study found lower Campylobacter prevalence in chicken meat from Malaysia wet markets compared to supermarkets, hypothesising that the chilling infrastructure in supermarkets favours survival of the bacteria whereas the ambient temperature of 29.6°C in wet markets is not favourable for growth [59]. On the other hand, a Kenyan study found lower levels of E. coli contamination in raw chicken meat sold in supermarkets compared to smaller-scale retailers [60]. This illustrates how specific freezing processes can influence risk reduction, and have a different effect depending on the pathogen. It highlights how significant the challenges linked to the cold chain can be in a context of limited resources, especially for a highly perishable product like chicken meat. Further research will need to investigate if the Campylobacter presence related to defrosted meat identified in this study originates from the freezing-thawing process with sub-optimal cold chain conditions, or from cross-contamination post-freezing. Since chicken meat freezing in Nairobi was well accepted by consumers, food safety interventions could capitalise on this practice and its potential for Campylobacter risk reduction. Awareness trainings regarding sanitary practices to avoid cross-contamination and promoting the freezing of small chicken pieces wrapped individually to minimise the handling and repetitive thawing-freezing could be considered.
In the farm univariable analysis, three variables were identified as having a significant association with the outcome of interest. The predictor “restricting access to the flock” was found protective. Arsenault et al. [61] specifically assessed the permanent locking of the broiler house, which was associated with a reduced risk of Campylobacter colonization in chickens. This practice is not readily applicable for indigenous free-ranging chickens. Even in the case of broilers, while greater access restriction could be encouraged by providing training to farmers, it is unlikely to result in any significant risk reduction without the general on-farm biosecurity being improved. This would require substantial investment, which in turn would demand external support or simultaneous improvement of small-producers’ market access and business profitability. Using a pen material “not easy to clean” and cleaning the pen without disinfectant were also found to increase the odds of Campylobacter presence, in line with literature citing inadequate cleaning and disinfection between flocks as risk factors [2]. Of interest, is a similar result from a 2018 Nairobi study, which only identified cleaning and disinfection of the chicken house before restocking as a risk factor (p<0.05) in the multivariable analysis [28].
Many of the risk factors identified in the literature for Campylobacter at farm-level (e.g. water source, thinning, biosecurity measures) [2,23,61,62], and retailer (or carcass) level (e.g. contact between different carcass parts (e.g. liver and meat), cross contamination via handling practices) [63], were tested in the univariable analysis, yet did not show any significant association with the presence of Campylobacter. The identification of few risk factors may be linked to the cross-sectional sampling design, less suited for risk factor analyses compared to case-control or cohort studies, and selected for the Campylobacter prevalence estimation objective. The limited number of observations, as well as the high number of potential risk factors, may have also limited the power of the study. In addition, the specificity of the Nairobi context and scarcity of similar studies in informal settings, are likely to explain some of the discrepancies with Western studies. The extreme variability in production practices and in the level of implementation of sanitary practices in this context is difficult to analyse accurately and illustrates the challenges of capturing risk factor data in messy settings. Overall, we can hypothesize that the minimal biosecurity and sanitary measures observed in small-scale Nairobi farms and retailers create an open system, with numerous sources of contamination, making individual risk factors hard to identify and isolate from the general environment. Still, the study, despite not following a risk assessment structure, provides useful risk indicators to be further investigated. While Roesel and Grace [15] have found that formal retailing settings in sub-Saharan Africa do not necessarily translate into a lower risk for consumers, repeating the analysis made for small-scale retailers in Nairobi’s high-end supermarkets, where stricter sanitary standards are applied, would enhance our understanding of the broad risk context. While the prevalence and risk factor analyses were designed to provide a system-wide picture of the risks of exposure to Campylobacter at farm and retailer levels, it should be noted that the lack of information regarding the origin of the carcasses at retailer-level limits our understanding of transmission dynamics in the chains. Indeed, retailers in Dagoretti have been found to source their carcasses locally, whereas Kibera retailers have reported selling low-value cuts from from major integrated broiler companies outside the informal settlement (see S4 Appendix for more information on the chicken meat supply).
In terms of recommendations arising from this study, the risk factors identified support training initiatives on biosecurity and food safety practices. Group feedback sessions are planned for farmers and retailer having participated in the study, including education on basic biosecurity principles, sanitation measures and safe handling of chicken meat. While gradual improvement of biosecurity measures (via appropriate cleaning and disinfection, better farming infrastructure and flock management) targeted at commercial farms should be supported, initiatives focusing on consumer education and enforcement of basic food safety principles seem more easily manageable, and with potentially greater impact as a first step, in informal settings. By using a risk-based sampling approach based on a value chain analysis to design the prevalence and risk factor analyses, this study presents methodological novelty. Substantial economic value chain studies in Africa can be found, but the combination of value chain analysis and risk identification, or disease investigation, remains limited. Finally, this study is the first to describe Campylobacter prevalence and risk factors both in chicken farms and chicken meat retailers at this level of detail in a peri-urban and informal settlement Kenyan setting, providing key insights into the specificities of Campylobacter epidemiology in quickly urbanising areas of East Africa.
|
10.1371/journal.ppat.0030052 | TGF-β Signaling Controls Embryo Development in the Parasitic Flatworm Schistosoma mansoni | Over 200 million people have, and another 600 million are at risk of contracting, schistosomiasis, one of the major neglected tropical diseases. Transmission of this infection, which is caused by helminth parasites of the genus Schistosoma, depends upon the release of parasite eggs from the human host. However, approximately 50% of eggs produced by schistosomes fail to reach the external environment, but instead become trapped in host tissues where pathological changes caused by the immune responses to secreted egg antigens precipitate disease. Despite the central importance of egg production in transmission and disease, relatively little is understood of the molecular processes underlying the development of this key life stage in schistosomes. Here, we describe a novel parasite-encoded TGF-β superfamily member, Schistosoma mansoni Inhibin/Activin (SmInAct), which is key to this process. In situ hybridization localizes SmInAct expression to the reproductive tissues of the adult female, and real-time RT-PCR analyses indicate that SmInAct is abundantly expressed in ovipositing females and the eggs they produce. Based on real-time RT-PCR analyses, SmInAct transcription continues, albeit at a reduced level, both in adult worms isolated from single-sex infections, where reproduction is absent, and in parasites from IL-7R−/− mice, in which viable egg production is severely compromised. Nevertheless, Western analyses demonstrate that SmInAct protein is undetectable in parasites from single-sex infections and from infections of IL-7R−/− mice, suggesting that SmInAct expression is tightly linked to the reproductive potential of the worms. A crucial role for SmInAct in successful embryogenesis is indicated by the finding that RNA interference–mediated knockdown of SmInAct expression in eggs aborts their development. Our results demonstrate that TGF-β signaling plays a major role in the embryogenesis of a metazoan parasite, and have implications for the development of new strategies for the treatment and prevention of an important and neglected human disease.
| Schistosomes are parasitic worms that infect hundreds of millions of people in developing countries. They cause disease by virtue of the fact that the eggs that they produce, which are intended for release from the host in order to allow transmission of infection, can become trapped in target organs such as the liver, where they induce damaging inflammation. Egg production by female schistosomes is critically dependent on the presence of male parasites, without which females never fully develop, and (counterintuitively) on the contribution of signals from the host's immune system. Very little is understood about the molecular basis of these interactions. Here, we describe a newly discovered schistosome gene, which is expressed in the reproductive tract of the female parasite and in parasite eggs. The protein encoded by this gene is made only when females are paired with males in an immunologically competent setting. Using recently developed tools that allow gene function to be inhibited in schistosomes, we show that the product of this gene plays a crucial role in egg development. Examining how the expression of this gene is controlled has the potential to provide insight into the molecular nature of the interactions between male and female parasites and their hosts. Moreover, the pivotal role of this gene in the egg makes it a potential target for blocking transmission and disease development.
| Amongst the Bilateria, transforming growth factor–β (TGF-β) signaling is recognized as playing an essential role in embryogenesis in deuterostomes and in arthropod protostomes, but its role in lophotrochozoan protostomes is unclear [1]. Schistosomes, the causative agents of schistosomiasis, one of the major neglected tropical diseases [2,3], are metazoan parasites that belong to the lophotrochozoan phylum Platyhelminthes.
Components of TGF-β signaling have been molecularly characterized in metazoans throughout the animal kingdom. Activation of this pathway begins at the cell surface when a dimeric ligand binds a complex consisting of types I and II receptor serine/threonine kinases [4]. Upon ligand binding, the constitutively active type II receptor phosphorylates and activates the type I receptor, which then phosphorylates cytoplasmic Smad proteins that translocate to the nucleus, where they mediate gene expression [4]. Components of a functional TGF-β pathway(s), including one type I receptor [5] (Schistosoma mansoni receptor kinase-1 [SmRK1], S. mansoni transforming growth factor–β type I receptor [SmTβ RI]), one type II receptor [6,7] (SmRK2, SmTβ RII), and three Smads [8–10], have been identified in S. mansoni, with nearly all components localized to either the surface of the worm or reproductive tissues of the female [5–9,11]. Nevertheless, while nearly the entire transcriptome of S. mansoni has been examined with the identification of 163,000 expressed sequence tags (ESTs) [12], a ligand of parasite origin for the TGF-β pathway(s) has remained elusive. This has led to the hypothesis that the ligands for schistosome TGF-β receptors are of host origin [5,13,14], and a suggestion that host TGF-β, signaling through SmRK2, plays a role in the pairing of male and female parasites [7].
Sexually mature S. mansoni live within the mesenteric vasculature, where each female produces approximately 300 eggs each day. Transmission of schistosomiasis depends upon the release of parasite eggs from the human host. Development of an immature egg into a mature egg containing a miracidium, the stage of the parasite that invades the intermediate fresh water snail host, occurs outside of the female worm, and takes approximately 5 d. Many of the eggs produced by schistosomes fail to reach the external environment, but instead become trapped in host tissues, where pathological changes caused by the immune responses to secreted egg antigens cause disease [15]. Despite the central importance of egg production in transmission and disease, and recent advances in proteomics and transcriptomics [12,16–18], essentially nothing is known of the molecular pathways involved in embryogenesis in schistosomes.
In this study, we describe the cloning and characterization of a S. mansoni TGF-β homolog, S. mansoni Inhibin/Activin (SmInAct). Although we found SmInAct to be expressed in adult male and female parasites, and in eggs, the localization of SmInAct expression to the reproductive organs of female parasites focused our attention on the role of this gene in egg production. A role for SmInAct in reproduction was supported by analyses of female parasites recovered from infertile infections, in which we found that SmInAct protein was undetectable. Confirmation of the importance of this TGF-β superfamily member in the reproductive process was obtained from RNA interference (RNAi) studies, in which targeted knockdown of SmInAct in female worms or directly in the eggs that they produce resulted in a marked cessation of embryogenesis.
SmInAct was identified through a tblastn search of the Wellcome Trust's Sanger Institute's S. mansoni genome sequence using the C-terminal region of the Drosophila melanogaster dActivin sequence. We were unable to identify SmInAct in EST databases regardless of whether we searched using the coding or 3′–untranslated region (UTR) sequences. The 5′ and 3′ ends of SmInAct were amplified via rapid amplification of cDNA ends (RACE) using primers designed from within putative coding sequence and adult S. mansoni cDNA as template. The 1.3-kb, full-length SmInAct transcript contains 10 base pairs (bp) of 5′UTR, 808 bp of 3′UTR, and a poly-A tail. The deduced amino acid sequence of SmInAct is 161 residues long and contains many of the molecular hallmarks for a TGF-β, including a putative basic proteolytic cleavage site located at position 32 as RQRR where the bioactive, C-terminal domain (126 amino acids) is enzymatically separated from the N-terminal pro-domain. Nine invariant cysteine moieties, and invariant proline and glycine residues (Figure 1A) essential for the proper dimerization and tertiary structure of a TGF-β homolog, are all predicted in SmInAct. The deduced amino acid sequence of SmInAct contains one putative N-linked glycosylation site at position 110. Within the bioactive domain, SmInAct is 27% identical to both DAF-7 from Caenorhabditis elegans and dActivin from D. melanogaster, and 29% identical to human TGF-β 1 (Figure 1A). Phylogenetic analysis of SmInAct among other TGF-β superfamily members groups this homolog with members of the TGF-β/Activin subfamily (Figure 1B), and further clusters SmInAct phylogenetically with TGF-β homologs from the free-living nematode C. elegans (DAF-7) and the parasitic nematodes Brugia malayi (Bm-TGH-2) and Strongyloides stercoralis (Ss-TGH-1).
To determine the expression of SmInAct at the transcript level, real-time reverse transcriptase–polymerase chain reaction (RT-PCR) was performed on cDNA from eggs, adult male parasites, and adult female parasites from mixed-sex infections. As seen in Figure 2A, SmInAct is expressed in all stages tested at relatively similar levels. Western analyses using polyclonal antibodies against recombinant SmInAct were used to determine the protein expression profile of SmInAct. The anti-SmInAct serum recognized a 28-kDa protein in egg antigen extracts and a doublet (32 kDa and 28 kDa) in adult male and female extracts (Figure 2B, lanes 1–3); these bands presumably represent the unprocessed (32 kDa) inactive and processed (28 kDa) active forms of the molecule. The relative molecular weights of the two bands recognized by anti-SmInAct antiserum in parasite extracts are larger than that predicted by the sequence, presumably due to detergent and reducing agent-resistant dimerization, and/or to glycosylation at amino acid 110. Glycosylation plays an important role in the solubility and secretion of other members of the TGF-β superfamily [19,20]. Eggs appear to contain only the lower molecular weight, putatively active form of SmInAct. To localize SmInAct within the parasite, we performed in situ hybridization on sections of adult worms. Anti-sense probes localized SmInAct transcripts to the reproductive tissues of the adult female, with strong signals in the vitellaria and ovary (Figure 2C), whereas in adult males, SmInAct transcripts localized to various subtegumental regions (Figure 2D).
The expression pattern in the female suggested a role for SmInAct in egg production. We focused on this possibility, and reasoned that if this were the case, SmInAct expression might be diminished in unfertile females. In vivo, successful oogenesis requires the presence of male schistosomes [21], and, for reasons that have remained unclear, an intact CD4+ T lymphocyte compartment within the host [22]. Therefore, we analyzed SmInAct expression in female parasites from mice harboring single-sex infections, and in parasites from severely lymphopenic interleukin-7 receptor knockout (IL-7R−/−) mice carrying mixed-sex infections, which produce a significant number of dead eggs [23,24]. Real-time RT-PCR demonstrated that SmInAct mRNA levels were significantly decreased, but not absent, in females from these infections (Figure 2E). Of particular interest, SmInAct protein was undetectable by Western analyses in females from single-sex infections as well as from infections of IL-7R−/− mice (Figure 2B).
While the localization of SmInAct transcripts to the male subtegumental region is not immediately informative in terms of function in the male, we nevertheless noted that male parasites recovered from infertile infections in IL-7R−/− mice were similar to female parasites in terms of transcriptional and post-transcriptional regulation of SmInAct expression (Figure 2B and 2F). Moreover, this was also the case for male parasites recovered from male single-sex infections (Figure 2B and 2F).
To gain a better understanding of the function of SmInAct and the signaling pathway it activates, this TGF-β homolog was targeted for knockdown via RNAi [25–27]. Pairs of adult males and females recovered from infected mice were soaked in double-stranded RNA (dsRNA) corresponding to SmInAct (1 μg/ml) or an irrelevant control dsRNA (luciferase) for 1 wk in vitro, followed by RNA extraction and real-time RT-PCR analyses. SmInAct dsRNA–treated worms showed a consistent and significant decrease in SmInAct expression of >40% when compared to SmInAct expression in worms soaked in the irrelevant control dsRNA (Figure 3A). No consistently significant difference in the numbers of eggs produced by control versus SmInAct dsRNA–treated worm pairs was observed, suggesting that SmInAct is not important for egg production per se. However, in examining these cultures, we noted that eggs produced by SmInAct dsRNA–treated parasites failed to develop (unpublished data). To specifically address the role of SmInAct in egg development, we treated eggs directly with SmInAct dsRNA. Approximately 20% of eggs laid by adult parasites during the first 2 d of in vitro culture will develop over the ensuing 5 d to contain miracidia [28], with a typical progression of development through six stages illustrated in Figure 3B. Therefore, eggs produced by worm pairs for the first 2 d ex vivo were collected and soaked in dsRNA (1 μg/ml) corresponding to SmInAct or an irrelevant dsRNA for 5 d, and their development was scored. Relative to eggs soaked in an irrelevant dsRNA, where ∼20% of the eggs developed through stage 6, eggs treated with SmInAct dsRNA aborted development at stage 2 (Figure 3C and 3D). An absence of SmInAct transcripts (Figure S1), and a nearly 10-fold decrease in SmInAct protein (Figure 3E), were associated with the failure of SmInAct dsRNA–treated eggs to develop. This phenotype was not observed when eggs were treated with dsRNA corresponding to luciferase, a sequence not encoded in the schistosome genome (Figure 3C and 3D), or to S. mansoni cathepsin B1 (SmCB1), a cathepsin B detectable in eggs (Table 1).
Multiple components of a TGF-β signaling pathway have been characterized in S. mansoni, but a ligand of parasite origin for the pathway has remained elusive. Additionally, while functions in host–parasite interactions have been proposed based on the expression of receptors on the parasite surface, and on the responsiveness of the parasite receptors to host TGF-β [5–7,14], the function that TGF-β signaling plays in S. mansoni has remained unclear. In this study, we report the expression of SmInAct, a TGF-β–like ligand in the parasitic flatworm S. mansoni, the production of which is coupled to the reproductive potential of the worms. We provide evidence that SmInAct plays a crucial role in embryogenesis.
Understanding of the developmental processes regulated by TGF-β in invertebrates is based largely on data from the model organisms D. melanogaster and C. elegans. Decapentaplegic, a bone morphogenetic protein (BMP)–like homolog in D. melanogaster, acts as a morphogen by determining cell fate along the dorsal–ventral axis in a gradient-dependent manner [29]. Also in D. melanogaster, a type I receptor, baboon, stimulates cellular proliferation and is essential for normal embryonic development [30]. Presumably, SmInAct could be fulfilling functions in the schistosome egg analogous to these known roles for decapentaplegic and/or baboon. None of the three characterized TGF-β homologs in C. elegans are important for patterning or growth of the embryo [31–33]; however, two TGF-β homologs have yet to be examined (tig-2 and Y46E12BL.1), and, intriguingly, serial analysis of gene expression (SAGE) tags for both homologs have been found in the C. elegans embryo [34]. Like the other C. elegans TGF-β homologs that are resistant to RNAi affects, tig-2 and Y46E12BL.1 have no phenotype in genome-wide RNAi screens [35,36]; therefore, direct mutagenesis will likely be required to determine the function of these genes.
The identification of SmInAct, a TGF-β superfamily member, as a key component of egg development in S. mansoni, a member of the Platyhelminthes, the earliest branch of the Bilateria [37], underscores the central role played by this pathway in embryogenesis. While one type I and one type II TGF-β receptor have been characterized for S. mansoni, there appears to be at least three type I receptors and two type II receptors present in the genome based on a preliminary blast search for homologs. It will be important to delineate which of the S. mansoni type I and type II TGF-β receptors are involved in SmInAct signaling and to identify the Smads important for transmitting the signal induced by this growth factor. Furthermore, identifying the genes regulated by SmInAct signaling will provide information regarding the precise function that this growth factor serves in egg maturation, as well as the functions the pathway may serve in other life stages of the parasite, including the adult male.
SmInAct protein was not detectable in infertile females recovered from single-sex infections or from IL-7R−/− mice, despite the fact that these parasites contained SmInAct transcripts (although at lower levels than in fecund parasites). This strongly indicates that SmInAct is both transcriptionally and post-transcriptionally regulated by worms of the opposite sex as well as by signals from the host. It is well established that parasites recovered from hosts lacking CD4+ T cells are developmentally stunted and produce significantly fewer fertile eggs than those recovered from mixed-sex infections of immunocompetent hosts. Translation of SmInAct mRNA is the first identified molecular process downstream of the effect of the host immune system on schistosome development [22–24], and as such, could open the way towards an increased understanding of this unusual feature of schistosome biology. The finding that the production of SmInAct in males is under the same constraints as in females is curious and perhaps indicates an additional function(s) for SmInAct in S. mansoni. We are unaware of a link between the site of expression of SmInAct in the male schistosome and reproductive events, and further work is required to elucidate the function of SmInAct in male worms.
In other settings, the uncoupling of transcription and translation is linked to the activation of the integrated stress response [38–41]. This mechanism, conserved in eukaryotes, re-programs cells to conserve energy in response to stress signals such as amino acid deficiency and oxidative stress by restricting the translation of transcripts requiring an active translation initiation complex [38–41]. Limited cellular energy is then used for the expression of genes necessary to maintain cell viability [42]. In this context, parasites in single-sex infections and in mice lacking CD4+ T cells may be considered stressed due to the lack of signals received from the opposite sex and immunocompetent host, thereby restricting the translation of non-essential transcripts. SmInAct protein expression may be considered expendable considering the role it plays in embryogenesis rather than in crucial cellular functions linked to the survival of the adult worm. A more thorough investigation of the S. mansoni homologs of translation factors involved in the stress response and of the regulation of other transcripts and protein expression will be required to evaluate this possibility.
Post-transcription regulation of eukaryotic transcripts is controlled in part by the 3′UTR [43]. This region can bind elements (including microRNAs and proteins) that inhibit the translation and/or decrease mRNA stability. For example, 3′UTRs of several mammalian cytokines contain adenosine- and uridine-rich elements (AREs) that bind ARE-binding proteins (ARE-BPs) (reviewed in [44]). The binding of ARE-BPs to these transcripts causes either rapid decay or inhibits their translation. While AREs are somewhat divergent in sequence, they often contain the consensus “AUUUA” and are found in a uridine-rich environment. Interestingly, the long 3′UTR of SmInAct has two exact repeats of “UUUCTAUUUA” that contain the consensus “AUUUA” ARE (underlined). Furthermore, the 3′UTR of SmInAct is U-rich (43% uridines). It will be interesting to determine whether these repeats, or other regions of the long 3′UTR, play a role in the post-transcriptional regulation of SmInAct expression.
It is of interest when considering the relationship of schistosomes with their mammalian hosts to note that in other systems, TGF-β superfamily members have been shown to function across phylum boundaries [45,46]. For example, the Drosophila BMP homologs DPP and 60A are able to induce bone development when injected into the skin of rats [45], and mammalian BMP-4 can rescue Drosophila DPP mutants [46]. Consequently, we believe that it is feasible that SmInAct could act as a ligand to initiate signaling in host cells. It is clear that proteins produced by eggs have distinct immunomodulatory functions [47], and SmInAct could conceivably participate in these effects if secreted/excreted from the schistosome egg. Our identification of SmInAct as a cytokine that is molecularly conserved between host and parasite, coupled with the description of an effective method for altering gene expression in the schistosome egg, allows these and other issues to now be addressed. Despite recent advances in vaccine design [48], a solution for schistosomiasis remains an elusive goal. Current attempts to control schistosomiasis depend on repeated administration of one drug, praziquantel, with no replacements waiting in the wings should resistance develop. Understanding how schistosome eggs develop could provide targets for intervention in the schistosome life cycle and for blocking disease progression.
The Puerto Rican/NMRI strain of S. mansoni was used in all experiments. Adult schistosomes were recovered by hepatic-portal perfusion from C57BL/6 female mice or B6 IL-7R−/− (The Jackson Laboratory, http://www.jax.org) that had each been percutaneously exposed to ∼60 cercariae 8 wk earlier. Adult parasites and eggs laid were maintained in vitro in M199 (Gibco, http://www.invitrogen.com), 10% fetal calf serum, 1% Antibiotic/Antimycotic (Gibco), and 1% HEPES in a 37 °C/5% CO2 atmosphere as previously described [11,28].
The C-terminal, translated region of the Drosophila activin homolog (dActivin) (amino acids 565–669) was used to search the Wellcome Trust's Sanger Institute's S. mansoni genome assembly using the tblastn algorithm. A contig (0020320) with significant similarity to dActivin was identified. Full-length cDNA corresponding to SmInAct was isolated using total RNA (1 μg) from adult parasites and the SuperScript III GeneRacer 5′ and 3′ RACE kit (Invitrogen, http://www.invitrogen.com) as per manufacturer's instructions. Gene-specific primers were designed for isolation of the 5′-end (5′-GGTTCAAAACTTTTCGGGTGTA-3′) and 3′-end (5′-AATCTTGTTGTCATCCAACTCAA-3′) of SmInAct and used in RT-PCR with GeneRacer 5′ and 3′ primers according to manufacturer's suggestions. Resulting amplicons were cloned into the TOPO cloning vector (Invitrogen) and sequenced. To verify the full-length sequence of SmInAct, primers designed from the 5′ and 3′ ends of the transcript were used in RT-PCR, and the resulting fragment was cloned and sequenced.
Sequence similarities between the deduced amino acid sequence of SmInAct and other members of the TGF-β superfamily were determined through multiple sequence alignments using the ClustalW algorithm, as well as the Align 2 sequences (bl2seq) program at the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov). An unrooted phylogram was drawn using amino acids within the conserved C-terminal domain of SmInAct, and known TGF-β superfamily members and distances were drawn using the Dayhoff Pam matrix and neighbor-joining algorithm in the PHYLIP software package developed by J. Felsenstein, University of Washington, Seattle, Washington, United States (http://evolution.genetics.washington.edu/phylip.html). Percentages at branch points are based on 1,000 bootstrap runs.
Total RNA was extracted from parasites using Qiagen's RNeasy Mini kit (http://www.qiagen.com), and contaminating genomic DNA was removed by DNase treatment using the Turbo DNA-free endonuclease (Ambion, http://www.ambion.com). First-strand cDNA was synthesized using 500 ng of RNA, SuperScript II reverse transcriptase (Invitrogen), and oligo dT as a primer. RT-minus controls were performed to confirm the absence of genomic DNA (unpublished data).
SmInAct transcript levels in egg and adult stages were quantified relative to α-tubulin using Applied Biosystems' 7500 real-time PCR system and SYBR green PCR Master Mix (Applied Biosystems, http://www.appliedbiosystems.com). Total reaction volume was 10 μl with 300 nM of each primer, 5 μl of SYBR green PCR Master Mix, and 0.5 μl of cDNA as template (or water as a negative control). SmInAct primers were: forward 5′-AATCTTGTTGTCATCCAACTCAA-3′ and reverse 5′-AACTACAAGCACATCCTAAAACAA-3′. α-Tubulin primers were: forward 5′-CCAGCAAATCAGATGGTGAA-3′ and reverse 5′-TTGACATCCTTGGGGACAAC-3′. PCR efficiency (E) was determined for both primer sets by plotting cycle thresholds from a 10-fold serial dilution of cDNA and inputting the slope in the equation E = 10(−1/slope). For expression analyses, quantification of SmInAct transcript relative to α-tubulin was calculated using the equation: ratio = (ESmα-tubulin)CT/(ESmInAct)CT where ESmα-tubulin is the PCR efficiency of the reference gene, ESmInAct is the PCR efficiency of target gene, and CT is the cycle threshold. For analysis of RNAi-induced knockdown, quantification of SmInAct transcript relative to paramyosin (paramyosin primers were: forward 5′-CGTGAAGGTCGTCGTATGGT-3′ and reverse 5′-GACGTTCAAATTTACGTGCTTG-3′) was calculated using the 2−ΔΔCt method. Dissociation curves were generated for each real-time RT-PCR to verify the amplification of only one product.
Eco RI (forward) and Xho I (reverse) tagged primers were designed to amplify the C-terminal bioactive region of SmInAct (forward 5′-GGAATTCTCATTAACTAAAGGAGATGA-3 and reverse 5′-CCGCTCGAGTTAACTACAAGCACATCCTA-3′). The amplified product was cloned into the expression vector pET28a+ (Novagen, http://www.emdbiosciences.com) and sequenced to verify the absence of any mutations. Expression of recombinant SmInAct was induced in Escherichia coli BL21(DE3) by addition of 1 mM IPTG when cultures reached an OD600 of 0.5 at 37 °C, followed by 3 hours of shaking at room temperature. Recombinant SmInAct was expressed in bacteria as insoluble inclusion bodies. Exhaustive attempts to refold the protein using gluathione and reduced glutathione proved unsuccessful. We therefore purified the protein via nickel column chromatography under denaturing conditions (6 M urea) as per the manufacturer's protocol (Novagen). Antiserum was generated by Cocalico Biologicals (http://www.cocalicobiologicals.com) through subcutaneous inoculation of a rabbit with 100 μg of purified protein in complete Freund's adjuvant, followed by three boosts of 50 μg in incomplete Freund's adjuvant on days 14, 21, and 49, followed by exsanguinations on day 64.
For detection of SmInAct protein, 10 μg of protein extracted from eggs, adult males, and adult females via Dounce homogenizing in lysis buffer (1% Triton-X 100, 20 mM HEPES, 10% glycerol, 150 mM NaCl) supplemented with a protease inhibitor cocktail (Sigma, http://www.sigmaaldrich.com) were separated by SDS-PAGE, electroblotted, and probed with anti-SmInAct antiserum (1:10,000), pre-immune serum (1:10,000), or a monoclonal antibody (4B1) against paramyosin. Affinity purified HRP-conjugated goat anti-rabbit IgG (Cell Signaling Technology, http://www.cellsignal.com) was used to detect bound rabbit antibodies, while an affinity purified HRP-conjugated horse anti-mouse IgG (Cell Signaling Technology) was used to detect the anti-paramyosin monoclonal antibody. The secondary antibodies were detected using ECL reagents as per manufacturer's instructions (GE Healthcare, http://www.gehealthcare.com).
Localization of SmInAct in 5-μm sections of adult S. mansoni was performed as previously described [49]. DIG-labeled sense and anti-sense transcripts were generated using Roche's DIG RNA labeling mix (http://www.roche.com) as per manufacturer's instructions with T7-tagged amplicons as template (sense: forward 5′-TAATACGACTCACTATAGGGTTGATCCAAAAAAGGTTGTTATGG-3′, reverse 5′-TTAACTACAAGCAGCTCCTA −3′; anti-sense: forward 5′-ATAATATGTAATAATTGTGA −3′ reverse 5′- TAATACGACTCACTATAGGGAACTACAAGCACATCCTAAAACAA-3′). The hybridized DIG-probes were detected using an alkaline-phosphatase conjugated anti-DIG antibody (Roche), and visualized using NBT (337.5 μg/ml) and BCIP (175 μg/ml) in 0.1M Tris-HCl, 0.1M NaCl, 0.05 MgCl2. Worm sections were photographed using a Leica DMIRB microscope and DC500 camera (Leica, http://www.leica.com).
dsRNA was synthesized using the T7 Megascript kit (Ambion) as per manufacturer's instructions. T7-tagged primers were used to generate a 381-bp SmInAct-dsRNA template encompassing the active ligand domain (forward 5′-TAATACGACTCACTATAGGGCGATCATTAACTAAAGGAGATGAG-3′, reverse 5′-TAATACGACTCACTATAGGGAACTACAAGCACATCCTAAAACAA-3′). Luciferase and SmCB1 dsRNAs (negative controls) were generated as described [25]. For dsRNA treatment of worms, five adult pairs were cultured in the presence of 1 μg/ml dsRNA for 7 d with medium and dsRNA changes occurring every other day. For dsRNA treatment of eggs, five adult pairs were cultured as above (in the absence of dsRNA) for 2 d, worms were removed, and dsRNA was added at 1 μg/ml. Eggs were photographed using a Leica DMIRB microscope and DC500 camera.
Student t-test was used for statistical analyses of dsRNA-induced knockdown of SmInAct expression, change in expression of SmInAct in single-sex and IL-7R−/− mice, and egg developmental phenotypes (control versus SmInAct dsRNA). Chi-square analyses were used to test the statistical significance of the egg developmental phenotype. The Yates correction was applied because we specified only two categories: undeveloped and developed (Table 1).
Sequence data reported in this manuscript are available from GenBank (http://www.ncbi.nlm.nih.gov/Genbank) under accession number DQ863513. Other GenBank accession numbers of genes and sequences used in this study include: B. malayi TGH-1 (AAB71839); B. malayi TGH-2 (AAD19903); C. elegans DAF-7 (NP_497265); C. elegans DBL-1 (NP_504709); Danio rerio Activinβ A isoform 2 (AAX68505); D. melanogaster Activin (NP_651942); D. melanogaster dActivin (AF454392); D. melanogaster decapentaplegic (NP_477311); Homo sapiens Activinβ E (NP_113667); H. sapiens BMP-2 (NP_001191); H. sapiens BMP-3 (NP_001192); H. sapiens BMP-4 (NP_031580); H. sapiens BMP-5 (NP_066551); H. sapiens BMP-6 (NP_001709); H. sapiens BMP-7 (NP_001710); H. sapiens BMP-8 (NP_861525); H. sapiens GDF-5 (NP_000548); H. sapiens GDF-6 (NP_001001557); H. sapiens GDF-7 (NP_878248); H. sapiens GDF-10 (NP_004953); H. sapiens Inhibinβ A precursor (NP_002183); H. sapiens Inhibinβ B (NP_002184); H. sapiens Inhibinβ C (NP_005529); H. sapiens TGF-β 1 (NP_000651); H. sapiens TGF-β 2 (NP_003229); H. sapiens TGF-β 3 (NP_003230); Mus musculus BMP-2 (NP_031579); M. musculus BMP-3 (NP_775580); M. musculus BMP-4 (NP_031580); M. musculus GDF-10 (NP_665684); M. musculus Inhibinβ A (NP_032406); M. musculus Inhibinβ B (NP_032407); M. musculus TGF-β 1 (NP_035707); M. musculus TGF-β 2 (NP_033393); M. musculus TGF-β 3 (NP_033394); S. mansoni α-tubulin (M80214); S. mansoni paramyosin (M35499); and Strongyloides stercoralis TGH-1 (AAV84743). |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.